From 70704b4ec30b69952c70372e0549ff9f5541eddc Mon Sep 17 00:00:00 2001
From: uiop07558 <uiop07558@gmail.com>
Date: Tue, 6 Feb 2024 17:03:11 +0300
Subject: [PATCH 01/15] add reference implementation for IDW

---
 .vscode/settings.json |  3 +++
 CMakeLists.txt        |  1 +
 include/algos.hpp     |  3 +++
 src/bgr2gray.cpp      | 47 +++++++++++++++++++++---------------------
 src/idw.cpp           | 48 +++++++++++++++++++++++++++++++++++++++++++
 test/test_main.cpp    | 19 +++++++++++++++++
 6 files changed, 98 insertions(+), 23 deletions(-)
 create mode 100644 .vscode/settings.json
 create mode 100644 src/idw.cpp

diff --git a/.vscode/settings.json b/.vscode/settings.json
new file mode 100644
index 0000000..70e34ec
--- /dev/null
+++ b/.vscode/settings.json
@@ -0,0 +1,3 @@
+{
+    "C_Cpp.errorSquiggles": "disabled"
+}
\ No newline at end of file
diff --git a/CMakeLists.txt b/CMakeLists.txt
index af70cbb..85f2881 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -114,6 +114,7 @@ add_executable(test_algo ${PERF_SOURCES})
 target_link_libraries(test_algo
   algos
   "${binary_dir}/lib/libopencv_ts.a"
+  "${binary_dir}/lib/libopencv_imgcodecs.so"
   pthread
 )
 target_include_directories(test_algo PUBLIC
diff --git a/include/algos.hpp b/include/algos.hpp
index c8517bc..a95cf5d 100644
--- a/include/algos.hpp
+++ b/include/algos.hpp
@@ -32,3 +32,6 @@ void convolution_nhwc_halide(float* src, float* kernel, float* dst,
 void convolution_opencv(const cv::Mat& src, const cv::Mat& weights, cv::Mat& dst,
                         int inpChannels, int outChannels);
 #endif  // HAVE_OPENCV_DNN
+
+
+void idw_ref(const uint8_t* src, uint8_t* dst, int height, int width, int* points, float* weights);
diff --git a/src/bgr2gray.cpp b/src/bgr2gray.cpp
index 67049db..ad7ef5d 100644
--- a/src/bgr2gray.cpp
+++ b/src/bgr2gray.cpp
@@ -60,29 +60,30 @@ void bgr2gray_interleaved_halide(uint8_t* src, uint8_t* dst, int height, int wid
 
         int factor = 16;
         f.vectorize(x, factor);
-
-        // Compile
-        Target target;
-        target.os = Target::OS::Linux;
-        target.arch = Target::Arch::RISCV;
-        target.bits = 64;
-        target.vector_bits = factor * sizeof(uint8_t) * 8;
-
-        // Tested XuanTie C906 has 128-bit vector unit
-        CV_Assert(target.vector_bits <= 128);
-
-        std::vector<Target::Feature> features;
-        features.push_back(Target::RVV);
-        features.push_back(Target::NoAsserts);
-        features.push_back(Target::NoRuntime);
-        target.set_features(features);
-
-        std::cout << target << std::endl;
-        f.print_loop_nest();
-
-        // Dump AOT code
-        f.compile_to_header("bgr2gray_interleaved.h", {input}, "bgr2gray_interleaved", target);
-        f.compile_to_assembly("bgr2gray_interleaved.s", {input}, "bgr2gray_interleaved", target);
+        f.realize(output);
+
+    //     // Compile
+    //     Target target;
+    //     target.os = Target::OS::Linux;
+    //     target.arch = Target::Arch::RISCV;
+    //     target.bits = 64;
+    //     target.vector_bits = factor * sizeof(uint8_t) * 8;
+
+    //     // Tested XuanTie C906 has 128-bit vector unit
+    //     CV_Assert(target.vector_bits <= 128);
+
+    //     std::vector<Target::Feature> features;
+    //     features.push_back(Target::RVV);
+    //     features.push_back(Target::NoAsserts);
+    //     features.push_back(Target::NoRuntime);
+    //     target.set_features(features);
+
+    //     std::cout << target << std::endl;
+    //     f.print_loop_nest();
+
+    //     // Dump AOT code
+    //     f.compile_to_header("bgr2gray_interleaved.h", {input}, "bgr2gray_interleaved", target);
+    //     f.compile_to_assembly("bgr2gray_interleaved.s", {input}, "bgr2gray_interleaved", target);
     }
 #endif
 }
diff --git a/src/idw.cpp b/src/idw.cpp
new file mode 100644
index 0000000..28be2cd
--- /dev/null
+++ b/src/idw.cpp
@@ -0,0 +1,48 @@
+#include <opencv2/opencv.hpp>
+
+#include "algos.hpp"
+
+#include <limits>
+
+using namespace std;
+
+static const int pointCount = 100; 
+
+void idw_ref(const uint8_t* src, uint8_t* dst, int height, int width, int* points, float* weights) {
+    float* mask = new float[height * width]();
+
+    float maxVal = -numeric_limits<float>::infinity();
+    float minVal = numeric_limits<float>::infinity();
+
+    for (int y = 0; y < height; y++) {
+        for (int x = 0; x < width; x++) {
+            float dot = 0;
+            for (int i = 0; i < pointCount; i++) {
+                int x0 = points[i];
+                int y0 = points[i + 1];
+                int dx = x - x0;
+                int dy = y - y0;
+
+                dot += sqrt(dx*dx + dy*dy) * weights[i];
+            }
+
+            mask[y*width + x] = dot;
+            if (dot < minVal) {
+                minVal = dot;
+            }
+            if (dot > maxVal) {
+                maxVal = dot;
+            }
+        }
+    }
+
+    float diff = maxVal - minVal;
+
+    for (int y = 0; y < height; y++) {
+        for (int x = 0; x < width; x++) {
+            dst[y*width + x] = (uint8_t) (255 * (mask[y*width + x] - minVal) / diff);
+        }
+    }
+
+    delete[] mask;
+}
\ No newline at end of file
diff --git a/test/test_main.cpp b/test/test_main.cpp
index 87a4fa6..49e4035 100644
--- a/test/test_main.cpp
+++ b/test/test_main.cpp
@@ -143,3 +143,22 @@ TEST(convolution_nhwc, halide) {
 }
 
 #endif  // HAVE_OPENCV_DNN
+
+
+TEST(idw_ref, halide) {
+    Mat src(height, width, CV_8U), dst(height, width, CV_8U), cl_dst(height, width, CV_8UC3);
+    // randu(src, 0, 256);
+
+    // 300 elems (100 points)
+    int points[] = {0, 0, 72, 0, 213, 79, 0, 426, 60, 0, 640, 76, 0, 853, 128, 0, 1066, 67, 0, 1280, 65, 0, 1493, 64, 0, 1706, 60, 0, 1920, 61, 120, 0, 81, 120, 213, 79, 120, 426, 58, 120, 640, 132, 120, 853, 149, 120, 1066, 142, 120, 1280, 64, 120, 1493, 69, 120, 1706, 65, 120, 1920, 64, 240, 0, 75, 240, 213, 68, 240, 426, 140, 240, 640, 153, 240, 853, 145, 240, 1066, 132, 240, 1280, 152, 240, 1493, 125, 240, 1706, 66, 240, 1920, 58, 360, 0, 78, 360, 213, 60, 360, 426, 139, 360, 640, 168, 360, 853, 154, 360, 1066, 138, 360, 1280, 145, 360, 1493, 160, 360, 1706, 68, 360, 1920, 60, 480, 0, 77, 480, 213, 59, 480, 426, 165, 480, 640, 183, 480, 853, 166, 480, 1066, 142, 480, 1280, 123, 480, 1493, 155, 480, 1706, 144, 480, 1920, 60, 600, 0, 83, 600, 213, 65, 600, 426, 178, 600, 640, 184, 600, 853, 138, 600, 1066, 124, 600, 1280, 132, 600, 1493, 175, 600, 1706, 175, 600, 1920, 60, 720, 0, 86, 720, 213, 60, 720, 426, 182, 720, 640, 179, 720, 853, 152, 720, 1066, 140, 720, 1280, 115, 720, 1493, 156, 720, 1706, 172, 720, 1920, 65, 840, 0, 90, 840, 213, 69, 840, 426, 164, 840, 640, 186, 840, 853, 146, 840, 1066, 147, 840, 1280, 122, 840, 1493, 152, 840, 1706, 166, 840, 1920, 69, 960, 0, 85, 960, 213, 67, 960, 426, 186, 960, 640, 169, 960, 853, 162, 960, 1066, 147, 960, 1280, 124, 960, 1493, 149, 960, 1706, 157, 960, 1920, 69, 1080, 0, 85, 1080, 213, 70, 1080, 426, 193, 1080, 640, 192, 1080, 853, 155, 1080, 1066, 135, 1080, 1280, 120, 1080, 1493, 138, 1080, 1706, 128, 1080, 1920, 70};
+    // 100 elems
+    float weights[] = {0.04053167, -0.07571915, 0.0025228, 0.1974248, -0.14461569, 0.30541419, -0.06167972, -0.04958807, 0.00050641, -0.02023852, -0.05910183, -0.05512609, 0.37617088, -0.15904428, -0.10229038, -0.34355378, 0.39128499, 0.16502984, -0.00916588, -0.05264022, 0.01936275, 0.07156495, -0.35144818, -0.01955447, 0.07818357, 0.10850975, -0.34972095, -0.12186563, 0.04116577, 0.00378971, -0.05382689, 0.12480152, 0.0629668, 0.01070963, 0.03499341, 0.03218783, -0.00559502, -0.21214646, 0.34402987, -0.0246006, -0.01532894, 0.14944613, -0.12848858, -0.04539923, -0.1401344, -0.08143395, 0.18241941, 0.08353522, -0.20102434, 0.05168503, -0.05897654, 0.09669851, -0.13024191, -0.02125988, 0.19499928, 0.12311092, -0.05857943, -0.14954968, -0.19763496, 0.1172914, -0.04584096, 0.20130274, -0.198312, 0.04763513, -0.05524272, -0.07034198, 0.15155036, 0.0718368, -0.09612551, 0.09076872, -0.0863682, 0.08457486, 0.05520895, -0.10551672, 0.12577908, -0.05857014, 0.03529613, 0.01300365, -0.09736055, 0.06580921, -0.03074898, 0.1528181, -0.18295799, 0.18713497, -0.07368328, -0.06431868, 0.02485816, -0.02782657, -0.15182979, 0.06845911, -0.01974711, 0.23508139, -0.23124445, -0.1202987, 0.05380124, 0.04331648, 0.06285662, -0.01171756, 0.01128749, 0.06012665};
+
+    idw_ref(NULL, dst.ptr<uint8_t>(), height, width, points, weights);
+
+    applyColorMap(dst, cl_dst, COLORMAP_JET);
+
+    // imwrite("src.png", src);
+    imwrite("res.png", dst);
+    imwrite("cres.png", cl_dst);
+}
\ No newline at end of file

From c143a56a27473659b9c6bbf1af12e89ff44e4870 Mon Sep 17 00:00:00 2001
From: Dmitry Kurtaev <dmitry.kurtaev@gmail.com>
Date: Tue, 6 Feb 2024 22:20:13 +0300
Subject: [PATCH 02/15] Minor fixes for reference IDW

---
 .vscode/settings.json |  3 ---
 CMakeLists.txt        |  2 +-
 src/bgr2gray.cpp      | 47 +++++++++++++++++++++----------------------
 src/idw.cpp           |  8 ++++----
 4 files changed, 28 insertions(+), 32 deletions(-)
 delete mode 100644 .vscode/settings.json

diff --git a/.vscode/settings.json b/.vscode/settings.json
deleted file mode 100644
index 70e34ec..0000000
--- a/.vscode/settings.json
+++ /dev/null
@@ -1,3 +0,0 @@
-{
-    "C_Cpp.errorSquiggles": "disabled"
-}
\ No newline at end of file
diff --git a/CMakeLists.txt b/CMakeLists.txt
index 85f2881..27127bd 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -92,6 +92,7 @@ target_include_directories(algos PRIVATE
 target_link_libraries(algos
   "${binary_dir}/install/lib/libopencv_core.so"
   "${binary_dir}/install/lib/libopencv_imgproc.so"
+  "${binary_dir}/install/lib/libopencv_imgcodecs.so"
 )
 
 
@@ -114,7 +115,6 @@ add_executable(test_algo ${PERF_SOURCES})
 target_link_libraries(test_algo
   algos
   "${binary_dir}/lib/libopencv_ts.a"
-  "${binary_dir}/lib/libopencv_imgcodecs.so"
   pthread
 )
 target_include_directories(test_algo PUBLIC
diff --git a/src/bgr2gray.cpp b/src/bgr2gray.cpp
index ad7ef5d..67049db 100644
--- a/src/bgr2gray.cpp
+++ b/src/bgr2gray.cpp
@@ -60,30 +60,29 @@ void bgr2gray_interleaved_halide(uint8_t* src, uint8_t* dst, int height, int wid
 
         int factor = 16;
         f.vectorize(x, factor);
-        f.realize(output);
-
-    //     // Compile
-    //     Target target;
-    //     target.os = Target::OS::Linux;
-    //     target.arch = Target::Arch::RISCV;
-    //     target.bits = 64;
-    //     target.vector_bits = factor * sizeof(uint8_t) * 8;
-
-    //     // Tested XuanTie C906 has 128-bit vector unit
-    //     CV_Assert(target.vector_bits <= 128);
-
-    //     std::vector<Target::Feature> features;
-    //     features.push_back(Target::RVV);
-    //     features.push_back(Target::NoAsserts);
-    //     features.push_back(Target::NoRuntime);
-    //     target.set_features(features);
-
-    //     std::cout << target << std::endl;
-    //     f.print_loop_nest();
-
-    //     // Dump AOT code
-    //     f.compile_to_header("bgr2gray_interleaved.h", {input}, "bgr2gray_interleaved", target);
-    //     f.compile_to_assembly("bgr2gray_interleaved.s", {input}, "bgr2gray_interleaved", target);
+
+        // Compile
+        Target target;
+        target.os = Target::OS::Linux;
+        target.arch = Target::Arch::RISCV;
+        target.bits = 64;
+        target.vector_bits = factor * sizeof(uint8_t) * 8;
+
+        // Tested XuanTie C906 has 128-bit vector unit
+        CV_Assert(target.vector_bits <= 128);
+
+        std::vector<Target::Feature> features;
+        features.push_back(Target::RVV);
+        features.push_back(Target::NoAsserts);
+        features.push_back(Target::NoRuntime);
+        target.set_features(features);
+
+        std::cout << target << std::endl;
+        f.print_loop_nest();
+
+        // Dump AOT code
+        f.compile_to_header("bgr2gray_interleaved.h", {input}, "bgr2gray_interleaved", target);
+        f.compile_to_assembly("bgr2gray_interleaved.s", {input}, "bgr2gray_interleaved", target);
     }
 #endif
 }
diff --git a/src/idw.cpp b/src/idw.cpp
index 28be2cd..8c738f1 100644
--- a/src/idw.cpp
+++ b/src/idw.cpp
@@ -6,7 +6,7 @@
 
 using namespace std;
 
-static const int pointCount = 100; 
+static const int pointCount = 100;
 
 void idw_ref(const uint8_t* src, uint8_t* dst, int height, int width, int* points, float* weights) {
     float* mask = new float[height * width]();
@@ -18,8 +18,8 @@ void idw_ref(const uint8_t* src, uint8_t* dst, int height, int width, int* point
         for (int x = 0; x < width; x++) {
             float dot = 0;
             for (int i = 0; i < pointCount; i++) {
-                int x0 = points[i];
-                int y0 = points[i + 1];
+                int x0 = points[3 * i + 1];
+                int y0 = points[3 * i];
                 int dx = x - x0;
                 int dy = y - y0;
 
@@ -45,4 +45,4 @@ void idw_ref(const uint8_t* src, uint8_t* dst, int height, int width, int* point
     }
 
     delete[] mask;
-}
\ No newline at end of file
+}

From f11f48820f49b277f413b22538c63c8236122c03 Mon Sep 17 00:00:00 2001
From: uiop07558 <uiop07558@gmail.com>
Date: Wed, 7 Feb 2024 17:27:08 +0300
Subject: [PATCH 03/15] add idw_halide

---
 .vscode/c_cpp_properties.json |  16 +++++
 .vscode/settings.json         |   9 +++
 include/algos.hpp             |   3 +-
 perf/perf_main.cpp            |  43 +++++++++++++
 src/idw.cpp                   | 117 ++++++++++++++++++++++++++++++++++
 test/test_main.cpp            |   9 ++-
 6 files changed, 194 insertions(+), 3 deletions(-)
 create mode 100644 .vscode/c_cpp_properties.json
 create mode 100644 .vscode/settings.json

diff --git a/.vscode/c_cpp_properties.json b/.vscode/c_cpp_properties.json
new file mode 100644
index 0000000..4039bef
--- /dev/null
+++ b/.vscode/c_cpp_properties.json
@@ -0,0 +1,16 @@
+{
+    "configurations": [
+        {
+            "name": "Linux",
+            "includePath": [
+                "${workspaceFolder}/**"
+            ],
+            "defines": [],
+            "compilerPath": "/usr/bin/gcc",
+            "cStandard": "c17",
+            "cppStandard": "gnu++17",
+            "intelliSenseMode": "linux-gcc-x64"
+        }
+    ],
+    "version": 4
+}
\ No newline at end of file
diff --git a/.vscode/settings.json b/.vscode/settings.json
new file mode 100644
index 0000000..8104e93
--- /dev/null
+++ b/.vscode/settings.json
@@ -0,0 +1,9 @@
+{
+    "C_Cpp.errorSquiggles": "disabled",
+    "files.associations": {
+        "limits": "cpp",
+        "iostream": "cpp",
+        "cmath": "cpp",
+        "optional": "cpp"
+    }
+}
\ No newline at end of file
diff --git a/include/algos.hpp b/include/algos.hpp
index a95cf5d..7ae4307 100644
--- a/include/algos.hpp
+++ b/include/algos.hpp
@@ -33,5 +33,6 @@ void convolution_opencv(const cv::Mat& src, const cv::Mat& weights, cv::Mat& dst
                         int inpChannels, int outChannels);
 #endif  // HAVE_OPENCV_DNN
 
-
 void idw_ref(const uint8_t* src, uint8_t* dst, int height, int width, int* points, float* weights);
+
+void idw_halide(const uint8_t* src, uint8_t* dst, int height, int width, int* pointsBuf, float* weightsBuf);
\ No newline at end of file
diff --git a/perf/perf_main.cpp b/perf/perf_main.cpp
index f99b78b..dc7296a 100644
--- a/perf/perf_main.cpp
+++ b/perf/perf_main.cpp
@@ -14,8 +14,14 @@
 
 #include "algos.hpp"
 
+#include <Halide.h>
+
+// #include <iostream>
+
 using namespace cv;
 
+// using namespace std;
+
 CV_PERF_TEST_MAIN("")
 
 Mat src(1080, 1920, CV_8UC3);
@@ -181,3 +187,40 @@ PERF_TEST(convolution_nhwc, halide) {
 }
 
 #endif  // HAVE_OPENCV_DNN
+
+
+PERF_TEST(idw, reference) {
+    const int width = 1920;
+    const int height = 1080;
+
+    // 300 elems (100 points)
+    int points[] = {0, 0, 72, 0, 213, 79, 0, 426, 60, 0, 640, 76, 0, 853, 128, 0, 1066, 67, 0, 1280, 65, 0, 1493, 64, 0, 1706, 60, 0, 1920, 61, 120, 0, 81, 120, 213, 79, 120, 426, 58, 120, 640, 132, 120, 853, 149, 120, 1066, 142, 120, 1280, 64, 120, 1493, 69, 120, 1706, 65, 120, 1920, 64, 240, 0, 75, 240, 213, 68, 240, 426, 140, 240, 640, 153, 240, 853, 145, 240, 1066, 132, 240, 1280, 152, 240, 1493, 125, 240, 1706, 66, 240, 1920, 58, 360, 0, 78, 360, 213, 60, 360, 426, 139, 360, 640, 168, 360, 853, 154, 360, 1066, 138, 360, 1280, 145, 360, 1493, 160, 360, 1706, 68, 360, 1920, 60, 480, 0, 77, 480, 213, 59, 480, 426, 165, 480, 640, 183, 480, 853, 166, 480, 1066, 142, 480, 1280, 123, 480, 1493, 155, 480, 1706, 144, 480, 1920, 60, 600, 0, 83, 600, 213, 65, 600, 426, 178, 600, 640, 184, 600, 853, 138, 600, 1066, 124, 600, 1280, 132, 600, 1493, 175, 600, 1706, 175, 600, 1920, 60, 720, 0, 86, 720, 213, 60, 720, 426, 182, 720, 640, 179, 720, 853, 152, 720, 1066, 140, 720, 1280, 115, 720, 1493, 156, 720, 1706, 172, 720, 1920, 65, 840, 0, 90, 840, 213, 69, 840, 426, 164, 840, 640, 186, 840, 853, 146, 840, 1066, 147, 840, 1280, 122, 840, 1493, 152, 840, 1706, 166, 840, 1920, 69, 960, 0, 85, 960, 213, 67, 960, 426, 186, 960, 640, 169, 960, 853, 162, 960, 1066, 147, 960, 1280, 124, 960, 1493, 149, 960, 1706, 157, 960, 1920, 69, 1080, 0, 85, 1080, 213, 70, 1080, 426, 193, 1080, 640, 192, 1080, 853, 155, 1080, 1066, 135, 1080, 1280, 120, 1080, 1493, 138, 1080, 1706, 128, 1080, 1920, 70};
+    // 100 elems
+    float weights[] = {0.04053167, -0.07571915, 0.0025228, 0.1974248, -0.14461569, 0.30541419, -0.06167972, -0.04958807, 0.00050641, -0.02023852, -0.05910183, -0.05512609, 0.37617088, -0.15904428, -0.10229038, -0.34355378, 0.39128499, 0.16502984, -0.00916588, -0.05264022, 0.01936275, 0.07156495, -0.35144818, -0.01955447, 0.07818357, 0.10850975, -0.34972095, -0.12186563, 0.04116577, 0.00378971, -0.05382689, 0.12480152, 0.0629668, 0.01070963, 0.03499341, 0.03218783, -0.00559502, -0.21214646, 0.34402987, -0.0246006, -0.01532894, 0.14944613, -0.12848858, -0.04539923, -0.1401344, -0.08143395, 0.18241941, 0.08353522, -0.20102434, 0.05168503, -0.05897654, 0.09669851, -0.13024191, -0.02125988, 0.19499928, 0.12311092, -0.05857943, -0.14954968, -0.19763496, 0.1172914, -0.04584096, 0.20130274, -0.198312, 0.04763513, -0.05524272, -0.07034198, 0.15155036, 0.0718368, -0.09612551, 0.09076872, -0.0863682, 0.08457486, 0.05520895, -0.10551672, 0.12577908, -0.05857014, 0.03529613, 0.01300365, -0.09736055, 0.06580921, -0.03074898, 0.1528181, -0.18295799, 0.18713497, -0.07368328, -0.06431868, 0.02485816, -0.02782657, -0.15182979, 0.06845911, -0.01974711, 0.23508139, -0.23124445, -0.1202987, 0.05380124, 0.04331648, 0.06285662, -0.01171756, 0.01128749, 0.06012665};
+
+    Mat dst(height, width, CV_8U);
+
+    PERF_SAMPLE_BEGIN()
+        idw_ref(NULL, dst.ptr<uint8_t>(), height, width, points, weights);
+    PERF_SAMPLE_END()
+
+    SANITY_CHECK_NOTHING();
+}
+
+PERF_TEST(idw, halide) {
+    const int width = 1920;
+    const int height = 1080;
+
+    // 300 elems (100 points)
+    int points[] = {0, 0, 72, 0, 213, 79, 0, 426, 60, 0, 640, 76, 0, 853, 128, 0, 1066, 67, 0, 1280, 65, 0, 1493, 64, 0, 1706, 60, 0, 1920, 61, 120, 0, 81, 120, 213, 79, 120, 426, 58, 120, 640, 132, 120, 853, 149, 120, 1066, 142, 120, 1280, 64, 120, 1493, 69, 120, 1706, 65, 120, 1920, 64, 240, 0, 75, 240, 213, 68, 240, 426, 140, 240, 640, 153, 240, 853, 145, 240, 1066, 132, 240, 1280, 152, 240, 1493, 125, 240, 1706, 66, 240, 1920, 58, 360, 0, 78, 360, 213, 60, 360, 426, 139, 360, 640, 168, 360, 853, 154, 360, 1066, 138, 360, 1280, 145, 360, 1493, 160, 360, 1706, 68, 360, 1920, 60, 480, 0, 77, 480, 213, 59, 480, 426, 165, 480, 640, 183, 480, 853, 166, 480, 1066, 142, 480, 1280, 123, 480, 1493, 155, 480, 1706, 144, 480, 1920, 60, 600, 0, 83, 600, 213, 65, 600, 426, 178, 600, 640, 184, 600, 853, 138, 600, 1066, 124, 600, 1280, 132, 600, 1493, 175, 600, 1706, 175, 600, 1920, 60, 720, 0, 86, 720, 213, 60, 720, 426, 182, 720, 640, 179, 720, 853, 152, 720, 1066, 140, 720, 1280, 115, 720, 1493, 156, 720, 1706, 172, 720, 1920, 65, 840, 0, 90, 840, 213, 69, 840, 426, 164, 840, 640, 186, 840, 853, 146, 840, 1066, 147, 840, 1280, 122, 840, 1493, 152, 840, 1706, 166, 840, 1920, 69, 960, 0, 85, 960, 213, 67, 960, 426, 186, 960, 640, 169, 960, 853, 162, 960, 1066, 147, 960, 1280, 124, 960, 1493, 149, 960, 1706, 157, 960, 1920, 69, 1080, 0, 85, 1080, 213, 70, 1080, 426, 193, 1080, 640, 192, 1080, 853, 155, 1080, 1066, 135, 1080, 1280, 120, 1080, 1493, 138, 1080, 1706, 128, 1080, 1920, 70};
+    // 100 elems
+    float weights[] = {0.04053167, -0.07571915, 0.0025228, 0.1974248, -0.14461569, 0.30541419, -0.06167972, -0.04958807, 0.00050641, -0.02023852, -0.05910183, -0.05512609, 0.37617088, -0.15904428, -0.10229038, -0.34355378, 0.39128499, 0.16502984, -0.00916588, -0.05264022, 0.01936275, 0.07156495, -0.35144818, -0.01955447, 0.07818357, 0.10850975, -0.34972095, -0.12186563, 0.04116577, 0.00378971, -0.05382689, 0.12480152, 0.0629668, 0.01070963, 0.03499341, 0.03218783, -0.00559502, -0.21214646, 0.34402987, -0.0246006, -0.01532894, 0.14944613, -0.12848858, -0.04539923, -0.1401344, -0.08143395, 0.18241941, 0.08353522, -0.20102434, 0.05168503, -0.05897654, 0.09669851, -0.13024191, -0.02125988, 0.19499928, 0.12311092, -0.05857943, -0.14954968, -0.19763496, 0.1172914, -0.04584096, 0.20130274, -0.198312, 0.04763513, -0.05524272, -0.07034198, 0.15155036, 0.0718368, -0.09612551, 0.09076872, -0.0863682, 0.08457486, 0.05520895, -0.10551672, 0.12577908, -0.05857014, 0.03529613, 0.01300365, -0.09736055, 0.06580921, -0.03074898, 0.1528181, -0.18295799, 0.18713497, -0.07368328, -0.06431868, 0.02485816, -0.02782657, -0.15182979, 0.06845911, -0.01974711, 0.23508139, -0.23124445, -0.1202987, 0.05380124, 0.04331648, 0.06285662, -0.01171756, 0.01128749, 0.06012665};
+
+    Mat dst(height, width, CV_8U);
+
+    PERF_SAMPLE_BEGIN()
+        idw_halide(NULL, dst.ptr<uint8_t>(), height, width, points, weights);
+    PERF_SAMPLE_END()
+
+    SANITY_CHECK_NOTHING();
+}
\ No newline at end of file
diff --git a/src/idw.cpp b/src/idw.cpp
index 8c738f1..7ef9887 100644
--- a/src/idw.cpp
+++ b/src/idw.cpp
@@ -3,11 +3,128 @@
 #include "algos.hpp"
 
 #include <limits>
+#include <iostream>
 
 using namespace std;
 
+#ifdef __riscv
+  #include <HalideBuffer.h>
+  #include "histogram.h"
+  using namespace Halide::Runtime;
+#else
+  #include <Halide.h>
+  using namespace Halide;
+#endif
+
 static const int pointCount = 100;
 
+void idw_halide(const uint8_t* src, uint8_t* dst, int height, int width, int* pointsBuf, float* weightsBuf) {
+    float* maskBuf = new float[height * width]();
+
+    Buffer<float> mask(maskBuf, {width, height});
+    Buffer<uint8_t> output(dst, {width, height});
+
+    Buffer<int> points(pointsBuf, {pointCount*3});
+    Buffer<float> weights(weightsBuf, {pointCount});
+#ifdef __riscv
+    idw(input, output);
+#else
+    static Func f("idw");
+    
+    // try {
+    if (!f.defined()) {
+        Var x("x"), y("y");
+        RDom r(0, pointCount);
+        
+        f(x, y) = 0.F;
+        Expr x0 = points(3*r+1);
+        Expr y0 = points(3*r);
+        Expr dx = x - x0;
+        Expr dy = y - y0;
+        Expr weight = weights(r);
+        f(x, y) += hypot(dx, dy) * weight;
+
+        // f.vectorize(r, 8);
+        const int factor = 4; 
+        f.update().atomic().vectorize(r, factor);
+
+        // Compile
+        Target target;
+        target.os = Target::OS::Linux;
+        target.arch = Target::Arch::RISCV;
+        target.bits = 64;
+        target.vector_bits = factor * sizeof(float) * 8;
+
+        // Tested XuanTie C906 has 128-bit vector unit
+        CV_Assert(target.vector_bits <= 128);
+
+        std::vector<Target::Feature> features;
+        features.push_back(Target::RVV);
+        features.push_back(Target::NoAsserts);
+        features.push_back(Target::NoRuntime);
+        target.set_features(features);
+
+        std::cout << target << std::endl;
+        f.print_loop_nest();
+
+        // Dump AOT code
+        f.compile_to_header("idw.h", {}, "idw", target);
+        f.compile_to_assembly("idw.s", {}, "idw", target);
+    }
+    // }
+    // catch (Halide::Error &e) {
+    //     cout << e.what() << '\n';
+    // }
+
+    f.realize(mask);
+#endif
+
+    //
+    float maxVal = -numeric_limits<float>::infinity();
+    float minVal = numeric_limits<float>::infinity();
+    for (int y = 0; y < height; y++) {
+        for (int x = 0; x < width; x++) {
+            if (maskBuf[y*width + x] < minVal) {
+                minVal = maskBuf[y*width + x];
+            }
+            if (maskBuf[y*width + x] > maxVal) {
+                maxVal = maskBuf[y*width + x];
+            }
+        }
+    }
+
+    float diff = maxVal - minVal;
+    for (int y = 0; y < height; y++) {
+        for (int x = 0; x < width; x++) {
+            dst[y*width + x] = (uint8_t) (255 * (maskBuf[y*width + x] - minVal) / diff);
+        }
+    }
+    //
+
+    delete[] maskBuf;
+// #ifdef __riscv
+//     idw(input, output);
+// #else
+//     static Func f("idw");
+//     if (!f.defined()) {
+//         Var x("x"), y("y");
+//         RDom r(0, pointCount);
+        
+//         f(x, y) = 0;
+//         Expr x0 = points(r.x, 0);
+//         Expr y0 = points(r.x, 1);
+//         Expr dx = x - x0;
+//         Expr dy = y - y0;
+//         f(x, y) += sqrt(dx*dx + dy*dy) * weights(r.x);
+//         f(x, y) = cast<uint8_t>(f(x, y));
+        
+//         // Dump AOT code
+//         f.compile_to_header("idw.h", {input}, "idw", target);
+//         f.compile_to_assembly("idw.s", {input}, "idw", target);
+//     }
+// #endif
+}
+
 void idw_ref(const uint8_t* src, uint8_t* dst, int height, int width, int* points, float* weights) {
     float* mask = new float[height * width]();
 
diff --git a/test/test_main.cpp b/test/test_main.cpp
index 49e4035..6d591fd 100644
--- a/test/test_main.cpp
+++ b/test/test_main.cpp
@@ -145,8 +145,9 @@ TEST(convolution_nhwc, halide) {
 #endif  // HAVE_OPENCV_DNN
 
 
-TEST(idw_ref, halide) {
-    Mat src(height, width, CV_8U), dst(height, width, CV_8U), cl_dst(height, width, CV_8UC3);
+TEST(idw, halide) {
+    Mat src(height, width, CV_8U); 
+    Mat dst(height, width, CV_8U), cl_dst(height, width, CV_8UC3), dst_h(height, width, CV_8U), cl_dst_h(height, width, CV_8UC3);
     // randu(src, 0, 256);
 
     // 300 elems (100 points)
@@ -155,10 +156,14 @@ TEST(idw_ref, halide) {
     float weights[] = {0.04053167, -0.07571915, 0.0025228, 0.1974248, -0.14461569, 0.30541419, -0.06167972, -0.04958807, 0.00050641, -0.02023852, -0.05910183, -0.05512609, 0.37617088, -0.15904428, -0.10229038, -0.34355378, 0.39128499, 0.16502984, -0.00916588, -0.05264022, 0.01936275, 0.07156495, -0.35144818, -0.01955447, 0.07818357, 0.10850975, -0.34972095, -0.12186563, 0.04116577, 0.00378971, -0.05382689, 0.12480152, 0.0629668, 0.01070963, 0.03499341, 0.03218783, -0.00559502, -0.21214646, 0.34402987, -0.0246006, -0.01532894, 0.14944613, -0.12848858, -0.04539923, -0.1401344, -0.08143395, 0.18241941, 0.08353522, -0.20102434, 0.05168503, -0.05897654, 0.09669851, -0.13024191, -0.02125988, 0.19499928, 0.12311092, -0.05857943, -0.14954968, -0.19763496, 0.1172914, -0.04584096, 0.20130274, -0.198312, 0.04763513, -0.05524272, -0.07034198, 0.15155036, 0.0718368, -0.09612551, 0.09076872, -0.0863682, 0.08457486, 0.05520895, -0.10551672, 0.12577908, -0.05857014, 0.03529613, 0.01300365, -0.09736055, 0.06580921, -0.03074898, 0.1528181, -0.18295799, 0.18713497, -0.07368328, -0.06431868, 0.02485816, -0.02782657, -0.15182979, 0.06845911, -0.01974711, 0.23508139, -0.23124445, -0.1202987, 0.05380124, 0.04331648, 0.06285662, -0.01171756, 0.01128749, 0.06012665};
 
     idw_ref(NULL, dst.ptr<uint8_t>(), height, width, points, weights);
+    idw_halide(NULL, dst_h.ptr<uint8_t>(), height, width, points, weights);
 
     applyColorMap(dst, cl_dst, COLORMAP_JET);
+    applyColorMap(dst_h, cl_dst_h, COLORMAP_JET);
 
     // imwrite("src.png", src);
     imwrite("res.png", dst);
     imwrite("cres.png", cl_dst);
+    imwrite("res_halide.png", dst_h);
+    imwrite("cres_halide.png", cl_dst_h);
 }
\ No newline at end of file

From 3a6c7a49a63c33b5834eb011e9abb80ff3cb9300 Mon Sep 17 00:00:00 2001
From: uiop07558 <uiop07558@gmail.com>
Date: Wed, 7 Feb 2024 17:33:51 +0300
Subject: [PATCH 04/15] add aot files for idw_halide

---
 aot/idw.h |  56 +++++++
 aot/idw.s | 450 ++++++++++++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 506 insertions(+)
 create mode 100644 aot/idw.h
 create mode 100644 aot/idw.s

diff --git a/aot/idw.h b/aot/idw.h
new file mode 100644
index 0000000..94728e2
--- /dev/null
+++ b/aot/idw.h
@@ -0,0 +1,56 @@
+#ifndef HALIDE__idw_h
+#define HALIDE__idw_h
+#include <stdint.h>
+
+// Forward declarations of the types used in the interface
+// to the Halide pipeline.
+//
+// For the definitions of these structs, include HalideRuntime.h
+
+// Halide's representation of a multi-dimensional array.
+// Halide::Runtime::Buffer is a more user-friendly wrapper
+// around this. Its declaration is in HalideBuffer.h
+struct halide_buffer_t;
+
+// Metadata describing the arguments to the generated function.
+// Used to construct calls to the _argv version of the function.
+struct halide_filter_metadata_t;
+
+#ifndef HALIDE_MUST_USE_RESULT
+#ifdef __has_attribute
+#if __has_attribute(nodiscard)
+#define HALIDE_MUST_USE_RESULT [[nodiscard]]
+#elif __has_attribute(warn_unused_result)
+#define HALIDE_MUST_USE_RESULT __attribute__((warn_unused_result))
+#else
+#define HALIDE_MUST_USE_RESULT
+#endif
+#else
+#define HALIDE_MUST_USE_RESULT
+#endif
+#endif
+
+#ifndef HALIDE_FUNCTION_ATTRS
+#define HALIDE_FUNCTION_ATTRS
+#endif
+
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+HALIDE_FUNCTION_ATTRS
+int idw(struct halide_buffer_t *_idw_buffer);
+
+HALIDE_FUNCTION_ATTRS
+int idw_argv(void **args);
+
+HALIDE_FUNCTION_ATTRS
+const struct halide_filter_metadata_t *idw_metadata();
+
+#ifdef __cplusplus
+}  // extern "C"
+#endif
+
+#endif
diff --git a/aot/idw.s b/aot/idw.s
new file mode 100644
index 0000000..45ac36e
--- /dev/null
+++ b/aot/idw.s
@@ -0,0 +1,450 @@
+	.text
+	.attribute	4, 16
+	.attribute	5, "rv64gcv0p7"
+	.file	"halide_buffer_t.cpp"
+	.section	.text.idw,"ax",@progbits
+	.globl	idw                             # -- Begin function idw
+	.p2align	1
+	.type	idw,@function
+idw:                                    # @idw
+# %bb.0:                                # %entry
+	addi	sp, sp, -192
+	sd	ra, 184(sp)                     # 8-byte Folded Spill
+	sd	s0, 176(sp)                     # 8-byte Folded Spill
+	sd	s1, 168(sp)                     # 8-byte Folded Spill
+	sd	s2, 160(sp)                     # 8-byte Folded Spill
+	sd	s3, 152(sp)                     # 8-byte Folded Spill
+	sd	s4, 144(sp)                     # 8-byte Folded Spill
+	sd	s5, 136(sp)                     # 8-byte Folded Spill
+	sd	s6, 128(sp)                     # 8-byte Folded Spill
+	sd	s7, 120(sp)                     # 8-byte Folded Spill
+	sd	s8, 112(sp)                     # 8-byte Folded Spill
+	sd	s9, 104(sp)                     # 8-byte Folded Spill
+	sd	s10, 96(sp)                     # 8-byte Folded Spill
+	sd	s11, 88(sp)                     # 8-byte Folded Spill
+.Lpcrel_hi0:
+	auipc	a1, %pcrel_hi(.Lb2.buffer)
+	addi	a1, a1, %pcrel_lo(.Lpcrel_hi0)
+	ld	s6, 16(a1)
+	ld	a3, 40(a0)
+.Lpcrel_hi1:
+	auipc	a2, %pcrel_hi(.Lb3.buffer)
+	addi	a2, a2, %pcrel_lo(.Lpcrel_hi1)
+	ld	s9, 16(a2)
+	lw	s4, 0(a3)
+	lwu	s8, 4(a3)
+	lw	a4, 16(a3)
+	sd	a4, 8(sp)                       # 8-byte Folded Spill
+	ld	a4, 0(a1)
+	lwu	a5, 20(a3)
+	sd	a5, 32(sp)                      # 8-byte Folded Spill
+	lw	s1, 24(a3)
+	or	a5, s6, a4
+	mv	a4, s9
+	bnez	a5, .LBB0_2
+# %bb.1:                                # %then_bb
+	lui	a4, 128
+	addiw	a4, a4, 9
+	slli	a4, a4, 13
+	sd	a4, 32(a1)
+	ld	a4, 40(a1)
+	sd	zero, 0(a1)
+	sd	zero, 8(a1)
+	sd	zero, 16(a1)
+	sw	zero, 0(a4)
+	li	a5, 300
+	sw	a5, 4(a4)
+	li	a5, 1
+	sw	a5, 8(a4)
+	sw	zero, 12(a4)
+	ld	a4, 16(a2)
+	sd	zero, 24(a1)
+.LBB0_2:                                # %after_bb
+	ld	a5, 0(a2)
+	ld	s11, 16(a0)
+	or	a4, a4, a5
+	bnez	a4, .LBB0_4
+# %bb.3:                                # %then_bb2
+	sd	zero, 16(a2)
+	sd	zero, 8(a2)
+	sd	zero, 0(a2)
+	lui	a4, 128
+	addiw	a4, a4, 9
+	ld	a5, 40(a2)
+	slli	a4, a4, 13
+	addi	a4, a4, 2
+	sd	a4, 32(a2)
+	sw	zero, 0(a5)
+	li	a4, 100
+	sw	a4, 4(a5)
+	li	a4, 1
+	sw	a4, 8(a5)
+	sw	zero, 12(a5)
+	sd	zero, 24(a2)
+.LBB0_4:                                # %after_bb1
+	sext.w	a4, s8
+	sd	a4, 24(sp)                      # 8-byte Folded Spill
+	lw	a4, 32(sp)                      # 8-byte Folded Reload
+	sd	s1, 16(sp)                      # 8-byte Folded Spill
+	beqz	s11, .LBB0_6
+# %bb.5:
+	li	a0, 0
+	j	.LBB0_9
+.LBB0_6:                                # %_halide_buffer_is_bounds_query.exit36
+	ld	a5, 0(a0)
+	bnez	a5, .LBB0_8
+# %bb.7:                                # %then_bb5
+	sd	zero, 16(a0)
+	sd	zero, 8(a0)
+	sd	zero, 0(a0)
+	lui	a5, 256
+	addiw	a5, a5, 9
+	slli	a5, a5, 13
+	addi	a5, a5, 2
+	sd	a5, 32(a0)
+	sw	s4, 0(a3)
+	ld	s1, 24(sp)                      # 8-byte Folded Reload
+	sw	s1, 4(a3)
+	li	a5, 1
+	sw	a5, 8(a3)
+	sw	zero, 12(a3)
+	ld	a5, 8(sp)                       # 8-byte Folded Reload
+	sw	a5, 16(a3)
+	sw	a4, 20(a3)
+	sw	s1, 24(a3)
+	sw	zero, 28(a3)
+	sd	zero, 24(a0)
+.LBB0_8:                                # %land.rhs.i49
+	ld	a0, 0(a0)
+	seqz	a0, a0
+.LBB0_9:                                # %_halide_buffer_is_bounds_query.exit50
+	ld	a3, 16(a1)
+	ld	a1, 0(a1)
+	ld	a5, 16(a2)
+	ld	a2, 0(a2)
+	or	a1, a1, a3
+	seqz	a1, a1
+	or	a2, a2, a5
+	seqz	a2, a2
+	or	a1, a1, a2
+	or	a0, a0, a1
+	slti	a1, a4, 1
+	or	a0, a0, a1
+	bnez	a0, .LBB0_20
+# %bb.10:                               # %"for idw.s0.y.rebased.preheader"
+	ld	s3, 16(sp)                      # 8-byte Folded Reload
+	ld	a0, 24(sp)                      # 8-byte Folded Reload
+	blez	a0, .LBB0_13
+# %bb.11:                               # %"for idw.s0.y.rebased.us.preheader"
+	li	s0, 0
+	ld	s2, 24(sp)                      # 8-byte Folded Reload
+	slli	s2, s2, 2
+	ld	s1, 32(sp)                      # 8-byte Folded Reload
+.LBB0_12:                               # %"for idw.s0.y.rebased.us"
+                                        # =>This Inner Loop Header: Depth=1
+	slli	a0, s0, 2
+	add	a0, a0, s11
+	li	a1, 0
+	mv	a2, s2
+	call	memset@plt
+	addi	s1, s1, -1
+	addw	s0, s0, s3
+	bnez	s1, .LBB0_12
+.LBB0_13:                               # %"for idw.s1.y.rebased.preheader"
+	sd	zero, 40(sp)                    # 8-byte Folded Spill
+	li	a1, 6
+	li	a2, 1
+	addi	a3, sp, 64
+	addi	a7, sp, 76
+	addi	t0, sp, 72
+	addi	t1, sp, 68
+	li	s0, 4
+	addi	s1, sp, 48
+	addi	t2, sp, 52
+	addi	t3, sp, 60
+	addi	t6, sp, 56
+	li	ra, 2
+	li	a6, 3
+.LBB0_14:                               # %"for idw.s1.y.rebased"
+                                        # =>This Loop Header: Depth=1
+                                        #     Child Loop BB0_16 Depth 2
+                                        #       Child Loop BB0_17 Depth 3
+	ld	a0, 24(sp)                      # 8-byte Folded Reload
+	blez	a0, .LBB0_19
+# %bb.15:                               # %"for idw.s1.x.rebased.preheader"
+                                        #   in Loop: Header=BB0_14 Depth=1
+	li	t4, 0
+	ld	a0, 16(sp)                      # 8-byte Folded Reload
+	ld	a4, 40(sp)                      # 8-byte Folded Reload
+	mulw	t5, a0, a4
+	ld	a0, 8(sp)                       # 8-byte Folded Reload
+	addw	a0, a0, a4
+.LBB0_16:                               # %"for idw.s1.x.rebased"
+                                        #   Parent Loop BB0_14 Depth=1
+                                        # =>  This Loop Header: Depth=2
+                                        #       Child Loop BB0_17 Depth 3
+	add	a4, t4, t5
+	slli	a4, a4, 2
+	add	s2, s11, a4
+	flw	ft0, 0(s2)
+	addw	a4, s4, t4
+	li	s5, 25
+	mv	s3, s6
+	mv	s10, s9
+.LBB0_17:                               # %"for idw.s1.r4$x.r4$x"
+                                        #   Parent Loop BB0_14 Depth=1
+                                        #     Parent Loop BB0_16 Depth=2
+                                        # =>    This Inner Loop Header: Depth=3
+	addi	s7, s3, 4
+	vsetvli	zero, a1, e32, m2
+	vlwu.v	v8, (s7)
+	vsetvli	zero, a2, e32, m2
+	vslidedown.vi	v10, v8, 3
+	addi	a5, s3, 20
+	vsetvli	zero, a1, e32, m2
+	vlwu.v	v12, (a5)
+	vsetvli	zero, a2, e32, m2
+	vslidedown.vi	v14, v12, 2
+	vslidedown.vi	v12, v12, 5
+	vsw.v	v8, (a3)
+	vsw.v	v12, (a7)
+	vsw.v	v14, (t0)
+	vsw.v	v10, (t1)
+	vsetvli	zero, s0, e32, m1
+	vlwu.v	v8, (a3)
+	vsetvli	a5, zero, e32, m1
+	vrsub.vx	v8, v8, a4
+	vsetvli	zero, a1, e32, m2
+	vlwu.v	v10, (s3)
+	vsetvli	zero, a2, e32, m2
+	vslidedown.vi	v12, v10, 3
+	addi	a5, s3, 16
+	vsetvli	zero, a1, e32, m2
+	vlwu.v	v14, (a5)
+	vsetvli	zero, a2, e32, m2
+	vslidedown.vi	v16, v14, 2
+	vslidedown.vi	v14, v14, 5
+	vsw.v	v10, (s1)
+	vsw.v	v12, (t2)
+	vsw.v	v14, (t3)
+	vsw.v	v16, (t6)
+	vsetvli	zero, s0, e32, m1
+	vlwu.v	v9, (s1)
+	vsetvli	a5, zero, e32, m1
+	vrsub.vx	v9, v9, a0
+	vmul.vv	v9, v9, v9
+	vmacc.vv	v9, v8, v8
+	vfcvt.f.x.v	v8, v9
+	vfmv.f.s	ft1, v8
+	fsqrt.s	ft1, ft1
+	vfmv.s.f	v9, ft1
+	vsetvli	zero, a2, e32, m1
+	vslidedown.vi	v10, v8, 1
+	vfmv.f.s	ft1, v10
+	fsqrt.s	ft1, ft1
+	vsetvli	a5, zero, e32, m1
+	vfmv.s.f	v10, ft1
+	vsetvli	zero, ra, e32, m1
+	vslideup.vi	v9, v10, 1
+	vsetvli	zero, a2, e32, m1
+	vslidedown.vi	v10, v8, 2
+	vfmv.f.s	ft1, v10
+	fsqrt.s	ft1, ft1
+	vsetvli	a5, zero, e32, m1
+	vfmv.s.f	v10, ft1
+	vsetvli	zero, a6, e32, m1
+	vslideup.vi	v9, v10, 2
+	vsetvli	zero, a2, e32, m1
+	vslidedown.vi	v8, v8, 3
+	vfmv.f.s	ft1, v8
+	fsqrt.s	ft1, ft1
+	vsetvli	a5, zero, e32, m1
+	vfmv.s.f	v8, ft1
+	vsetvli	zero, s0, e32, m1
+	vslideup.vi	v9, v8, 3
+	vsetvli	zero, zero, e32, m1
+	vlwu.v	v8, (s10)
+	vfmul.vv	v8, v9, v8
+	vfmv.s.f	v9, ft0
+	vfredusum.vs	v8, v8, v9
+	vfmv.f.s	ft0, v8
+	addi	s5, s5, -1
+	addi	s10, s10, 16
+	addi	s3, s3, 48
+	bnez	s5, .LBB0_17
+# %bb.18:                               # %"end for idw.s1.r4$x.r4$x"
+                                        #   in Loop: Header=BB0_16 Depth=2
+	addi	t4, t4, 1
+	fsw	ft0, 0(s2)
+	bne	t4, s8, .LBB0_16
+.LBB0_19:                               # %"end for idw.s1.x.rebased"
+                                        #   in Loop: Header=BB0_14 Depth=1
+	ld	a4, 40(sp)                      # 8-byte Folded Reload
+	addi	a4, a4, 1
+	ld	a0, 32(sp)                      # 8-byte Folded Reload
+	sd	a4, 40(sp)                      # 8-byte Folded Spill
+	bne	a4, a0, .LBB0_14
+.LBB0_20:                               # %destructor_block
+	li	a0, 0
+	ld	ra, 184(sp)                     # 8-byte Folded Reload
+	ld	s0, 176(sp)                     # 8-byte Folded Reload
+	ld	s1, 168(sp)                     # 8-byte Folded Reload
+	ld	s2, 160(sp)                     # 8-byte Folded Reload
+	ld	s3, 152(sp)                     # 8-byte Folded Reload
+	ld	s4, 144(sp)                     # 8-byte Folded Reload
+	ld	s5, 136(sp)                     # 8-byte Folded Reload
+	ld	s6, 128(sp)                     # 8-byte Folded Reload
+	ld	s7, 120(sp)                     # 8-byte Folded Reload
+	ld	s8, 112(sp)                     # 8-byte Folded Reload
+	ld	s9, 104(sp)                     # 8-byte Folded Reload
+	ld	s10, 96(sp)                     # 8-byte Folded Reload
+	ld	s11, 88(sp)                     # 8-byte Folded Reload
+	addi	sp, sp, 192
+	ret
+.Lfunc_end0:
+	.size	idw, .Lfunc_end0-idw
+                                        # -- End function
+	.section	.text.idw_argv,"ax",@progbits
+	.globl	idw_argv                        # -- Begin function idw_argv
+	.p2align	1
+	.type	idw_argv,@function
+idw_argv:                               # @idw_argv
+# %bb.0:                                # %entry
+	addi	sp, sp, -16
+	sd	ra, 8(sp)                       # 8-byte Folded Spill
+	ld	a0, 0(a0)
+	call	idw@plt
+	li	a0, 0
+	ld	ra, 8(sp)                       # 8-byte Folded Reload
+	addi	sp, sp, 16
+	ret
+.Lfunc_end1:
+	.size	idw_argv, .Lfunc_end1-idw_argv
+                                        # -- End function
+	.section	.text.idw_metadata,"ax",@progbits
+	.globl	idw_metadata                    # -- Begin function idw_metadata
+	.p2align	1
+	.type	idw_metadata,@function
+idw_metadata:                           # @idw_metadata
+# %bb.0:                                # %entry
+.Lpcrel_hi2:
+	auipc	a0, %pcrel_hi(.Lidw_metadata_storage)
+	addi	a0, a0, %pcrel_lo(.Lpcrel_hi2)
+	ret
+.Lfunc_end2:
+	.size	idw_metadata, .Lfunc_end2-idw_metadata
+                                        # -- End function
+	.type	.Lb2.shape,@object              # @b2.shape
+	.section	.rodata,"a",@progbits
+	.p2align	5, 0x0
+.Lb2.shape:
+	.asciz	"\000\000\000\000,\001\000\000\001\000\000\000\000\000\000"
+	.size	.Lb2.shape, 16
+
+	.type	.Lb2.data,@object               # @b2.data
+	.p2align	5, 0x0
+.Lb2.data:
+	.asciz	"\000\000\000\000\000\000\000\000H\000\000\000\000\000\000\000\325\000\000\000O\000\000\000\000\000\000\000\252\001\000\000<\000\000\000\000\000\000\000\200\002\000\000L\000\000\000\000\000\000\000U\003\000\000\200\000\000\000\000\000\000\000*\004\000\000C\000\000\000\000\000\000\000\000\005\000\000A\000\000\000\000\000\000\000\325\005\000\000@\000\000\000\000\000\000\000\252\006\000\000<\000\000\000\000\000\000\000\200\007\000\000=\000\000\000x\000\000\000\000\000\000\000Q\000\000\000x\000\000\000\325\000\000\000O\000\000\000x\000\000\000\252\001\000\000:\000\000\000x\000\000\000\200\002\000\000\204\000\000\000x\000\000\000U\003\000\000\225\000\000\000x\000\000\000*\004\000\000\216\000\000\000x\000\000\000\000\005\000\000@\000\000\000x\000\000\000\325\005\000\000E\000\000\000x\000\000\000\252\006\000\000A\000\000\000x\000\000\000\200\007\000\000@\000\000\000\360\000\000\000\000\000\000\000K\000\000\000\360\000\000\000\325\000\000\000D\000\000\000\360\000\000\000\252\001\000\000\214\000\000\000\360\000\000\000\200\002\000\000\231\000\000\000\360\000\000\000U\003\000\000\221\000\000\000\360\000\000\000*\004\000\000\204\000\000\000\360\000\000\000\000\005\000\000\230\000\000\000\360\000\000\000\325\005\000\000}\000\000\000\360\000\000\000\252\006\000\000B\000\000\000\360\000\000\000\200\007\000\000:\000\000\000h\001\000\000\000\000\000\000N\000\000\000h\001\000\000\325\000\000\000<\000\000\000h\001\000\000\252\001\000\000\213\000\000\000h\001\000\000\200\002\000\000\250\000\000\000h\001\000\000U\003\000\000\232\000\000\000h\001\000\000*\004\000\000\212\000\000\000h\001\000\000\000\005\000\000\221\000\000\000h\001\000\000\325\005\000\000\240\000\000\000h\001\000\000\252\006\000\000D\000\000\000h\001\000\000\200\007\000\000<\000\000\000\340\001\000\000\000\000\000\000M\000\000\000\340\001\000\000\325\000\000\000;\000\000\000\340\001\000\000\252\001\000\000\245\000\000\000\340\001\000\000\200\002\000\000\267\000\000\000\340\001\000\000U\003\000\000\246\000\000\000\340\001\000\000*\004\000\000\216\000\000\000\340\001\000\000\000\005\000\000{\000\000\000\340\001\000\000\325\005\000\000\233\000\000\000\340\001\000\000\252\006\000\000\220\000\000\000\340\001\000\000\200\007\000\000<\000\000\000X\002\000\000\000\000\000\000S\000\000\000X\002\000\000\325\000\000\000A\000\000\000X\002\000\000\252\001\000\000\262\000\000\000X\002\000\000\200\002\000\000\270\000\000\000X\002\000\000U\003\000\000\212\000\000\000X\002\000\000*\004\000\000|\000\000\000X\002\000\000\000\005\000\000\204\000\000\000X\002\000\000\325\005\000\000\257\000\000\000X\002\000\000\252\006\000\000\257\000\000\000X\002\000\000\200\007\000\000<\000\000\000\320\002\000\000\000\000\000\000V\000\000\000\320\002\000\000\325\000\000\000<\000\000\000\320\002\000\000\252\001\000\000\266\000\000\000\320\002\000\000\200\002\000\000\263\000\000\000\320\002\000\000U\003\000\000\230\000\000\000\320\002\000\000*\004\000\000\214\000\000\000\320\002\000\000\000\005\000\000s\000\000\000\320\002\000\000\325\005\000\000\234\000\000\000\320\002\000\000\252\006\000\000\254\000\000\000\320\002\000\000\200\007\000\000A\000\000\000H\003\000\000\000\000\000\000Z\000\000\000H\003\000\000\325\000\000\000E\000\000\000H\003\000\000\252\001\000\000\244\000\000\000H\003\000\000\200\002\000\000\272\000\000\000H\003\000\000U\003\000\000\222\000\000\000H\003\000\000*\004\000\000\223\000\000\000H\003\000\000\000\005\000\000z\000\000\000H\003\000\000\325\005\000\000\230\000\000\000H\003\000\000\252\006\000\000\246\000\000\000H\003\000\000\200\007\000\000E\000\000\000\300\003\000\000\000\000\000\000U\000\000\000\300\003\000\000\325\000\000\000C\000\000\000\300\003\000\000\252\001\000\000\272\000\000\000\300\003\000\000\200\002\000\000\251\000\000\000\300\003\000\000U\003\000\000\242\000\000\000\300\003\000\000*\004\000\000\223\000\000\000\300\003\000\000\000\005\000\000|\000\000\000\300\003\000\000\325\005\000\000\225\000\000\000\300\003\000\000\252\006\000\000\235\000\000\000\300\003\000\000\200\007\000\000E\000\000\0008\004\000\000\000\000\000\000U\000\000\0008\004\000\000\325\000\000\000F\000\000\0008\004\000\000\252\001\000\000\301\000\000\0008\004\000\000\200\002\000\000\300\000\000\0008\004\000\000U\003\000\000\233\000\000\0008\004\000\000*\004\000\000\207\000\000\0008\004\000\000\000\005\000\000x\000\000\0008\004\000\000\325\005\000\000\212\000\000\0008\004\000\000\252\006\000\000\200\000\000\0008\004\000\000\200\007\000\000F\000\000"
+	.size	.Lb2.data, 1200
+
+	.type	.Lb2.buffer,@object             # @b2.buffer
+	.data
+	.p2align	4, 0x0
+.Lb2.buffer:
+	.quad	0                               # 0x0
+	.quad	0
+	.quad	.Lb2.data
+	.quad	1                               # 0x1
+	.byte	0                               # 0x0
+	.byte	32                              # 0x20
+	.half	1                               # 0x1
+	.word	1                               # 0x1
+	.quad	.Lb2.shape
+	.quad	0
+	.size	.Lb2.buffer, 56
+
+	.type	.Lb3.shape,@object              # @b3.shape
+	.section	.rodata,"a",@progbits
+	.p2align	5, 0x0
+.Lb3.shape:
+	.asciz	"\000\000\000\000d\000\000\000\001\000\000\000\000\000\000"
+	.size	.Lb3.shape, 16
+
+	.type	.Lb3.data,@object               # @b3.data
+	.p2align	5, 0x0
+.Lb3.data:
+	.ascii	"\211\004&=\244\022\233\275\217U%;\272)J>#\026\024\276@_\234>\340\243|\275\334\034K\275\232\300\004:A\313\245\274\303\024r\275\345\313a\275x\231\300>\201\334\"\276\236}\321\275H\346\257\276\202V\310>\225\375(>},\026\274E\235W\275\241\236\236<\245\220\222=\004\361\263\276\2620\240\274\265\036\240=\\:\336=\240\016\263\276\260\224\371\275p\235(=\311\\x;\226y\\\275\360\227\377=\275\364\200=rw/<@U\017=c\327\003=nV\267\273\354<Y\276\257$\260>3\207\311\274<&{\274h\b\031>\203\222\003\276\213\3649\275d\177\017\276\330\306\246\275'\314:>\203\024\253=S\331M\276\257\263S=b\221q\275\336\t\306=#^\005\2763)\256\274\344\255G>\224!\374=\374\360o\275\215#\031\276\322`J\276y6\360=\273\303;\275N\"N>M\022K\276\016\035C=1Fb\275u\017\220\275\0040\033>,\037\223=t\335\304\275\363\344\271=\320\341\260\275\2265\255=\310\"b=&\031\330\275;\314\000>>\347o\275\255\222\020=C\rU<\370d\307\275\373\306\206=I\345\373\274Y|\034>WY;\276O\240?>B\347\226\275\203\271\203\275W\243\313<\214\364\343\274Ey\033\276J4\214=\261\304\241\274-\271p>X\313l\276*_\366\275\260^\\=\237l1=\371\272\200=\002\373?\274*\3578<]Gv="
+	.size	.Lb3.data, 400
+
+	.type	.Lb3.buffer,@object             # @b3.buffer
+	.data
+	.p2align	4, 0x0
+.Lb3.buffer:
+	.quad	0                               # 0x0
+	.quad	0
+	.quad	.Lb3.data
+	.quad	1                               # 0x1
+	.byte	2                               # 0x2
+	.byte	32                              # 0x20
+	.half	1                               # 0x1
+	.word	1                               # 0x1
+	.quad	.Lb3.shape
+	.quad	0
+	.size	.Lb3.buffer, 56
+
+	.type	.L__unnamed_1,@object           # @0
+	.section	.rodata,"a",@progbits
+	.p2align	4, 0x0
+.L__unnamed_1:
+	.zero	32
+	.size	.L__unnamed_1, 32
+
+	.type	.Lstr,@object                   # @str
+	.p2align	5, 0x0
+.Lstr:
+	.asciz	"idw"
+	.size	.Lstr, 4
+
+	.type	.L__unnamed_2,@object           # @1
+	.section	.data.rel.ro,"aw",@progbits
+	.p2align	4, 0x0
+.L__unnamed_2:
+	.quad	.Lstr
+	.word	2                               # 0x2
+	.word	2                               # 0x2
+	.byte	2                               # 0x2
+	.byte	32                              # 0x20
+	.half	1                               # 0x1
+	.zero	4
+	.quad	0
+	.quad	0
+	.quad	0
+	.quad	0
+	.quad	.L__unnamed_1
+	.size	.L__unnamed_2, 64
+
+	.type	.Lstr.4,@object                 # @str.4
+	.section	.rodata,"a",@progbits
+	.p2align	5, 0x0
+.Lstr.4:
+	.asciz	"riscv-64-linux-no_asserts-no_runtime-rvv-vector_bits_128"
+	.size	.Lstr.4, 57
+
+	.type	.Lidw_metadata_storage,@object  # @idw_metadata_storage
+	.section	.data.rel.ro,"aw",@progbits
+	.p2align	4, 0x0
+.Lidw_metadata_storage:
+	.word	1                               # 0x1
+	.word	1                               # 0x1
+	.quad	.L__unnamed_2
+	.quad	.Lstr.4
+	.quad	.Lstr
+	.size	.Lidw_metadata_storage, 32
+
+	.ident	"clang version 16.0.0 (https://github.com/dkurt/llvm-rvv-071 b027aa1b59c9f53240bdc836f39656723fdf9df0)"
+	.ident	"clang version 16.0.0 (https://github.com/dkurt/llvm-rvv-071 b027aa1b59c9f53240bdc836f39656723fdf9df0)"
+	.ident	"clang version 16.0.0 (https://github.com/dkurt/llvm-rvv-071 b027aa1b59c9f53240bdc836f39656723fdf9df0)"
+	.ident	"clang version 16.0.0 (https://github.com/dkurt/llvm-rvv-071 b027aa1b59c9f53240bdc836f39656723fdf9df0)"
+	.ident	"clang version 16.0.0 (https://github.com/dkurt/llvm-rvv-071 b027aa1b59c9f53240bdc836f39656723fdf9df0)"
+	.ident	"clang version 16.0.0 (https://github.com/dkurt/llvm-rvv-071 b027aa1b59c9f53240bdc836f39656723fdf9df0)"
+	.section	".note.GNU-stack","",@progbits

From 1efccdffdc3a37b1f482f3674f36f935ff718f1e Mon Sep 17 00:00:00 2001
From: uiop07558 <uiop07558@gmail.com>
Date: Wed, 7 Feb 2024 17:54:23 +0300
Subject: [PATCH 05/15] hardcoded maxval and minval

---
 .vscode/settings.json |  3 ++-
 src/idw.cpp           | 40 ++++++++++++++++++++--------------------
 2 files changed, 22 insertions(+), 21 deletions(-)

diff --git a/.vscode/settings.json b/.vscode/settings.json
index 8104e93..9b18adb 100644
--- a/.vscode/settings.json
+++ b/.vscode/settings.json
@@ -4,6 +4,7 @@
         "limits": "cpp",
         "iostream": "cpp",
         "cmath": "cpp",
-        "optional": "cpp"
+        "optional": "cpp",
+        "new": "cpp"
     }
 }
\ No newline at end of file
diff --git a/src/idw.cpp b/src/idw.cpp
index 7ef9887..7a08ac8 100644
--- a/src/idw.cpp
+++ b/src/idw.cpp
@@ -80,18 +80,18 @@ void idw_halide(const uint8_t* src, uint8_t* dst, int height, int width, int* po
 #endif
 
     //
-    float maxVal = -numeric_limits<float>::infinity();
-    float minVal = numeric_limits<float>::infinity();
-    for (int y = 0; y < height; y++) {
-        for (int x = 0; x < width; x++) {
-            if (maskBuf[y*width + x] < minVal) {
-                minVal = maskBuf[y*width + x];
-            }
-            if (maskBuf[y*width + x] > maxVal) {
-                maxVal = maskBuf[y*width + x];
-            }
-        }
-    }
+    float maxVal = 193.0 //-numeric_limits<float>::infinity();
+    float minVal = 58.0 //numeric_limits<float>::infinity();
+    // for (int y = 0; y < height; y++) {
+    //     for (int x = 0; x < width; x++) {
+    //         if (maskBuf[y*width + x] < minVal) {
+    //             minVal = maskBuf[y*width + x];
+    //         }
+    //         if (maskBuf[y*width + x] > maxVal) {
+    //             maxVal = maskBuf[y*width + x];
+    //         }
+    //     }
+    // }
 
     float diff = maxVal - minVal;
     for (int y = 0; y < height; y++) {
@@ -128,8 +128,8 @@ void idw_halide(const uint8_t* src, uint8_t* dst, int height, int width, int* po
 void idw_ref(const uint8_t* src, uint8_t* dst, int height, int width, int* points, float* weights) {
     float* mask = new float[height * width]();
 
-    float maxVal = -numeric_limits<float>::infinity();
-    float minVal = numeric_limits<float>::infinity();
+    float maxVal = 193.0 //-numeric_limits<float>::infinity();
+    float minVal = 58.0 //numeric_limits<float>::infinity();
 
     for (int y = 0; y < height; y++) {
         for (int x = 0; x < width; x++) {
@@ -144,12 +144,12 @@ void idw_ref(const uint8_t* src, uint8_t* dst, int height, int width, int* point
             }
 
             mask[y*width + x] = dot;
-            if (dot < minVal) {
-                minVal = dot;
-            }
-            if (dot > maxVal) {
-                maxVal = dot;
-            }
+            // if (dot < minVal) {
+            //     minVal = dot;
+            // }
+            // if (dot > maxVal) {
+            //     maxVal = dot;
+            // }
         }
     }
 

From bda6cc0b8e8dda9f2352ee653c061126ae1a8113 Mon Sep 17 00:00:00 2001
From: uiop07558 <uiop07558@gmail.com>
Date: Tue, 6 Feb 2024 17:03:11 +0300
Subject: [PATCH 06/15] add reference implementation for IDW

---
 .vscode/settings.json |  3 +++
 src/bgr2gray.cpp      | 47 ++++++++++++++++++++++---------------------
 2 files changed, 27 insertions(+), 23 deletions(-)
 create mode 100644 .vscode/settings.json

diff --git a/.vscode/settings.json b/.vscode/settings.json
new file mode 100644
index 0000000..70e34ec
--- /dev/null
+++ b/.vscode/settings.json
@@ -0,0 +1,3 @@
+{
+    "C_Cpp.errorSquiggles": "disabled"
+}
\ No newline at end of file
diff --git a/src/bgr2gray.cpp b/src/bgr2gray.cpp
index 67049db..ad7ef5d 100644
--- a/src/bgr2gray.cpp
+++ b/src/bgr2gray.cpp
@@ -60,29 +60,30 @@ void bgr2gray_interleaved_halide(uint8_t* src, uint8_t* dst, int height, int wid
 
         int factor = 16;
         f.vectorize(x, factor);
-
-        // Compile
-        Target target;
-        target.os = Target::OS::Linux;
-        target.arch = Target::Arch::RISCV;
-        target.bits = 64;
-        target.vector_bits = factor * sizeof(uint8_t) * 8;
-
-        // Tested XuanTie C906 has 128-bit vector unit
-        CV_Assert(target.vector_bits <= 128);
-
-        std::vector<Target::Feature> features;
-        features.push_back(Target::RVV);
-        features.push_back(Target::NoAsserts);
-        features.push_back(Target::NoRuntime);
-        target.set_features(features);
-
-        std::cout << target << std::endl;
-        f.print_loop_nest();
-
-        // Dump AOT code
-        f.compile_to_header("bgr2gray_interleaved.h", {input}, "bgr2gray_interleaved", target);
-        f.compile_to_assembly("bgr2gray_interleaved.s", {input}, "bgr2gray_interleaved", target);
+        f.realize(output);
+
+    //     // Compile
+    //     Target target;
+    //     target.os = Target::OS::Linux;
+    //     target.arch = Target::Arch::RISCV;
+    //     target.bits = 64;
+    //     target.vector_bits = factor * sizeof(uint8_t) * 8;
+
+    //     // Tested XuanTie C906 has 128-bit vector unit
+    //     CV_Assert(target.vector_bits <= 128);
+
+    //     std::vector<Target::Feature> features;
+    //     features.push_back(Target::RVV);
+    //     features.push_back(Target::NoAsserts);
+    //     features.push_back(Target::NoRuntime);
+    //     target.set_features(features);
+
+    //     std::cout << target << std::endl;
+    //     f.print_loop_nest();
+
+    //     // Dump AOT code
+    //     f.compile_to_header("bgr2gray_interleaved.h", {input}, "bgr2gray_interleaved", target);
+    //     f.compile_to_assembly("bgr2gray_interleaved.s", {input}, "bgr2gray_interleaved", target);
     }
 #endif
 }

From 086ea7b5acb670ea911ec94aec2242b16ab093f3 Mon Sep 17 00:00:00 2001
From: Dmitry Kurtaev <dmitry.kurtaev@gmail.com>
Date: Tue, 6 Feb 2024 22:20:13 +0300
Subject: [PATCH 07/15] Minor fixes for reference IDW

---
 .vscode/settings.json |  3 ---
 CMakeLists.txt        |  1 -
 src/bgr2gray.cpp      | 47 +++++++++++++++++++++----------------------
 3 files changed, 23 insertions(+), 28 deletions(-)
 delete mode 100644 .vscode/settings.json

diff --git a/.vscode/settings.json b/.vscode/settings.json
deleted file mode 100644
index 70e34ec..0000000
--- a/.vscode/settings.json
+++ /dev/null
@@ -1,3 +0,0 @@
-{
-    "C_Cpp.errorSquiggles": "disabled"
-}
\ No newline at end of file
diff --git a/CMakeLists.txt b/CMakeLists.txt
index a20e3f2..b4ee5cc 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -116,7 +116,6 @@ add_executable(test_algo ${PERF_SOURCES})
 target_link_libraries(test_algo
   algos
   "${binary_dir}/lib/libopencv_ts.a"
-  "${binary_dir}/install/lib/libopencv_imgcodecs.so"
   pthread
 )
 target_include_directories(test_algo PUBLIC
diff --git a/src/bgr2gray.cpp b/src/bgr2gray.cpp
index ad7ef5d..67049db 100644
--- a/src/bgr2gray.cpp
+++ b/src/bgr2gray.cpp
@@ -60,30 +60,29 @@ void bgr2gray_interleaved_halide(uint8_t* src, uint8_t* dst, int height, int wid
 
         int factor = 16;
         f.vectorize(x, factor);
-        f.realize(output);
-
-    //     // Compile
-    //     Target target;
-    //     target.os = Target::OS::Linux;
-    //     target.arch = Target::Arch::RISCV;
-    //     target.bits = 64;
-    //     target.vector_bits = factor * sizeof(uint8_t) * 8;
-
-    //     // Tested XuanTie C906 has 128-bit vector unit
-    //     CV_Assert(target.vector_bits <= 128);
-
-    //     std::vector<Target::Feature> features;
-    //     features.push_back(Target::RVV);
-    //     features.push_back(Target::NoAsserts);
-    //     features.push_back(Target::NoRuntime);
-    //     target.set_features(features);
-
-    //     std::cout << target << std::endl;
-    //     f.print_loop_nest();
-
-    //     // Dump AOT code
-    //     f.compile_to_header("bgr2gray_interleaved.h", {input}, "bgr2gray_interleaved", target);
-    //     f.compile_to_assembly("bgr2gray_interleaved.s", {input}, "bgr2gray_interleaved", target);
+
+        // Compile
+        Target target;
+        target.os = Target::OS::Linux;
+        target.arch = Target::Arch::RISCV;
+        target.bits = 64;
+        target.vector_bits = factor * sizeof(uint8_t) * 8;
+
+        // Tested XuanTie C906 has 128-bit vector unit
+        CV_Assert(target.vector_bits <= 128);
+
+        std::vector<Target::Feature> features;
+        features.push_back(Target::RVV);
+        features.push_back(Target::NoAsserts);
+        features.push_back(Target::NoRuntime);
+        target.set_features(features);
+
+        std::cout << target << std::endl;
+        f.print_loop_nest();
+
+        // Dump AOT code
+        f.compile_to_header("bgr2gray_interleaved.h", {input}, "bgr2gray_interleaved", target);
+        f.compile_to_assembly("bgr2gray_interleaved.s", {input}, "bgr2gray_interleaved", target);
     }
 #endif
 }

From dde043eea1b06e786c536969a1c40e65f3f9f095 Mon Sep 17 00:00:00 2001
From: uiop07558 <uiop07558@gmail.com>
Date: Wed, 7 Feb 2024 17:27:08 +0300
Subject: [PATCH 08/15] add idw_halide

---
 .vscode/c_cpp_properties.json |  16 +++++
 .vscode/settings.json         |   9 +++
 include/algos.hpp             |   3 +-
 perf/perf_main.cpp            |  43 +++++++++++++
 src/idw.cpp                   | 117 ++++++++++++++++++++++++++++++++++
 test/test_main.cpp            |   9 ++-
 6 files changed, 194 insertions(+), 3 deletions(-)
 create mode 100644 .vscode/c_cpp_properties.json
 create mode 100644 .vscode/settings.json

diff --git a/.vscode/c_cpp_properties.json b/.vscode/c_cpp_properties.json
new file mode 100644
index 0000000..4039bef
--- /dev/null
+++ b/.vscode/c_cpp_properties.json
@@ -0,0 +1,16 @@
+{
+    "configurations": [
+        {
+            "name": "Linux",
+            "includePath": [
+                "${workspaceFolder}/**"
+            ],
+            "defines": [],
+            "compilerPath": "/usr/bin/gcc",
+            "cStandard": "c17",
+            "cppStandard": "gnu++17",
+            "intelliSenseMode": "linux-gcc-x64"
+        }
+    ],
+    "version": 4
+}
\ No newline at end of file
diff --git a/.vscode/settings.json b/.vscode/settings.json
new file mode 100644
index 0000000..8104e93
--- /dev/null
+++ b/.vscode/settings.json
@@ -0,0 +1,9 @@
+{
+    "C_Cpp.errorSquiggles": "disabled",
+    "files.associations": {
+        "limits": "cpp",
+        "iostream": "cpp",
+        "cmath": "cpp",
+        "optional": "cpp"
+    }
+}
\ No newline at end of file
diff --git a/include/algos.hpp b/include/algos.hpp
index 7a291b9..9982b03 100644
--- a/include/algos.hpp
+++ b/include/algos.hpp
@@ -36,5 +36,6 @@ void convolution_opencv(const cv::Mat& src, const cv::Mat& weights, cv::Mat& dst
                         int inpChannels, int outChannels);
 #endif  // HAVE_OPENCV_DNN
 
-
 void idw_ref(const uint8_t* src, uint8_t* dst, int height, int width, int* points, float* weights);
+
+void idw_halide(const uint8_t* src, uint8_t* dst, int height, int width, int* pointsBuf, float* weightsBuf);
\ No newline at end of file
diff --git a/perf/perf_main.cpp b/perf/perf_main.cpp
index 4a4f4f2..9952623 100644
--- a/perf/perf_main.cpp
+++ b/perf/perf_main.cpp
@@ -14,8 +14,14 @@
 
 #include "algos.hpp"
 
+#include <Halide.h>
+
+// #include <iostream>
+
 using namespace cv;
 
+// using namespace std;
+
 CV_PERF_TEST_MAIN("")
 
 Mat src(1080, 1920, CV_8UC3);
@@ -204,3 +210,40 @@ PERF_TEST(convolution_nhwc, halide) {
 }
 
 #endif  // HAVE_OPENCV_DNN
+
+
+PERF_TEST(idw, reference) {
+    const int width = 1920;
+    const int height = 1080;
+
+    // 300 elems (100 points)
+    int points[] = {0, 0, 72, 0, 213, 79, 0, 426, 60, 0, 640, 76, 0, 853, 128, 0, 1066, 67, 0, 1280, 65, 0, 1493, 64, 0, 1706, 60, 0, 1920, 61, 120, 0, 81, 120, 213, 79, 120, 426, 58, 120, 640, 132, 120, 853, 149, 120, 1066, 142, 120, 1280, 64, 120, 1493, 69, 120, 1706, 65, 120, 1920, 64, 240, 0, 75, 240, 213, 68, 240, 426, 140, 240, 640, 153, 240, 853, 145, 240, 1066, 132, 240, 1280, 152, 240, 1493, 125, 240, 1706, 66, 240, 1920, 58, 360, 0, 78, 360, 213, 60, 360, 426, 139, 360, 640, 168, 360, 853, 154, 360, 1066, 138, 360, 1280, 145, 360, 1493, 160, 360, 1706, 68, 360, 1920, 60, 480, 0, 77, 480, 213, 59, 480, 426, 165, 480, 640, 183, 480, 853, 166, 480, 1066, 142, 480, 1280, 123, 480, 1493, 155, 480, 1706, 144, 480, 1920, 60, 600, 0, 83, 600, 213, 65, 600, 426, 178, 600, 640, 184, 600, 853, 138, 600, 1066, 124, 600, 1280, 132, 600, 1493, 175, 600, 1706, 175, 600, 1920, 60, 720, 0, 86, 720, 213, 60, 720, 426, 182, 720, 640, 179, 720, 853, 152, 720, 1066, 140, 720, 1280, 115, 720, 1493, 156, 720, 1706, 172, 720, 1920, 65, 840, 0, 90, 840, 213, 69, 840, 426, 164, 840, 640, 186, 840, 853, 146, 840, 1066, 147, 840, 1280, 122, 840, 1493, 152, 840, 1706, 166, 840, 1920, 69, 960, 0, 85, 960, 213, 67, 960, 426, 186, 960, 640, 169, 960, 853, 162, 960, 1066, 147, 960, 1280, 124, 960, 1493, 149, 960, 1706, 157, 960, 1920, 69, 1080, 0, 85, 1080, 213, 70, 1080, 426, 193, 1080, 640, 192, 1080, 853, 155, 1080, 1066, 135, 1080, 1280, 120, 1080, 1493, 138, 1080, 1706, 128, 1080, 1920, 70};
+    // 100 elems
+    float weights[] = {0.04053167, -0.07571915, 0.0025228, 0.1974248, -0.14461569, 0.30541419, -0.06167972, -0.04958807, 0.00050641, -0.02023852, -0.05910183, -0.05512609, 0.37617088, -0.15904428, -0.10229038, -0.34355378, 0.39128499, 0.16502984, -0.00916588, -0.05264022, 0.01936275, 0.07156495, -0.35144818, -0.01955447, 0.07818357, 0.10850975, -0.34972095, -0.12186563, 0.04116577, 0.00378971, -0.05382689, 0.12480152, 0.0629668, 0.01070963, 0.03499341, 0.03218783, -0.00559502, -0.21214646, 0.34402987, -0.0246006, -0.01532894, 0.14944613, -0.12848858, -0.04539923, -0.1401344, -0.08143395, 0.18241941, 0.08353522, -0.20102434, 0.05168503, -0.05897654, 0.09669851, -0.13024191, -0.02125988, 0.19499928, 0.12311092, -0.05857943, -0.14954968, -0.19763496, 0.1172914, -0.04584096, 0.20130274, -0.198312, 0.04763513, -0.05524272, -0.07034198, 0.15155036, 0.0718368, -0.09612551, 0.09076872, -0.0863682, 0.08457486, 0.05520895, -0.10551672, 0.12577908, -0.05857014, 0.03529613, 0.01300365, -0.09736055, 0.06580921, -0.03074898, 0.1528181, -0.18295799, 0.18713497, -0.07368328, -0.06431868, 0.02485816, -0.02782657, -0.15182979, 0.06845911, -0.01974711, 0.23508139, -0.23124445, -0.1202987, 0.05380124, 0.04331648, 0.06285662, -0.01171756, 0.01128749, 0.06012665};
+
+    Mat dst(height, width, CV_8U);
+
+    PERF_SAMPLE_BEGIN()
+        idw_ref(NULL, dst.ptr<uint8_t>(), height, width, points, weights);
+    PERF_SAMPLE_END()
+
+    SANITY_CHECK_NOTHING();
+}
+
+PERF_TEST(idw, halide) {
+    const int width = 1920;
+    const int height = 1080;
+
+    // 300 elems (100 points)
+    int points[] = {0, 0, 72, 0, 213, 79, 0, 426, 60, 0, 640, 76, 0, 853, 128, 0, 1066, 67, 0, 1280, 65, 0, 1493, 64, 0, 1706, 60, 0, 1920, 61, 120, 0, 81, 120, 213, 79, 120, 426, 58, 120, 640, 132, 120, 853, 149, 120, 1066, 142, 120, 1280, 64, 120, 1493, 69, 120, 1706, 65, 120, 1920, 64, 240, 0, 75, 240, 213, 68, 240, 426, 140, 240, 640, 153, 240, 853, 145, 240, 1066, 132, 240, 1280, 152, 240, 1493, 125, 240, 1706, 66, 240, 1920, 58, 360, 0, 78, 360, 213, 60, 360, 426, 139, 360, 640, 168, 360, 853, 154, 360, 1066, 138, 360, 1280, 145, 360, 1493, 160, 360, 1706, 68, 360, 1920, 60, 480, 0, 77, 480, 213, 59, 480, 426, 165, 480, 640, 183, 480, 853, 166, 480, 1066, 142, 480, 1280, 123, 480, 1493, 155, 480, 1706, 144, 480, 1920, 60, 600, 0, 83, 600, 213, 65, 600, 426, 178, 600, 640, 184, 600, 853, 138, 600, 1066, 124, 600, 1280, 132, 600, 1493, 175, 600, 1706, 175, 600, 1920, 60, 720, 0, 86, 720, 213, 60, 720, 426, 182, 720, 640, 179, 720, 853, 152, 720, 1066, 140, 720, 1280, 115, 720, 1493, 156, 720, 1706, 172, 720, 1920, 65, 840, 0, 90, 840, 213, 69, 840, 426, 164, 840, 640, 186, 840, 853, 146, 840, 1066, 147, 840, 1280, 122, 840, 1493, 152, 840, 1706, 166, 840, 1920, 69, 960, 0, 85, 960, 213, 67, 960, 426, 186, 960, 640, 169, 960, 853, 162, 960, 1066, 147, 960, 1280, 124, 960, 1493, 149, 960, 1706, 157, 960, 1920, 69, 1080, 0, 85, 1080, 213, 70, 1080, 426, 193, 1080, 640, 192, 1080, 853, 155, 1080, 1066, 135, 1080, 1280, 120, 1080, 1493, 138, 1080, 1706, 128, 1080, 1920, 70};
+    // 100 elems
+    float weights[] = {0.04053167, -0.07571915, 0.0025228, 0.1974248, -0.14461569, 0.30541419, -0.06167972, -0.04958807, 0.00050641, -0.02023852, -0.05910183, -0.05512609, 0.37617088, -0.15904428, -0.10229038, -0.34355378, 0.39128499, 0.16502984, -0.00916588, -0.05264022, 0.01936275, 0.07156495, -0.35144818, -0.01955447, 0.07818357, 0.10850975, -0.34972095, -0.12186563, 0.04116577, 0.00378971, -0.05382689, 0.12480152, 0.0629668, 0.01070963, 0.03499341, 0.03218783, -0.00559502, -0.21214646, 0.34402987, -0.0246006, -0.01532894, 0.14944613, -0.12848858, -0.04539923, -0.1401344, -0.08143395, 0.18241941, 0.08353522, -0.20102434, 0.05168503, -0.05897654, 0.09669851, -0.13024191, -0.02125988, 0.19499928, 0.12311092, -0.05857943, -0.14954968, -0.19763496, 0.1172914, -0.04584096, 0.20130274, -0.198312, 0.04763513, -0.05524272, -0.07034198, 0.15155036, 0.0718368, -0.09612551, 0.09076872, -0.0863682, 0.08457486, 0.05520895, -0.10551672, 0.12577908, -0.05857014, 0.03529613, 0.01300365, -0.09736055, 0.06580921, -0.03074898, 0.1528181, -0.18295799, 0.18713497, -0.07368328, -0.06431868, 0.02485816, -0.02782657, -0.15182979, 0.06845911, -0.01974711, 0.23508139, -0.23124445, -0.1202987, 0.05380124, 0.04331648, 0.06285662, -0.01171756, 0.01128749, 0.06012665};
+
+    Mat dst(height, width, CV_8U);
+
+    PERF_SAMPLE_BEGIN()
+        idw_halide(NULL, dst.ptr<uint8_t>(), height, width, points, weights);
+    PERF_SAMPLE_END()
+
+    SANITY_CHECK_NOTHING();
+}
\ No newline at end of file
diff --git a/src/idw.cpp b/src/idw.cpp
index 8c738f1..7ef9887 100644
--- a/src/idw.cpp
+++ b/src/idw.cpp
@@ -3,11 +3,128 @@
 #include "algos.hpp"
 
 #include <limits>
+#include <iostream>
 
 using namespace std;
 
+#ifdef __riscv
+  #include <HalideBuffer.h>
+  #include "histogram.h"
+  using namespace Halide::Runtime;
+#else
+  #include <Halide.h>
+  using namespace Halide;
+#endif
+
 static const int pointCount = 100;
 
+void idw_halide(const uint8_t* src, uint8_t* dst, int height, int width, int* pointsBuf, float* weightsBuf) {
+    float* maskBuf = new float[height * width]();
+
+    Buffer<float> mask(maskBuf, {width, height});
+    Buffer<uint8_t> output(dst, {width, height});
+
+    Buffer<int> points(pointsBuf, {pointCount*3});
+    Buffer<float> weights(weightsBuf, {pointCount});
+#ifdef __riscv
+    idw(input, output);
+#else
+    static Func f("idw");
+    
+    // try {
+    if (!f.defined()) {
+        Var x("x"), y("y");
+        RDom r(0, pointCount);
+        
+        f(x, y) = 0.F;
+        Expr x0 = points(3*r+1);
+        Expr y0 = points(3*r);
+        Expr dx = x - x0;
+        Expr dy = y - y0;
+        Expr weight = weights(r);
+        f(x, y) += hypot(dx, dy) * weight;
+
+        // f.vectorize(r, 8);
+        const int factor = 4; 
+        f.update().atomic().vectorize(r, factor);
+
+        // Compile
+        Target target;
+        target.os = Target::OS::Linux;
+        target.arch = Target::Arch::RISCV;
+        target.bits = 64;
+        target.vector_bits = factor * sizeof(float) * 8;
+
+        // Tested XuanTie C906 has 128-bit vector unit
+        CV_Assert(target.vector_bits <= 128);
+
+        std::vector<Target::Feature> features;
+        features.push_back(Target::RVV);
+        features.push_back(Target::NoAsserts);
+        features.push_back(Target::NoRuntime);
+        target.set_features(features);
+
+        std::cout << target << std::endl;
+        f.print_loop_nest();
+
+        // Dump AOT code
+        f.compile_to_header("idw.h", {}, "idw", target);
+        f.compile_to_assembly("idw.s", {}, "idw", target);
+    }
+    // }
+    // catch (Halide::Error &e) {
+    //     cout << e.what() << '\n';
+    // }
+
+    f.realize(mask);
+#endif
+
+    //
+    float maxVal = -numeric_limits<float>::infinity();
+    float minVal = numeric_limits<float>::infinity();
+    for (int y = 0; y < height; y++) {
+        for (int x = 0; x < width; x++) {
+            if (maskBuf[y*width + x] < minVal) {
+                minVal = maskBuf[y*width + x];
+            }
+            if (maskBuf[y*width + x] > maxVal) {
+                maxVal = maskBuf[y*width + x];
+            }
+        }
+    }
+
+    float diff = maxVal - minVal;
+    for (int y = 0; y < height; y++) {
+        for (int x = 0; x < width; x++) {
+            dst[y*width + x] = (uint8_t) (255 * (maskBuf[y*width + x] - minVal) / diff);
+        }
+    }
+    //
+
+    delete[] maskBuf;
+// #ifdef __riscv
+//     idw(input, output);
+// #else
+//     static Func f("idw");
+//     if (!f.defined()) {
+//         Var x("x"), y("y");
+//         RDom r(0, pointCount);
+        
+//         f(x, y) = 0;
+//         Expr x0 = points(r.x, 0);
+//         Expr y0 = points(r.x, 1);
+//         Expr dx = x - x0;
+//         Expr dy = y - y0;
+//         f(x, y) += sqrt(dx*dx + dy*dy) * weights(r.x);
+//         f(x, y) = cast<uint8_t>(f(x, y));
+        
+//         // Dump AOT code
+//         f.compile_to_header("idw.h", {input}, "idw", target);
+//         f.compile_to_assembly("idw.s", {input}, "idw", target);
+//     }
+// #endif
+}
+
 void idw_ref(const uint8_t* src, uint8_t* dst, int height, int width, int* points, float* weights) {
     float* mask = new float[height * width]();
 
diff --git a/test/test_main.cpp b/test/test_main.cpp
index 05580f8..9fadfbd 100644
--- a/test/test_main.cpp
+++ b/test/test_main.cpp
@@ -164,8 +164,9 @@ TEST(convolution_nhwc, halide) {
 #endif  // HAVE_OPENCV_DNN
 
 
-TEST(idw_ref, halide) {
-    Mat src(height, width, CV_8U), dst(height, width, CV_8U), cl_dst(height, width, CV_8UC3);
+TEST(idw, halide) {
+    Mat src(height, width, CV_8U); 
+    Mat dst(height, width, CV_8U), cl_dst(height, width, CV_8UC3), dst_h(height, width, CV_8U), cl_dst_h(height, width, CV_8UC3);
     // randu(src, 0, 256);
 
     // 300 elems (100 points)
@@ -174,10 +175,14 @@ TEST(idw_ref, halide) {
     float weights[] = {0.04053167, -0.07571915, 0.0025228, 0.1974248, -0.14461569, 0.30541419, -0.06167972, -0.04958807, 0.00050641, -0.02023852, -0.05910183, -0.05512609, 0.37617088, -0.15904428, -0.10229038, -0.34355378, 0.39128499, 0.16502984, -0.00916588, -0.05264022, 0.01936275, 0.07156495, -0.35144818, -0.01955447, 0.07818357, 0.10850975, -0.34972095, -0.12186563, 0.04116577, 0.00378971, -0.05382689, 0.12480152, 0.0629668, 0.01070963, 0.03499341, 0.03218783, -0.00559502, -0.21214646, 0.34402987, -0.0246006, -0.01532894, 0.14944613, -0.12848858, -0.04539923, -0.1401344, -0.08143395, 0.18241941, 0.08353522, -0.20102434, 0.05168503, -0.05897654, 0.09669851, -0.13024191, -0.02125988, 0.19499928, 0.12311092, -0.05857943, -0.14954968, -0.19763496, 0.1172914, -0.04584096, 0.20130274, -0.198312, 0.04763513, -0.05524272, -0.07034198, 0.15155036, 0.0718368, -0.09612551, 0.09076872, -0.0863682, 0.08457486, 0.05520895, -0.10551672, 0.12577908, -0.05857014, 0.03529613, 0.01300365, -0.09736055, 0.06580921, -0.03074898, 0.1528181, -0.18295799, 0.18713497, -0.07368328, -0.06431868, 0.02485816, -0.02782657, -0.15182979, 0.06845911, -0.01974711, 0.23508139, -0.23124445, -0.1202987, 0.05380124, 0.04331648, 0.06285662, -0.01171756, 0.01128749, 0.06012665};
 
     idw_ref(NULL, dst.ptr<uint8_t>(), height, width, points, weights);
+    idw_halide(NULL, dst_h.ptr<uint8_t>(), height, width, points, weights);
 
     applyColorMap(dst, cl_dst, COLORMAP_JET);
+    applyColorMap(dst_h, cl_dst_h, COLORMAP_JET);
 
     // imwrite("src.png", src);
     imwrite("res.png", dst);
     imwrite("cres.png", cl_dst);
+    imwrite("res_halide.png", dst_h);
+    imwrite("cres_halide.png", cl_dst_h);
 }
\ No newline at end of file

From 0f6398316c692a485e7b55a251836fa63fef1e8e Mon Sep 17 00:00:00 2001
From: uiop07558 <uiop07558@gmail.com>
Date: Wed, 7 Feb 2024 17:33:51 +0300
Subject: [PATCH 09/15] add aot files for idw_halide

---
 aot/idw.h |  56 +++++++
 aot/idw.s | 450 ++++++++++++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 506 insertions(+)
 create mode 100644 aot/idw.h
 create mode 100644 aot/idw.s

diff --git a/aot/idw.h b/aot/idw.h
new file mode 100644
index 0000000..94728e2
--- /dev/null
+++ b/aot/idw.h
@@ -0,0 +1,56 @@
+#ifndef HALIDE__idw_h
+#define HALIDE__idw_h
+#include <stdint.h>
+
+// Forward declarations of the types used in the interface
+// to the Halide pipeline.
+//
+// For the definitions of these structs, include HalideRuntime.h
+
+// Halide's representation of a multi-dimensional array.
+// Halide::Runtime::Buffer is a more user-friendly wrapper
+// around this. Its declaration is in HalideBuffer.h
+struct halide_buffer_t;
+
+// Metadata describing the arguments to the generated function.
+// Used to construct calls to the _argv version of the function.
+struct halide_filter_metadata_t;
+
+#ifndef HALIDE_MUST_USE_RESULT
+#ifdef __has_attribute
+#if __has_attribute(nodiscard)
+#define HALIDE_MUST_USE_RESULT [[nodiscard]]
+#elif __has_attribute(warn_unused_result)
+#define HALIDE_MUST_USE_RESULT __attribute__((warn_unused_result))
+#else
+#define HALIDE_MUST_USE_RESULT
+#endif
+#else
+#define HALIDE_MUST_USE_RESULT
+#endif
+#endif
+
+#ifndef HALIDE_FUNCTION_ATTRS
+#define HALIDE_FUNCTION_ATTRS
+#endif
+
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+HALIDE_FUNCTION_ATTRS
+int idw(struct halide_buffer_t *_idw_buffer);
+
+HALIDE_FUNCTION_ATTRS
+int idw_argv(void **args);
+
+HALIDE_FUNCTION_ATTRS
+const struct halide_filter_metadata_t *idw_metadata();
+
+#ifdef __cplusplus
+}  // extern "C"
+#endif
+
+#endif
diff --git a/aot/idw.s b/aot/idw.s
new file mode 100644
index 0000000..45ac36e
--- /dev/null
+++ b/aot/idw.s
@@ -0,0 +1,450 @@
+	.text
+	.attribute	4, 16
+	.attribute	5, "rv64gcv0p7"
+	.file	"halide_buffer_t.cpp"
+	.section	.text.idw,"ax",@progbits
+	.globl	idw                             # -- Begin function idw
+	.p2align	1
+	.type	idw,@function
+idw:                                    # @idw
+# %bb.0:                                # %entry
+	addi	sp, sp, -192
+	sd	ra, 184(sp)                     # 8-byte Folded Spill
+	sd	s0, 176(sp)                     # 8-byte Folded Spill
+	sd	s1, 168(sp)                     # 8-byte Folded Spill
+	sd	s2, 160(sp)                     # 8-byte Folded Spill
+	sd	s3, 152(sp)                     # 8-byte Folded Spill
+	sd	s4, 144(sp)                     # 8-byte Folded Spill
+	sd	s5, 136(sp)                     # 8-byte Folded Spill
+	sd	s6, 128(sp)                     # 8-byte Folded Spill
+	sd	s7, 120(sp)                     # 8-byte Folded Spill
+	sd	s8, 112(sp)                     # 8-byte Folded Spill
+	sd	s9, 104(sp)                     # 8-byte Folded Spill
+	sd	s10, 96(sp)                     # 8-byte Folded Spill
+	sd	s11, 88(sp)                     # 8-byte Folded Spill
+.Lpcrel_hi0:
+	auipc	a1, %pcrel_hi(.Lb2.buffer)
+	addi	a1, a1, %pcrel_lo(.Lpcrel_hi0)
+	ld	s6, 16(a1)
+	ld	a3, 40(a0)
+.Lpcrel_hi1:
+	auipc	a2, %pcrel_hi(.Lb3.buffer)
+	addi	a2, a2, %pcrel_lo(.Lpcrel_hi1)
+	ld	s9, 16(a2)
+	lw	s4, 0(a3)
+	lwu	s8, 4(a3)
+	lw	a4, 16(a3)
+	sd	a4, 8(sp)                       # 8-byte Folded Spill
+	ld	a4, 0(a1)
+	lwu	a5, 20(a3)
+	sd	a5, 32(sp)                      # 8-byte Folded Spill
+	lw	s1, 24(a3)
+	or	a5, s6, a4
+	mv	a4, s9
+	bnez	a5, .LBB0_2
+# %bb.1:                                # %then_bb
+	lui	a4, 128
+	addiw	a4, a4, 9
+	slli	a4, a4, 13
+	sd	a4, 32(a1)
+	ld	a4, 40(a1)
+	sd	zero, 0(a1)
+	sd	zero, 8(a1)
+	sd	zero, 16(a1)
+	sw	zero, 0(a4)
+	li	a5, 300
+	sw	a5, 4(a4)
+	li	a5, 1
+	sw	a5, 8(a4)
+	sw	zero, 12(a4)
+	ld	a4, 16(a2)
+	sd	zero, 24(a1)
+.LBB0_2:                                # %after_bb
+	ld	a5, 0(a2)
+	ld	s11, 16(a0)
+	or	a4, a4, a5
+	bnez	a4, .LBB0_4
+# %bb.3:                                # %then_bb2
+	sd	zero, 16(a2)
+	sd	zero, 8(a2)
+	sd	zero, 0(a2)
+	lui	a4, 128
+	addiw	a4, a4, 9
+	ld	a5, 40(a2)
+	slli	a4, a4, 13
+	addi	a4, a4, 2
+	sd	a4, 32(a2)
+	sw	zero, 0(a5)
+	li	a4, 100
+	sw	a4, 4(a5)
+	li	a4, 1
+	sw	a4, 8(a5)
+	sw	zero, 12(a5)
+	sd	zero, 24(a2)
+.LBB0_4:                                # %after_bb1
+	sext.w	a4, s8
+	sd	a4, 24(sp)                      # 8-byte Folded Spill
+	lw	a4, 32(sp)                      # 8-byte Folded Reload
+	sd	s1, 16(sp)                      # 8-byte Folded Spill
+	beqz	s11, .LBB0_6
+# %bb.5:
+	li	a0, 0
+	j	.LBB0_9
+.LBB0_6:                                # %_halide_buffer_is_bounds_query.exit36
+	ld	a5, 0(a0)
+	bnez	a5, .LBB0_8
+# %bb.7:                                # %then_bb5
+	sd	zero, 16(a0)
+	sd	zero, 8(a0)
+	sd	zero, 0(a0)
+	lui	a5, 256
+	addiw	a5, a5, 9
+	slli	a5, a5, 13
+	addi	a5, a5, 2
+	sd	a5, 32(a0)
+	sw	s4, 0(a3)
+	ld	s1, 24(sp)                      # 8-byte Folded Reload
+	sw	s1, 4(a3)
+	li	a5, 1
+	sw	a5, 8(a3)
+	sw	zero, 12(a3)
+	ld	a5, 8(sp)                       # 8-byte Folded Reload
+	sw	a5, 16(a3)
+	sw	a4, 20(a3)
+	sw	s1, 24(a3)
+	sw	zero, 28(a3)
+	sd	zero, 24(a0)
+.LBB0_8:                                # %land.rhs.i49
+	ld	a0, 0(a0)
+	seqz	a0, a0
+.LBB0_9:                                # %_halide_buffer_is_bounds_query.exit50
+	ld	a3, 16(a1)
+	ld	a1, 0(a1)
+	ld	a5, 16(a2)
+	ld	a2, 0(a2)
+	or	a1, a1, a3
+	seqz	a1, a1
+	or	a2, a2, a5
+	seqz	a2, a2
+	or	a1, a1, a2
+	or	a0, a0, a1
+	slti	a1, a4, 1
+	or	a0, a0, a1
+	bnez	a0, .LBB0_20
+# %bb.10:                               # %"for idw.s0.y.rebased.preheader"
+	ld	s3, 16(sp)                      # 8-byte Folded Reload
+	ld	a0, 24(sp)                      # 8-byte Folded Reload
+	blez	a0, .LBB0_13
+# %bb.11:                               # %"for idw.s0.y.rebased.us.preheader"
+	li	s0, 0
+	ld	s2, 24(sp)                      # 8-byte Folded Reload
+	slli	s2, s2, 2
+	ld	s1, 32(sp)                      # 8-byte Folded Reload
+.LBB0_12:                               # %"for idw.s0.y.rebased.us"
+                                        # =>This Inner Loop Header: Depth=1
+	slli	a0, s0, 2
+	add	a0, a0, s11
+	li	a1, 0
+	mv	a2, s2
+	call	memset@plt
+	addi	s1, s1, -1
+	addw	s0, s0, s3
+	bnez	s1, .LBB0_12
+.LBB0_13:                               # %"for idw.s1.y.rebased.preheader"
+	sd	zero, 40(sp)                    # 8-byte Folded Spill
+	li	a1, 6
+	li	a2, 1
+	addi	a3, sp, 64
+	addi	a7, sp, 76
+	addi	t0, sp, 72
+	addi	t1, sp, 68
+	li	s0, 4
+	addi	s1, sp, 48
+	addi	t2, sp, 52
+	addi	t3, sp, 60
+	addi	t6, sp, 56
+	li	ra, 2
+	li	a6, 3
+.LBB0_14:                               # %"for idw.s1.y.rebased"
+                                        # =>This Loop Header: Depth=1
+                                        #     Child Loop BB0_16 Depth 2
+                                        #       Child Loop BB0_17 Depth 3
+	ld	a0, 24(sp)                      # 8-byte Folded Reload
+	blez	a0, .LBB0_19
+# %bb.15:                               # %"for idw.s1.x.rebased.preheader"
+                                        #   in Loop: Header=BB0_14 Depth=1
+	li	t4, 0
+	ld	a0, 16(sp)                      # 8-byte Folded Reload
+	ld	a4, 40(sp)                      # 8-byte Folded Reload
+	mulw	t5, a0, a4
+	ld	a0, 8(sp)                       # 8-byte Folded Reload
+	addw	a0, a0, a4
+.LBB0_16:                               # %"for idw.s1.x.rebased"
+                                        #   Parent Loop BB0_14 Depth=1
+                                        # =>  This Loop Header: Depth=2
+                                        #       Child Loop BB0_17 Depth 3
+	add	a4, t4, t5
+	slli	a4, a4, 2
+	add	s2, s11, a4
+	flw	ft0, 0(s2)
+	addw	a4, s4, t4
+	li	s5, 25
+	mv	s3, s6
+	mv	s10, s9
+.LBB0_17:                               # %"for idw.s1.r4$x.r4$x"
+                                        #   Parent Loop BB0_14 Depth=1
+                                        #     Parent Loop BB0_16 Depth=2
+                                        # =>    This Inner Loop Header: Depth=3
+	addi	s7, s3, 4
+	vsetvli	zero, a1, e32, m2
+	vlwu.v	v8, (s7)
+	vsetvli	zero, a2, e32, m2
+	vslidedown.vi	v10, v8, 3
+	addi	a5, s3, 20
+	vsetvli	zero, a1, e32, m2
+	vlwu.v	v12, (a5)
+	vsetvli	zero, a2, e32, m2
+	vslidedown.vi	v14, v12, 2
+	vslidedown.vi	v12, v12, 5
+	vsw.v	v8, (a3)
+	vsw.v	v12, (a7)
+	vsw.v	v14, (t0)
+	vsw.v	v10, (t1)
+	vsetvli	zero, s0, e32, m1
+	vlwu.v	v8, (a3)
+	vsetvli	a5, zero, e32, m1
+	vrsub.vx	v8, v8, a4
+	vsetvli	zero, a1, e32, m2
+	vlwu.v	v10, (s3)
+	vsetvli	zero, a2, e32, m2
+	vslidedown.vi	v12, v10, 3
+	addi	a5, s3, 16
+	vsetvli	zero, a1, e32, m2
+	vlwu.v	v14, (a5)
+	vsetvli	zero, a2, e32, m2
+	vslidedown.vi	v16, v14, 2
+	vslidedown.vi	v14, v14, 5
+	vsw.v	v10, (s1)
+	vsw.v	v12, (t2)
+	vsw.v	v14, (t3)
+	vsw.v	v16, (t6)
+	vsetvli	zero, s0, e32, m1
+	vlwu.v	v9, (s1)
+	vsetvli	a5, zero, e32, m1
+	vrsub.vx	v9, v9, a0
+	vmul.vv	v9, v9, v9
+	vmacc.vv	v9, v8, v8
+	vfcvt.f.x.v	v8, v9
+	vfmv.f.s	ft1, v8
+	fsqrt.s	ft1, ft1
+	vfmv.s.f	v9, ft1
+	vsetvli	zero, a2, e32, m1
+	vslidedown.vi	v10, v8, 1
+	vfmv.f.s	ft1, v10
+	fsqrt.s	ft1, ft1
+	vsetvli	a5, zero, e32, m1
+	vfmv.s.f	v10, ft1
+	vsetvli	zero, ra, e32, m1
+	vslideup.vi	v9, v10, 1
+	vsetvli	zero, a2, e32, m1
+	vslidedown.vi	v10, v8, 2
+	vfmv.f.s	ft1, v10
+	fsqrt.s	ft1, ft1
+	vsetvli	a5, zero, e32, m1
+	vfmv.s.f	v10, ft1
+	vsetvli	zero, a6, e32, m1
+	vslideup.vi	v9, v10, 2
+	vsetvli	zero, a2, e32, m1
+	vslidedown.vi	v8, v8, 3
+	vfmv.f.s	ft1, v8
+	fsqrt.s	ft1, ft1
+	vsetvli	a5, zero, e32, m1
+	vfmv.s.f	v8, ft1
+	vsetvli	zero, s0, e32, m1
+	vslideup.vi	v9, v8, 3
+	vsetvli	zero, zero, e32, m1
+	vlwu.v	v8, (s10)
+	vfmul.vv	v8, v9, v8
+	vfmv.s.f	v9, ft0
+	vfredusum.vs	v8, v8, v9
+	vfmv.f.s	ft0, v8
+	addi	s5, s5, -1
+	addi	s10, s10, 16
+	addi	s3, s3, 48
+	bnez	s5, .LBB0_17
+# %bb.18:                               # %"end for idw.s1.r4$x.r4$x"
+                                        #   in Loop: Header=BB0_16 Depth=2
+	addi	t4, t4, 1
+	fsw	ft0, 0(s2)
+	bne	t4, s8, .LBB0_16
+.LBB0_19:                               # %"end for idw.s1.x.rebased"
+                                        #   in Loop: Header=BB0_14 Depth=1
+	ld	a4, 40(sp)                      # 8-byte Folded Reload
+	addi	a4, a4, 1
+	ld	a0, 32(sp)                      # 8-byte Folded Reload
+	sd	a4, 40(sp)                      # 8-byte Folded Spill
+	bne	a4, a0, .LBB0_14
+.LBB0_20:                               # %destructor_block
+	li	a0, 0
+	ld	ra, 184(sp)                     # 8-byte Folded Reload
+	ld	s0, 176(sp)                     # 8-byte Folded Reload
+	ld	s1, 168(sp)                     # 8-byte Folded Reload
+	ld	s2, 160(sp)                     # 8-byte Folded Reload
+	ld	s3, 152(sp)                     # 8-byte Folded Reload
+	ld	s4, 144(sp)                     # 8-byte Folded Reload
+	ld	s5, 136(sp)                     # 8-byte Folded Reload
+	ld	s6, 128(sp)                     # 8-byte Folded Reload
+	ld	s7, 120(sp)                     # 8-byte Folded Reload
+	ld	s8, 112(sp)                     # 8-byte Folded Reload
+	ld	s9, 104(sp)                     # 8-byte Folded Reload
+	ld	s10, 96(sp)                     # 8-byte Folded Reload
+	ld	s11, 88(sp)                     # 8-byte Folded Reload
+	addi	sp, sp, 192
+	ret
+.Lfunc_end0:
+	.size	idw, .Lfunc_end0-idw
+                                        # -- End function
+	.section	.text.idw_argv,"ax",@progbits
+	.globl	idw_argv                        # -- Begin function idw_argv
+	.p2align	1
+	.type	idw_argv,@function
+idw_argv:                               # @idw_argv
+# %bb.0:                                # %entry
+	addi	sp, sp, -16
+	sd	ra, 8(sp)                       # 8-byte Folded Spill
+	ld	a0, 0(a0)
+	call	idw@plt
+	li	a0, 0
+	ld	ra, 8(sp)                       # 8-byte Folded Reload
+	addi	sp, sp, 16
+	ret
+.Lfunc_end1:
+	.size	idw_argv, .Lfunc_end1-idw_argv
+                                        # -- End function
+	.section	.text.idw_metadata,"ax",@progbits
+	.globl	idw_metadata                    # -- Begin function idw_metadata
+	.p2align	1
+	.type	idw_metadata,@function
+idw_metadata:                           # @idw_metadata
+# %bb.0:                                # %entry
+.Lpcrel_hi2:
+	auipc	a0, %pcrel_hi(.Lidw_metadata_storage)
+	addi	a0, a0, %pcrel_lo(.Lpcrel_hi2)
+	ret
+.Lfunc_end2:
+	.size	idw_metadata, .Lfunc_end2-idw_metadata
+                                        # -- End function
+	.type	.Lb2.shape,@object              # @b2.shape
+	.section	.rodata,"a",@progbits
+	.p2align	5, 0x0
+.Lb2.shape:
+	.asciz	"\000\000\000\000,\001\000\000\001\000\000\000\000\000\000"
+	.size	.Lb2.shape, 16
+
+	.type	.Lb2.data,@object               # @b2.data
+	.p2align	5, 0x0
+.Lb2.data:
+	.asciz	"\000\000\000\000\000\000\000\000H\000\000\000\000\000\000\000\325\000\000\000O\000\000\000\000\000\000\000\252\001\000\000<\000\000\000\000\000\000\000\200\002\000\000L\000\000\000\000\000\000\000U\003\000\000\200\000\000\000\000\000\000\000*\004\000\000C\000\000\000\000\000\000\000\000\005\000\000A\000\000\000\000\000\000\000\325\005\000\000@\000\000\000\000\000\000\000\252\006\000\000<\000\000\000\000\000\000\000\200\007\000\000=\000\000\000x\000\000\000\000\000\000\000Q\000\000\000x\000\000\000\325\000\000\000O\000\000\000x\000\000\000\252\001\000\000:\000\000\000x\000\000\000\200\002\000\000\204\000\000\000x\000\000\000U\003\000\000\225\000\000\000x\000\000\000*\004\000\000\216\000\000\000x\000\000\000\000\005\000\000@\000\000\000x\000\000\000\325\005\000\000E\000\000\000x\000\000\000\252\006\000\000A\000\000\000x\000\000\000\200\007\000\000@\000\000\000\360\000\000\000\000\000\000\000K\000\000\000\360\000\000\000\325\000\000\000D\000\000\000\360\000\000\000\252\001\000\000\214\000\000\000\360\000\000\000\200\002\000\000\231\000\000\000\360\000\000\000U\003\000\000\221\000\000\000\360\000\000\000*\004\000\000\204\000\000\000\360\000\000\000\000\005\000\000\230\000\000\000\360\000\000\000\325\005\000\000}\000\000\000\360\000\000\000\252\006\000\000B\000\000\000\360\000\000\000\200\007\000\000:\000\000\000h\001\000\000\000\000\000\000N\000\000\000h\001\000\000\325\000\000\000<\000\000\000h\001\000\000\252\001\000\000\213\000\000\000h\001\000\000\200\002\000\000\250\000\000\000h\001\000\000U\003\000\000\232\000\000\000h\001\000\000*\004\000\000\212\000\000\000h\001\000\000\000\005\000\000\221\000\000\000h\001\000\000\325\005\000\000\240\000\000\000h\001\000\000\252\006\000\000D\000\000\000h\001\000\000\200\007\000\000<\000\000\000\340\001\000\000\000\000\000\000M\000\000\000\340\001\000\000\325\000\000\000;\000\000\000\340\001\000\000\252\001\000\000\245\000\000\000\340\001\000\000\200\002\000\000\267\000\000\000\340\001\000\000U\003\000\000\246\000\000\000\340\001\000\000*\004\000\000\216\000\000\000\340\001\000\000\000\005\000\000{\000\000\000\340\001\000\000\325\005\000\000\233\000\000\000\340\001\000\000\252\006\000\000\220\000\000\000\340\001\000\000\200\007\000\000<\000\000\000X\002\000\000\000\000\000\000S\000\000\000X\002\000\000\325\000\000\000A\000\000\000X\002\000\000\252\001\000\000\262\000\000\000X\002\000\000\200\002\000\000\270\000\000\000X\002\000\000U\003\000\000\212\000\000\000X\002\000\000*\004\000\000|\000\000\000X\002\000\000\000\005\000\000\204\000\000\000X\002\000\000\325\005\000\000\257\000\000\000X\002\000\000\252\006\000\000\257\000\000\000X\002\000\000\200\007\000\000<\000\000\000\320\002\000\000\000\000\000\000V\000\000\000\320\002\000\000\325\000\000\000<\000\000\000\320\002\000\000\252\001\000\000\266\000\000\000\320\002\000\000\200\002\000\000\263\000\000\000\320\002\000\000U\003\000\000\230\000\000\000\320\002\000\000*\004\000\000\214\000\000\000\320\002\000\000\000\005\000\000s\000\000\000\320\002\000\000\325\005\000\000\234\000\000\000\320\002\000\000\252\006\000\000\254\000\000\000\320\002\000\000\200\007\000\000A\000\000\000H\003\000\000\000\000\000\000Z\000\000\000H\003\000\000\325\000\000\000E\000\000\000H\003\000\000\252\001\000\000\244\000\000\000H\003\000\000\200\002\000\000\272\000\000\000H\003\000\000U\003\000\000\222\000\000\000H\003\000\000*\004\000\000\223\000\000\000H\003\000\000\000\005\000\000z\000\000\000H\003\000\000\325\005\000\000\230\000\000\000H\003\000\000\252\006\000\000\246\000\000\000H\003\000\000\200\007\000\000E\000\000\000\300\003\000\000\000\000\000\000U\000\000\000\300\003\000\000\325\000\000\000C\000\000\000\300\003\000\000\252\001\000\000\272\000\000\000\300\003\000\000\200\002\000\000\251\000\000\000\300\003\000\000U\003\000\000\242\000\000\000\300\003\000\000*\004\000\000\223\000\000\000\300\003\000\000\000\005\000\000|\000\000\000\300\003\000\000\325\005\000\000\225\000\000\000\300\003\000\000\252\006\000\000\235\000\000\000\300\003\000\000\200\007\000\000E\000\000\0008\004\000\000\000\000\000\000U\000\000\0008\004\000\000\325\000\000\000F\000\000\0008\004\000\000\252\001\000\000\301\000\000\0008\004\000\000\200\002\000\000\300\000\000\0008\004\000\000U\003\000\000\233\000\000\0008\004\000\000*\004\000\000\207\000\000\0008\004\000\000\000\005\000\000x\000\000\0008\004\000\000\325\005\000\000\212\000\000\0008\004\000\000\252\006\000\000\200\000\000\0008\004\000\000\200\007\000\000F\000\000"
+	.size	.Lb2.data, 1200
+
+	.type	.Lb2.buffer,@object             # @b2.buffer
+	.data
+	.p2align	4, 0x0
+.Lb2.buffer:
+	.quad	0                               # 0x0
+	.quad	0
+	.quad	.Lb2.data
+	.quad	1                               # 0x1
+	.byte	0                               # 0x0
+	.byte	32                              # 0x20
+	.half	1                               # 0x1
+	.word	1                               # 0x1
+	.quad	.Lb2.shape
+	.quad	0
+	.size	.Lb2.buffer, 56
+
+	.type	.Lb3.shape,@object              # @b3.shape
+	.section	.rodata,"a",@progbits
+	.p2align	5, 0x0
+.Lb3.shape:
+	.asciz	"\000\000\000\000d\000\000\000\001\000\000\000\000\000\000"
+	.size	.Lb3.shape, 16
+
+	.type	.Lb3.data,@object               # @b3.data
+	.p2align	5, 0x0
+.Lb3.data:
+	.ascii	"\211\004&=\244\022\233\275\217U%;\272)J>#\026\024\276@_\234>\340\243|\275\334\034K\275\232\300\004:A\313\245\274\303\024r\275\345\313a\275x\231\300>\201\334\"\276\236}\321\275H\346\257\276\202V\310>\225\375(>},\026\274E\235W\275\241\236\236<\245\220\222=\004\361\263\276\2620\240\274\265\036\240=\\:\336=\240\016\263\276\260\224\371\275p\235(=\311\\x;\226y\\\275\360\227\377=\275\364\200=rw/<@U\017=c\327\003=nV\267\273\354<Y\276\257$\260>3\207\311\274<&{\274h\b\031>\203\222\003\276\213\3649\275d\177\017\276\330\306\246\275'\314:>\203\024\253=S\331M\276\257\263S=b\221q\275\336\t\306=#^\005\2763)\256\274\344\255G>\224!\374=\374\360o\275\215#\031\276\322`J\276y6\360=\273\303;\275N\"N>M\022K\276\016\035C=1Fb\275u\017\220\275\0040\033>,\037\223=t\335\304\275\363\344\271=\320\341\260\275\2265\255=\310\"b=&\031\330\275;\314\000>>\347o\275\255\222\020=C\rU<\370d\307\275\373\306\206=I\345\373\274Y|\034>WY;\276O\240?>B\347\226\275\203\271\203\275W\243\313<\214\364\343\274Ey\033\276J4\214=\261\304\241\274-\271p>X\313l\276*_\366\275\260^\\=\237l1=\371\272\200=\002\373?\274*\3578<]Gv="
+	.size	.Lb3.data, 400
+
+	.type	.Lb3.buffer,@object             # @b3.buffer
+	.data
+	.p2align	4, 0x0
+.Lb3.buffer:
+	.quad	0                               # 0x0
+	.quad	0
+	.quad	.Lb3.data
+	.quad	1                               # 0x1
+	.byte	2                               # 0x2
+	.byte	32                              # 0x20
+	.half	1                               # 0x1
+	.word	1                               # 0x1
+	.quad	.Lb3.shape
+	.quad	0
+	.size	.Lb3.buffer, 56
+
+	.type	.L__unnamed_1,@object           # @0
+	.section	.rodata,"a",@progbits
+	.p2align	4, 0x0
+.L__unnamed_1:
+	.zero	32
+	.size	.L__unnamed_1, 32
+
+	.type	.Lstr,@object                   # @str
+	.p2align	5, 0x0
+.Lstr:
+	.asciz	"idw"
+	.size	.Lstr, 4
+
+	.type	.L__unnamed_2,@object           # @1
+	.section	.data.rel.ro,"aw",@progbits
+	.p2align	4, 0x0
+.L__unnamed_2:
+	.quad	.Lstr
+	.word	2                               # 0x2
+	.word	2                               # 0x2
+	.byte	2                               # 0x2
+	.byte	32                              # 0x20
+	.half	1                               # 0x1
+	.zero	4
+	.quad	0
+	.quad	0
+	.quad	0
+	.quad	0
+	.quad	.L__unnamed_1
+	.size	.L__unnamed_2, 64
+
+	.type	.Lstr.4,@object                 # @str.4
+	.section	.rodata,"a",@progbits
+	.p2align	5, 0x0
+.Lstr.4:
+	.asciz	"riscv-64-linux-no_asserts-no_runtime-rvv-vector_bits_128"
+	.size	.Lstr.4, 57
+
+	.type	.Lidw_metadata_storage,@object  # @idw_metadata_storage
+	.section	.data.rel.ro,"aw",@progbits
+	.p2align	4, 0x0
+.Lidw_metadata_storage:
+	.word	1                               # 0x1
+	.word	1                               # 0x1
+	.quad	.L__unnamed_2
+	.quad	.Lstr.4
+	.quad	.Lstr
+	.size	.Lidw_metadata_storage, 32
+
+	.ident	"clang version 16.0.0 (https://github.com/dkurt/llvm-rvv-071 b027aa1b59c9f53240bdc836f39656723fdf9df0)"
+	.ident	"clang version 16.0.0 (https://github.com/dkurt/llvm-rvv-071 b027aa1b59c9f53240bdc836f39656723fdf9df0)"
+	.ident	"clang version 16.0.0 (https://github.com/dkurt/llvm-rvv-071 b027aa1b59c9f53240bdc836f39656723fdf9df0)"
+	.ident	"clang version 16.0.0 (https://github.com/dkurt/llvm-rvv-071 b027aa1b59c9f53240bdc836f39656723fdf9df0)"
+	.ident	"clang version 16.0.0 (https://github.com/dkurt/llvm-rvv-071 b027aa1b59c9f53240bdc836f39656723fdf9df0)"
+	.ident	"clang version 16.0.0 (https://github.com/dkurt/llvm-rvv-071 b027aa1b59c9f53240bdc836f39656723fdf9df0)"
+	.section	".note.GNU-stack","",@progbits

From 366b68a9040cf411535406e8b5b93142cf29ef31 Mon Sep 17 00:00:00 2001
From: uiop07558 <uiop07558@gmail.com>
Date: Wed, 7 Feb 2024 17:54:23 +0300
Subject: [PATCH 10/15] hardcoded maxval and minval

---
 .vscode/settings.json |  3 ++-
 src/idw.cpp           | 40 ++++++++++++++++++++--------------------
 2 files changed, 22 insertions(+), 21 deletions(-)

diff --git a/.vscode/settings.json b/.vscode/settings.json
index 8104e93..9b18adb 100644
--- a/.vscode/settings.json
+++ b/.vscode/settings.json
@@ -4,6 +4,7 @@
         "limits": "cpp",
         "iostream": "cpp",
         "cmath": "cpp",
-        "optional": "cpp"
+        "optional": "cpp",
+        "new": "cpp"
     }
 }
\ No newline at end of file
diff --git a/src/idw.cpp b/src/idw.cpp
index 7ef9887..7a08ac8 100644
--- a/src/idw.cpp
+++ b/src/idw.cpp
@@ -80,18 +80,18 @@ void idw_halide(const uint8_t* src, uint8_t* dst, int height, int width, int* po
 #endif
 
     //
-    float maxVal = -numeric_limits<float>::infinity();
-    float minVal = numeric_limits<float>::infinity();
-    for (int y = 0; y < height; y++) {
-        for (int x = 0; x < width; x++) {
-            if (maskBuf[y*width + x] < minVal) {
-                minVal = maskBuf[y*width + x];
-            }
-            if (maskBuf[y*width + x] > maxVal) {
-                maxVal = maskBuf[y*width + x];
-            }
-        }
-    }
+    float maxVal = 193.0 //-numeric_limits<float>::infinity();
+    float minVal = 58.0 //numeric_limits<float>::infinity();
+    // for (int y = 0; y < height; y++) {
+    //     for (int x = 0; x < width; x++) {
+    //         if (maskBuf[y*width + x] < minVal) {
+    //             minVal = maskBuf[y*width + x];
+    //         }
+    //         if (maskBuf[y*width + x] > maxVal) {
+    //             maxVal = maskBuf[y*width + x];
+    //         }
+    //     }
+    // }
 
     float diff = maxVal - minVal;
     for (int y = 0; y < height; y++) {
@@ -128,8 +128,8 @@ void idw_halide(const uint8_t* src, uint8_t* dst, int height, int width, int* po
 void idw_ref(const uint8_t* src, uint8_t* dst, int height, int width, int* points, float* weights) {
     float* mask = new float[height * width]();
 
-    float maxVal = -numeric_limits<float>::infinity();
-    float minVal = numeric_limits<float>::infinity();
+    float maxVal = 193.0 //-numeric_limits<float>::infinity();
+    float minVal = 58.0 //numeric_limits<float>::infinity();
 
     for (int y = 0; y < height; y++) {
         for (int x = 0; x < width; x++) {
@@ -144,12 +144,12 @@ void idw_ref(const uint8_t* src, uint8_t* dst, int height, int width, int* point
             }
 
             mask[y*width + x] = dot;
-            if (dot < minVal) {
-                minVal = dot;
-            }
-            if (dot > maxVal) {
-                maxVal = dot;
-            }
+            // if (dot < minVal) {
+            //     minVal = dot;
+            // }
+            // if (dot > maxVal) {
+            //     maxVal = dot;
+            // }
         }
     }
 

From fac19e5f0b5c9fdf8a82624c3d191ec001119e2e Mon Sep 17 00:00:00 2001
From: Dmitry Kurtaev <dmitry.kurtaev@gmail.com>
Date: Wed, 7 Feb 2024 20:42:57 +0300
Subject: [PATCH 11/15] Remove commented code

---
 .vscode/c_cpp_properties.json | 16 ----------
 .vscode/settings.json         | 10 -------
 perf/perf_main.cpp            | 27 +++++------------
 src/idw.cpp                   | 55 +++++------------------------------
 4 files changed, 15 insertions(+), 93 deletions(-)
 delete mode 100644 .vscode/c_cpp_properties.json
 delete mode 100644 .vscode/settings.json

diff --git a/.vscode/c_cpp_properties.json b/.vscode/c_cpp_properties.json
deleted file mode 100644
index 4039bef..0000000
--- a/.vscode/c_cpp_properties.json
+++ /dev/null
@@ -1,16 +0,0 @@
-{
-    "configurations": [
-        {
-            "name": "Linux",
-            "includePath": [
-                "${workspaceFolder}/**"
-            ],
-            "defines": [],
-            "compilerPath": "/usr/bin/gcc",
-            "cStandard": "c17",
-            "cppStandard": "gnu++17",
-            "intelliSenseMode": "linux-gcc-x64"
-        }
-    ],
-    "version": 4
-}
\ No newline at end of file
diff --git a/.vscode/settings.json b/.vscode/settings.json
deleted file mode 100644
index 9b18adb..0000000
--- a/.vscode/settings.json
+++ /dev/null
@@ -1,10 +0,0 @@
-{
-    "C_Cpp.errorSquiggles": "disabled",
-    "files.associations": {
-        "limits": "cpp",
-        "iostream": "cpp",
-        "cmath": "cpp",
-        "optional": "cpp",
-        "new": "cpp"
-    }
-}
\ No newline at end of file
diff --git a/perf/perf_main.cpp b/perf/perf_main.cpp
index 9952623..8357ead 100644
--- a/perf/perf_main.cpp
+++ b/perf/perf_main.cpp
@@ -14,14 +14,8 @@
 
 #include "algos.hpp"
 
-#include <Halide.h>
-
-// #include <iostream>
-
 using namespace cv;
 
-// using namespace std;
-
 CV_PERF_TEST_MAIN("")
 
 Mat src(1080, 1920, CV_8UC3);
@@ -212,19 +206,19 @@ PERF_TEST(convolution_nhwc, halide) {
 #endif  // HAVE_OPENCV_DNN
 
 
+// 300 elems (100 points)
+static int idwPoints[] = {0, 0, 72, 0, 213, 79, 0, 426, 60, 0, 640, 76, 0, 853, 128, 0, 1066, 67, 0, 1280, 65, 0, 1493, 64, 0, 1706, 60, 0, 1920, 61, 120, 0, 81, 120, 213, 79, 120, 426, 58, 120, 640, 132, 120, 853, 149, 120, 1066, 142, 120, 1280, 64, 120, 1493, 69, 120, 1706, 65, 120, 1920, 64, 240, 0, 75, 240, 213, 68, 240, 426, 140, 240, 640, 153, 240, 853, 145, 240, 1066, 132, 240, 1280, 152, 240, 1493, 125, 240, 1706, 66, 240, 1920, 58, 360, 0, 78, 360, 213, 60, 360, 426, 139, 360, 640, 168, 360, 853, 154, 360, 1066, 138, 360, 1280, 145, 360, 1493, 160, 360, 1706, 68, 360, 1920, 60, 480, 0, 77, 480, 213, 59, 480, 426, 165, 480, 640, 183, 480, 853, 166, 480, 1066, 142, 480, 1280, 123, 480, 1493, 155, 480, 1706, 144, 480, 1920, 60, 600, 0, 83, 600, 213, 65, 600, 426, 178, 600, 640, 184, 600, 853, 138, 600, 1066, 124, 600, 1280, 132, 600, 1493, 175, 600, 1706, 175, 600, 1920, 60, 720, 0, 86, 720, 213, 60, 720, 426, 182, 720, 640, 179, 720, 853, 152, 720, 1066, 140, 720, 1280, 115, 720, 1493, 156, 720, 1706, 172, 720, 1920, 65, 840, 0, 90, 840, 213, 69, 840, 426, 164, 840, 640, 186, 840, 853, 146, 840, 1066, 147, 840, 1280, 122, 840, 1493, 152, 840, 1706, 166, 840, 1920, 69, 960, 0, 85, 960, 213, 67, 960, 426, 186, 960, 640, 169, 960, 853, 162, 960, 1066, 147, 960, 1280, 124, 960, 1493, 149, 960, 1706, 157, 960, 1920, 69, 1080, 0, 85, 1080, 213, 70, 1080, 426, 193, 1080, 640, 192, 1080, 853, 155, 1080, 1066, 135, 1080, 1280, 120, 1080, 1493, 138, 1080, 1706, 128, 1080, 1920, 70};
+// 100 elems
+float idwWeights[] = {0.04053167, -0.07571915, 0.0025228, 0.1974248, -0.14461569, 0.30541419, -0.06167972, -0.04958807, 0.00050641, -0.02023852, -0.05910183, -0.05512609, 0.37617088, -0.15904428, -0.10229038, -0.34355378, 0.39128499, 0.16502984, -0.00916588, -0.05264022, 0.01936275, 0.07156495, -0.35144818, -0.01955447, 0.07818357, 0.10850975, -0.34972095, -0.12186563, 0.04116577, 0.00378971, -0.05382689, 0.12480152, 0.0629668, 0.01070963, 0.03499341, 0.03218783, -0.00559502, -0.21214646, 0.34402987, -0.0246006, -0.01532894, 0.14944613, -0.12848858, -0.04539923, -0.1401344, -0.08143395, 0.18241941, 0.08353522, -0.20102434, 0.05168503, -0.05897654, 0.09669851, -0.13024191, -0.02125988, 0.19499928, 0.12311092, -0.05857943, -0.14954968, -0.19763496, 0.1172914, -0.04584096, 0.20130274, -0.198312, 0.04763513, -0.05524272, -0.07034198, 0.15155036, 0.0718368, -0.09612551, 0.09076872, -0.0863682, 0.08457486, 0.05520895, -0.10551672, 0.12577908, -0.05857014, 0.03529613, 0.01300365, -0.09736055, 0.06580921, -0.03074898, 0.1528181, -0.18295799, 0.18713497, -0.07368328, -0.06431868, 0.02485816, -0.02782657, -0.15182979, 0.06845911, -0.01974711, 0.23508139, -0.23124445, -0.1202987, 0.05380124, 0.04331648, 0.06285662, -0.01171756, 0.01128749, 0.06012665};
+
 PERF_TEST(idw, reference) {
     const int width = 1920;
     const int height = 1080;
 
-    // 300 elems (100 points)
-    int points[] = {0, 0, 72, 0, 213, 79, 0, 426, 60, 0, 640, 76, 0, 853, 128, 0, 1066, 67, 0, 1280, 65, 0, 1493, 64, 0, 1706, 60, 0, 1920, 61, 120, 0, 81, 120, 213, 79, 120, 426, 58, 120, 640, 132, 120, 853, 149, 120, 1066, 142, 120, 1280, 64, 120, 1493, 69, 120, 1706, 65, 120, 1920, 64, 240, 0, 75, 240, 213, 68, 240, 426, 140, 240, 640, 153, 240, 853, 145, 240, 1066, 132, 240, 1280, 152, 240, 1493, 125, 240, 1706, 66, 240, 1920, 58, 360, 0, 78, 360, 213, 60, 360, 426, 139, 360, 640, 168, 360, 853, 154, 360, 1066, 138, 360, 1280, 145, 360, 1493, 160, 360, 1706, 68, 360, 1920, 60, 480, 0, 77, 480, 213, 59, 480, 426, 165, 480, 640, 183, 480, 853, 166, 480, 1066, 142, 480, 1280, 123, 480, 1493, 155, 480, 1706, 144, 480, 1920, 60, 600, 0, 83, 600, 213, 65, 600, 426, 178, 600, 640, 184, 600, 853, 138, 600, 1066, 124, 600, 1280, 132, 600, 1493, 175, 600, 1706, 175, 600, 1920, 60, 720, 0, 86, 720, 213, 60, 720, 426, 182, 720, 640, 179, 720, 853, 152, 720, 1066, 140, 720, 1280, 115, 720, 1493, 156, 720, 1706, 172, 720, 1920, 65, 840, 0, 90, 840, 213, 69, 840, 426, 164, 840, 640, 186, 840, 853, 146, 840, 1066, 147, 840, 1280, 122, 840, 1493, 152, 840, 1706, 166, 840, 1920, 69, 960, 0, 85, 960, 213, 67, 960, 426, 186, 960, 640, 169, 960, 853, 162, 960, 1066, 147, 960, 1280, 124, 960, 1493, 149, 960, 1706, 157, 960, 1920, 69, 1080, 0, 85, 1080, 213, 70, 1080, 426, 193, 1080, 640, 192, 1080, 853, 155, 1080, 1066, 135, 1080, 1280, 120, 1080, 1493, 138, 1080, 1706, 128, 1080, 1920, 70};
-    // 100 elems
-    float weights[] = {0.04053167, -0.07571915, 0.0025228, 0.1974248, -0.14461569, 0.30541419, -0.06167972, -0.04958807, 0.00050641, -0.02023852, -0.05910183, -0.05512609, 0.37617088, -0.15904428, -0.10229038, -0.34355378, 0.39128499, 0.16502984, -0.00916588, -0.05264022, 0.01936275, 0.07156495, -0.35144818, -0.01955447, 0.07818357, 0.10850975, -0.34972095, -0.12186563, 0.04116577, 0.00378971, -0.05382689, 0.12480152, 0.0629668, 0.01070963, 0.03499341, 0.03218783, -0.00559502, -0.21214646, 0.34402987, -0.0246006, -0.01532894, 0.14944613, -0.12848858, -0.04539923, -0.1401344, -0.08143395, 0.18241941, 0.08353522, -0.20102434, 0.05168503, -0.05897654, 0.09669851, -0.13024191, -0.02125988, 0.19499928, 0.12311092, -0.05857943, -0.14954968, -0.19763496, 0.1172914, -0.04584096, 0.20130274, -0.198312, 0.04763513, -0.05524272, -0.07034198, 0.15155036, 0.0718368, -0.09612551, 0.09076872, -0.0863682, 0.08457486, 0.05520895, -0.10551672, 0.12577908, -0.05857014, 0.03529613, 0.01300365, -0.09736055, 0.06580921, -0.03074898, 0.1528181, -0.18295799, 0.18713497, -0.07368328, -0.06431868, 0.02485816, -0.02782657, -0.15182979, 0.06845911, -0.01974711, 0.23508139, -0.23124445, -0.1202987, 0.05380124, 0.04331648, 0.06285662, -0.01171756, 0.01128749, 0.06012665};
-
     Mat dst(height, width, CV_8U);
 
     PERF_SAMPLE_BEGIN()
-        idw_ref(NULL, dst.ptr<uint8_t>(), height, width, points, weights);
+        idw_ref(NULL, dst.ptr<uint8_t>(), height, width, idwPoints, idwWeights);
     PERF_SAMPLE_END()
 
     SANITY_CHECK_NOTHING();
@@ -234,16 +228,11 @@ PERF_TEST(idw, halide) {
     const int width = 1920;
     const int height = 1080;
 
-    // 300 elems (100 points)
-    int points[] = {0, 0, 72, 0, 213, 79, 0, 426, 60, 0, 640, 76, 0, 853, 128, 0, 1066, 67, 0, 1280, 65, 0, 1493, 64, 0, 1706, 60, 0, 1920, 61, 120, 0, 81, 120, 213, 79, 120, 426, 58, 120, 640, 132, 120, 853, 149, 120, 1066, 142, 120, 1280, 64, 120, 1493, 69, 120, 1706, 65, 120, 1920, 64, 240, 0, 75, 240, 213, 68, 240, 426, 140, 240, 640, 153, 240, 853, 145, 240, 1066, 132, 240, 1280, 152, 240, 1493, 125, 240, 1706, 66, 240, 1920, 58, 360, 0, 78, 360, 213, 60, 360, 426, 139, 360, 640, 168, 360, 853, 154, 360, 1066, 138, 360, 1280, 145, 360, 1493, 160, 360, 1706, 68, 360, 1920, 60, 480, 0, 77, 480, 213, 59, 480, 426, 165, 480, 640, 183, 480, 853, 166, 480, 1066, 142, 480, 1280, 123, 480, 1493, 155, 480, 1706, 144, 480, 1920, 60, 600, 0, 83, 600, 213, 65, 600, 426, 178, 600, 640, 184, 600, 853, 138, 600, 1066, 124, 600, 1280, 132, 600, 1493, 175, 600, 1706, 175, 600, 1920, 60, 720, 0, 86, 720, 213, 60, 720, 426, 182, 720, 640, 179, 720, 853, 152, 720, 1066, 140, 720, 1280, 115, 720, 1493, 156, 720, 1706, 172, 720, 1920, 65, 840, 0, 90, 840, 213, 69, 840, 426, 164, 840, 640, 186, 840, 853, 146, 840, 1066, 147, 840, 1280, 122, 840, 1493, 152, 840, 1706, 166, 840, 1920, 69, 960, 0, 85, 960, 213, 67, 960, 426, 186, 960, 640, 169, 960, 853, 162, 960, 1066, 147, 960, 1280, 124, 960, 1493, 149, 960, 1706, 157, 960, 1920, 69, 1080, 0, 85, 1080, 213, 70, 1080, 426, 193, 1080, 640, 192, 1080, 853, 155, 1080, 1066, 135, 1080, 1280, 120, 1080, 1493, 138, 1080, 1706, 128, 1080, 1920, 70};
-    // 100 elems
-    float weights[] = {0.04053167, -0.07571915, 0.0025228, 0.1974248, -0.14461569, 0.30541419, -0.06167972, -0.04958807, 0.00050641, -0.02023852, -0.05910183, -0.05512609, 0.37617088, -0.15904428, -0.10229038, -0.34355378, 0.39128499, 0.16502984, -0.00916588, -0.05264022, 0.01936275, 0.07156495, -0.35144818, -0.01955447, 0.07818357, 0.10850975, -0.34972095, -0.12186563, 0.04116577, 0.00378971, -0.05382689, 0.12480152, 0.0629668, 0.01070963, 0.03499341, 0.03218783, -0.00559502, -0.21214646, 0.34402987, -0.0246006, -0.01532894, 0.14944613, -0.12848858, -0.04539923, -0.1401344, -0.08143395, 0.18241941, 0.08353522, -0.20102434, 0.05168503, -0.05897654, 0.09669851, -0.13024191, -0.02125988, 0.19499928, 0.12311092, -0.05857943, -0.14954968, -0.19763496, 0.1172914, -0.04584096, 0.20130274, -0.198312, 0.04763513, -0.05524272, -0.07034198, 0.15155036, 0.0718368, -0.09612551, 0.09076872, -0.0863682, 0.08457486, 0.05520895, -0.10551672, 0.12577908, -0.05857014, 0.03529613, 0.01300365, -0.09736055, 0.06580921, -0.03074898, 0.1528181, -0.18295799, 0.18713497, -0.07368328, -0.06431868, 0.02485816, -0.02782657, -0.15182979, 0.06845911, -0.01974711, 0.23508139, -0.23124445, -0.1202987, 0.05380124, 0.04331648, 0.06285662, -0.01171756, 0.01128749, 0.06012665};
-
     Mat dst(height, width, CV_8U);
 
     PERF_SAMPLE_BEGIN()
-        idw_halide(NULL, dst.ptr<uint8_t>(), height, width, points, weights);
+        idw_halide(NULL, dst.ptr<uint8_t>(), height, width, idwPoints, idwWeights);
     PERF_SAMPLE_END()
 
     SANITY_CHECK_NOTHING();
-}
\ No newline at end of file
+}
diff --git a/src/idw.cpp b/src/idw.cpp
index 7a08ac8..4b9315f 100644
--- a/src/idw.cpp
+++ b/src/idw.cpp
@@ -30,12 +30,12 @@ void idw_halide(const uint8_t* src, uint8_t* dst, int height, int width, int* po
     idw(input, output);
 #else
     static Func f("idw");
-    
+
     // try {
     if (!f.defined()) {
         Var x("x"), y("y");
         RDom r(0, pointCount);
-        
+
         f(x, y) = 0.F;
         Expr x0 = points(3*r+1);
         Expr y0 = points(3*r);
@@ -45,7 +45,7 @@ void idw_halide(const uint8_t* src, uint8_t* dst, int height, int width, int* po
         f(x, y) += hypot(dx, dy) * weight;
 
         // f.vectorize(r, 8);
-        const int factor = 4; 
+        const int factor = 4;
         f.update().atomic().vectorize(r, factor);
 
         // Compile
@@ -79,57 +79,22 @@ void idw_halide(const uint8_t* src, uint8_t* dst, int height, int width, int* po
     f.realize(mask);
 #endif
 
-    //
-    float maxVal = 193.0 //-numeric_limits<float>::infinity();
-    float minVal = 58.0 //numeric_limits<float>::infinity();
-    // for (int y = 0; y < height; y++) {
-    //     for (int x = 0; x < width; x++) {
-    //         if (maskBuf[y*width + x] < minVal) {
-    //             minVal = maskBuf[y*width + x];
-    //         }
-    //         if (maskBuf[y*width + x] > maxVal) {
-    //             maxVal = maskBuf[y*width + x];
-    //         }
-    //     }
-    // }
-
+    float maxVal = 193.0;
+    float minVal = 58.0;
     float diff = maxVal - minVal;
     for (int y = 0; y < height; y++) {
         for (int x = 0; x < width; x++) {
             dst[y*width + x] = (uint8_t) (255 * (maskBuf[y*width + x] - minVal) / diff);
         }
     }
-    //
-
     delete[] maskBuf;
-// #ifdef __riscv
-//     idw(input, output);
-// #else
-//     static Func f("idw");
-//     if (!f.defined()) {
-//         Var x("x"), y("y");
-//         RDom r(0, pointCount);
-        
-//         f(x, y) = 0;
-//         Expr x0 = points(r.x, 0);
-//         Expr y0 = points(r.x, 1);
-//         Expr dx = x - x0;
-//         Expr dy = y - y0;
-//         f(x, y) += sqrt(dx*dx + dy*dy) * weights(r.x);
-//         f(x, y) = cast<uint8_t>(f(x, y));
-        
-//         // Dump AOT code
-//         f.compile_to_header("idw.h", {input}, "idw", target);
-//         f.compile_to_assembly("idw.s", {input}, "idw", target);
-//     }
-// #endif
 }
 
 void idw_ref(const uint8_t* src, uint8_t* dst, int height, int width, int* points, float* weights) {
     float* mask = new float[height * width]();
 
-    float maxVal = 193.0 //-numeric_limits<float>::infinity();
-    float minVal = 58.0 //numeric_limits<float>::infinity();
+    float maxVal = 193.0;
+    float minVal = 58.0;
 
     for (int y = 0; y < height; y++) {
         for (int x = 0; x < width; x++) {
@@ -144,12 +109,6 @@ void idw_ref(const uint8_t* src, uint8_t* dst, int height, int width, int* point
             }
 
             mask[y*width + x] = dot;
-            // if (dot < minVal) {
-            //     minVal = dot;
-            // }
-            // if (dot > maxVal) {
-            //     maxVal = dot;
-            // }
         }
     }
 

From 0bf2bfe4ae76b539361b0c985666c299ada3da25 Mon Sep 17 00:00:00 2001
From: Dmitry Kurtaev <dmitry.kurtaev@gmail.com>
Date: Wed, 7 Feb 2024 21:03:44 +0300
Subject: [PATCH 12/15] Fix rvv

---
 aot/idw.s   | 2 +-
 src/idw.cpp | 4 ++--
 2 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/aot/idw.s b/aot/idw.s
index 45ac36e..647cee3 100644
--- a/aot/idw.s
+++ b/aot/idw.s
@@ -266,7 +266,7 @@ idw:                                    # @idw
 	vlwu.v	v8, (s10)
 	vfmul.vv	v8, v9, v8
 	vfmv.s.f	v9, ft0
-	vfredusum.vs	v8, v8, v9
+	vfredsum.vs	v8, v8, v9
 	vfmv.f.s	ft0, v8
 	addi	s5, s5, -1
 	addi	s10, s10, 16
diff --git a/src/idw.cpp b/src/idw.cpp
index 4b9315f..d974449 100644
--- a/src/idw.cpp
+++ b/src/idw.cpp
@@ -9,7 +9,7 @@ using namespace std;
 
 #ifdef __riscv
   #include <HalideBuffer.h>
-  #include "histogram.h"
+  #include "idw.h"
   using namespace Halide::Runtime;
 #else
   #include <Halide.h>
@@ -27,7 +27,7 @@ void idw_halide(const uint8_t* src, uint8_t* dst, int height, int width, int* po
     Buffer<int> points(pointsBuf, {pointCount*3});
     Buffer<float> weights(weightsBuf, {pointCount});
 #ifdef __riscv
-    idw(input, output);
+    idw(mask);
 #else
     static Func f("idw");
 

From 3b97268b3450a5058fcae17b4ec4a816f514fcb9 Mon Sep 17 00:00:00 2001
From: Dmitry Kurtaev <dmitry.kurtaev@gmail.com>
Date: Wed, 7 Feb 2024 23:02:45 +0300
Subject: [PATCH 13/15] Update test.yaml

---
 .github/workflows/test.yaml | 1 +
 1 file changed, 1 insertion(+)

diff --git a/.github/workflows/test.yaml b/.github/workflows/test.yaml
index e1fd539..34d7e31 100644
--- a/.github/workflows/test.yaml
+++ b/.github/workflows/test.yaml
@@ -45,6 +45,7 @@ jobs:
         cd build
         rm ${{ github.workspace }}/aot/*
         ./perf_algo --gtest_filter=*halide*
+        sed -i 's|vfredusum|vfredsum|' idw.s
         mv *.h *.s halide_runtime.o ${{ github.workspace }}/aot
 
     - name: Build for RISC-V

From 7d7906140ef0c117696662e38d0b89b8ceeff67f Mon Sep 17 00:00:00 2001
From: uiop07558 <uiop07558@gmail.com>
Date: Thu, 8 Feb 2024 10:44:01 +0300
Subject: [PATCH 14/15] fix merge artifacts

---
 aot/idw.s | 4 ----
 1 file changed, 4 deletions(-)

diff --git a/aot/idw.s b/aot/idw.s
index 7fa4310..647cee3 100644
--- a/aot/idw.s
+++ b/aot/idw.s
@@ -266,11 +266,7 @@ idw:                                    # @idw
 	vlwu.v	v8, (s10)
 	vfmul.vv	v8, v9, v8
 	vfmv.s.f	v9, ft0
-<<<<<<< HEAD
-	vfredusum.vs	v8, v8, v9
-=======
 	vfredsum.vs	v8, v8, v9
->>>>>>> a6d731b66e883d018fbaefd444edc23e7aa09c80
 	vfmv.f.s	ft0, v8
 	addi	s5, s5, -1
 	addi	s10, s10, 16

From 1fcc642de0616d9fc74839b72beeb06244cd191e Mon Sep 17 00:00:00 2001
From: uiop07558 <uiop07558@gmail.com>
Date: Thu, 8 Feb 2024 13:08:26 +0300
Subject: [PATCH 15/15] add variants of idw_halide and minor bugfixes

---
 .vscode/settings.json         |   4 +-
 aot/idw.s                     | 296 +++++---------
 aot/idw_halide.h              |  56 +++
 aot/idw_halide.s              | 362 +++++++++++++++++
 aot/idw_halide_parallel.h     |  56 +++
 aot/idw_halide_parallel.s     | 598 +++++++++++++++++++++++++++
 aot/idw_halide_parallel_vec.h |  56 +++
 aot/idw_halide_parallel_vec.s | 739 ++++++++++++++++++++++++++++++++++
 aot/idw_halide_vec.h          |  56 +++
 aot/idw_halide_vec.s          | 450 +++++++++++++++++++++
 include/algos.hpp             |   6 +
 perf/perf_main.cpp            |  59 ++-
 src/idw.cpp                   | 234 ++++++++++-
 test/test_main.cpp            |  14 +
 14 files changed, 2778 insertions(+), 208 deletions(-)
 create mode 100644 aot/idw_halide.h
 create mode 100644 aot/idw_halide.s
 create mode 100644 aot/idw_halide_parallel.h
 create mode 100644 aot/idw_halide_parallel.s
 create mode 100644 aot/idw_halide_parallel_vec.h
 create mode 100644 aot/idw_halide_parallel_vec.s
 create mode 100644 aot/idw_halide_vec.h
 create mode 100644 aot/idw_halide_vec.s

diff --git a/.vscode/settings.json b/.vscode/settings.json
index 9b18adb..90a6474 100644
--- a/.vscode/settings.json
+++ b/.vscode/settings.json
@@ -5,6 +5,8 @@
         "iostream": "cpp",
         "cmath": "cpp",
         "optional": "cpp",
-        "new": "cpp"
+        "new": "cpp",
+        "ostream": "cpp",
+        "numbers": "cpp"
     }
 }
\ No newline at end of file
diff --git a/aot/idw.s b/aot/idw.s
index 647cee3..12c9a39 100644
--- a/aot/idw.s
+++ b/aot/idw.s
@@ -8,41 +8,49 @@
 	.type	idw,@function
 idw:                                    # @idw
 # %bb.0:                                # %entry
-	addi	sp, sp, -192
-	sd	ra, 184(sp)                     # 8-byte Folded Spill
-	sd	s0, 176(sp)                     # 8-byte Folded Spill
-	sd	s1, 168(sp)                     # 8-byte Folded Spill
-	sd	s2, 160(sp)                     # 8-byte Folded Spill
-	sd	s3, 152(sp)                     # 8-byte Folded Spill
-	sd	s4, 144(sp)                     # 8-byte Folded Spill
-	sd	s5, 136(sp)                     # 8-byte Folded Spill
-	sd	s6, 128(sp)                     # 8-byte Folded Spill
-	sd	s7, 120(sp)                     # 8-byte Folded Spill
-	sd	s8, 112(sp)                     # 8-byte Folded Spill
-	sd	s9, 104(sp)                     # 8-byte Folded Spill
-	sd	s10, 96(sp)                     # 8-byte Folded Spill
-	sd	s11, 88(sp)                     # 8-byte Folded Spill
+	addi	sp, sp, -96
+	sd	ra, 88(sp)                      # 8-byte Folded Spill
+	sd	s0, 80(sp)                      # 8-byte Folded Spill
+	sd	s1, 72(sp)                      # 8-byte Folded Spill
+	sd	s2, 64(sp)                      # 8-byte Folded Spill
+	sd	s3, 56(sp)                      # 8-byte Folded Spill
+	sd	s4, 48(sp)                      # 8-byte Folded Spill
+	sd	s5, 40(sp)                      # 8-byte Folded Spill
+	sd	s6, 32(sp)                      # 8-byte Folded Spill
+	sd	s7, 24(sp)                      # 8-byte Folded Spill
+	sd	s8, 16(sp)                      # 8-byte Folded Spill
+	sd	s9, 8(sp)                       # 8-byte Folded Spill
+	sd	s10, 0(sp)                      # 8-byte Folded Spill
 .Lpcrel_hi0:
 	auipc	a1, %pcrel_hi(.Lb2.buffer)
 	addi	a1, a1, %pcrel_lo(.Lpcrel_hi0)
-	ld	s6, 16(a1)
+	ld	s8, 16(a1)
 	ld	a3, 40(a0)
 .Lpcrel_hi1:
 	auipc	a2, %pcrel_hi(.Lb3.buffer)
 	addi	a2, a2, %pcrel_lo(.Lpcrel_hi1)
 	ld	s9, 16(a2)
-	lw	s4, 0(a3)
-	lwu	s8, 4(a3)
-	lw	a4, 16(a3)
-	sd	a4, 8(sp)                       # 8-byte Folded Spill
+	lw	s3, 0(a3)
+	lw	s7, 4(a3)
+	lw	s4, 16(a3)
 	ld	a4, 0(a1)
-	lwu	a5, 20(a3)
-	sd	a5, 32(sp)                      # 8-byte Folded Spill
-	lw	s1, 24(a3)
-	or	a5, s6, a4
+	lwu	s5, 20(a3)
+	lw	s6, 24(a3)
+	or	a5, s8, a4
 	mv	a4, s9
-	bnez	a5, .LBB0_2
-# %bb.1:                                # %then_bb
+	beqz	a5, .LBB0_4
+# %bb.1:                                # %after_bb
+	ld	a5, 0(a2)
+	ld	s10, 16(a0)
+	or	a4, a4, a5
+	beqz	a4, .LBB0_5
+.LBB0_2:                                # %after_bb1
+	sext.w	a4, s5
+	beqz	s10, .LBB0_6
+.LBB0_3:
+	li	a0, 0
+	j	.LBB0_9
+.LBB0_4:                                # %then_bb
 	lui	a4, 128
 	addiw	a4, a4, 9
 	slli	a4, a4, 13
@@ -59,12 +67,11 @@ idw:                                    # @idw
 	sw	zero, 12(a4)
 	ld	a4, 16(a2)
 	sd	zero, 24(a1)
-.LBB0_2:                                # %after_bb
 	ld	a5, 0(a2)
-	ld	s11, 16(a0)
+	ld	s10, 16(a0)
 	or	a4, a4, a5
-	bnez	a4, .LBB0_4
-# %bb.3:                                # %then_bb2
+	bnez	a4, .LBB0_2
+.LBB0_5:                                # %then_bb2
 	sd	zero, 16(a2)
 	sd	zero, 8(a2)
 	sd	zero, 0(a2)
@@ -81,15 +88,8 @@ idw:                                    # @idw
 	sw	a4, 8(a5)
 	sw	zero, 12(a5)
 	sd	zero, 24(a2)
-.LBB0_4:                                # %after_bb1
-	sext.w	a4, s8
-	sd	a4, 24(sp)                      # 8-byte Folded Spill
-	lw	a4, 32(sp)                      # 8-byte Folded Reload
-	sd	s1, 16(sp)                      # 8-byte Folded Spill
-	beqz	s11, .LBB0_6
-# %bb.5:
-	li	a0, 0
-	j	.LBB0_9
+	sext.w	a4, s5
+	bnez	s10, .LBB0_3
 .LBB0_6:                                # %_halide_buffer_is_bounds_query.exit36
 	ld	a5, 0(a0)
 	bnez	a5, .LBB0_8
@@ -102,16 +102,14 @@ idw:                                    # @idw
 	slli	a5, a5, 13
 	addi	a5, a5, 2
 	sd	a5, 32(a0)
-	sw	s4, 0(a3)
-	ld	s1, 24(sp)                      # 8-byte Folded Reload
-	sw	s1, 4(a3)
+	sw	s3, 0(a3)
+	sw	s7, 4(a3)
 	li	a5, 1
 	sw	a5, 8(a3)
 	sw	zero, 12(a3)
-	ld	a5, 8(sp)                       # 8-byte Folded Reload
-	sw	a5, 16(a3)
+	sw	s4, 16(a3)
 	sw	a4, 20(a3)
-	sw	s1, 24(a3)
+	sw	s7, 24(a3)
 	sw	zero, 28(a3)
 	sd	zero, 24(a0)
 .LBB0_8:                                # %land.rhs.i49
@@ -132,174 +130,88 @@ idw:                                    # @idw
 	or	a0, a0, a1
 	bnez	a0, .LBB0_20
 # %bb.10:                               # %"for idw.s0.y.rebased.preheader"
-	ld	s3, 16(sp)                      # 8-byte Folded Reload
-	ld	a0, 24(sp)                      # 8-byte Folded Reload
-	blez	a0, .LBB0_13
+	blez	s7, .LBB0_20
 # %bb.11:                               # %"for idw.s0.y.rebased.us.preheader"
 	li	s0, 0
-	ld	s2, 24(sp)                      # 8-byte Folded Reload
-	slli	s2, s2, 2
-	ld	s1, 32(sp)                      # 8-byte Folded Reload
+	slli	s2, s7, 2
+	mv	s1, s5
 .LBB0_12:                               # %"for idw.s0.y.rebased.us"
                                         # =>This Inner Loop Header: Depth=1
 	slli	a0, s0, 2
-	add	a0, a0, s11
+	add	a0, a0, s10
 	li	a1, 0
 	mv	a2, s2
 	call	memset@plt
 	addi	s1, s1, -1
-	addw	s0, s0, s3
+	addw	s0, s0, s6
 	bnez	s1, .LBB0_12
-.LBB0_13:                               # %"for idw.s1.y.rebased.preheader"
-	sd	zero, 40(sp)                    # 8-byte Folded Spill
-	li	a1, 6
-	li	a2, 1
-	addi	a3, sp, 64
-	addi	a7, sp, 76
-	addi	t0, sp, 72
-	addi	t1, sp, 68
-	li	s0, 4
-	addi	s1, sp, 48
-	addi	t2, sp, 52
-	addi	t3, sp, 60
-	addi	t6, sp, 56
-	li	ra, 2
-	li	a6, 3
-.LBB0_14:                               # %"for idw.s1.y.rebased"
+# %bb.13:                               # %"for idw.s1.y.rebased.preheader"
+	blez	s7, .LBB0_20
+# %bb.14:                               # %"for idw.s1.y.rebased.us.preheader"
+	li	a6, 0
+	addi	s8, s8, 4
+.LBB0_15:                               # %"for idw.s1.y.rebased.us"
                                         # =>This Loop Header: Depth=1
                                         #     Child Loop BB0_16 Depth 2
                                         #       Child Loop BB0_17 Depth 3
-	ld	a0, 24(sp)                      # 8-byte Folded Reload
-	blez	a0, .LBB0_19
-# %bb.15:                               # %"for idw.s1.x.rebased.preheader"
-                                        #   in Loop: Header=BB0_14 Depth=1
-	li	t4, 0
-	ld	a0, 16(sp)                      # 8-byte Folded Reload
-	ld	a4, 40(sp)                      # 8-byte Folded Reload
-	mulw	t5, a0, a4
-	ld	a0, 8(sp)                       # 8-byte Folded Reload
-	addw	a0, a0, a4
-.LBB0_16:                               # %"for idw.s1.x.rebased"
-                                        #   Parent Loop BB0_14 Depth=1
+	li	a1, 0
+	mulw	a7, s6, a6
+	add	a3, s4, a6
+.LBB0_16:                               # %"for idw.s1.x.rebased.us"
+                                        #   Parent Loop BB0_15 Depth=1
                                         # =>  This Loop Header: Depth=2
                                         #       Child Loop BB0_17 Depth 3
-	add	a4, t4, t5
-	slli	a4, a4, 2
-	add	s2, s11, a4
-	flw	ft0, 0(s2)
-	addw	a4, s4, t4
-	li	s5, 25
-	mv	s3, s6
-	mv	s10, s9
-.LBB0_17:                               # %"for idw.s1.r4$x.r4$x"
-                                        #   Parent Loop BB0_14 Depth=1
+	add	a0, a1, a7
+	slli	a0, a0, 2
+	add	t0, s10, a0
+	flw	ft0, 0(t0)
+	add	a5, s3, a1
+	li	a0, 100
+	mv	s1, s8
+	mv	s0, s9
+.LBB0_17:                               # %"for idw.s1.r4$x.us"
+                                        #   Parent Loop BB0_15 Depth=1
                                         #     Parent Loop BB0_16 Depth=2
                                         # =>    This Inner Loop Header: Depth=3
-	addi	s7, s3, 4
-	vsetvli	zero, a1, e32, m2
-	vlwu.v	v8, (s7)
-	vsetvli	zero, a2, e32, m2
-	vslidedown.vi	v10, v8, 3
-	addi	a5, s3, 20
-	vsetvli	zero, a1, e32, m2
-	vlwu.v	v12, (a5)
-	vsetvli	zero, a2, e32, m2
-	vslidedown.vi	v14, v12, 2
-	vslidedown.vi	v12, v12, 5
-	vsw.v	v8, (a3)
-	vsw.v	v12, (a7)
-	vsw.v	v14, (t0)
-	vsw.v	v10, (t1)
-	vsetvli	zero, s0, e32, m1
-	vlwu.v	v8, (a3)
-	vsetvli	a5, zero, e32, m1
-	vrsub.vx	v8, v8, a4
-	vsetvli	zero, a1, e32, m2
-	vlwu.v	v10, (s3)
-	vsetvli	zero, a2, e32, m2
-	vslidedown.vi	v12, v10, 3
-	addi	a5, s3, 16
-	vsetvli	zero, a1, e32, m2
-	vlwu.v	v14, (a5)
-	vsetvli	zero, a2, e32, m2
-	vslidedown.vi	v16, v14, 2
-	vslidedown.vi	v14, v14, 5
-	vsw.v	v10, (s1)
-	vsw.v	v12, (t2)
-	vsw.v	v14, (t3)
-	vsw.v	v16, (t6)
-	vsetvli	zero, s0, e32, m1
-	vlwu.v	v9, (s1)
-	vsetvli	a5, zero, e32, m1
-	vrsub.vx	v9, v9, a0
-	vmul.vv	v9, v9, v9
-	vmacc.vv	v9, v8, v8
-	vfcvt.f.x.v	v8, v9
-	vfmv.f.s	ft1, v8
-	fsqrt.s	ft1, ft1
-	vfmv.s.f	v9, ft1
-	vsetvli	zero, a2, e32, m1
-	vslidedown.vi	v10, v8, 1
-	vfmv.f.s	ft1, v10
-	fsqrt.s	ft1, ft1
-	vsetvli	a5, zero, e32, m1
-	vfmv.s.f	v10, ft1
-	vsetvli	zero, ra, e32, m1
-	vslideup.vi	v9, v10, 1
-	vsetvli	zero, a2, e32, m1
-	vslidedown.vi	v10, v8, 2
-	vfmv.f.s	ft1, v10
-	fsqrt.s	ft1, ft1
-	vsetvli	a5, zero, e32, m1
-	vfmv.s.f	v10, ft1
-	vsetvli	zero, a6, e32, m1
-	vslideup.vi	v9, v10, 2
-	vsetvli	zero, a2, e32, m1
-	vslidedown.vi	v8, v8, 3
-	vfmv.f.s	ft1, v8
-	fsqrt.s	ft1, ft1
-	vsetvli	a5, zero, e32, m1
-	vfmv.s.f	v8, ft1
-	vsetvli	zero, s0, e32, m1
-	vslideup.vi	v9, v8, 3
-	vsetvli	zero, zero, e32, m1
-	vlwu.v	v8, (s10)
-	vfmul.vv	v8, v9, v8
-	vfmv.s.f	v9, ft0
-	vfredsum.vs	v8, v8, v9
-	vfmv.f.s	ft0, v8
-	addi	s5, s5, -1
-	addi	s10, s10, 16
-	addi	s3, s3, 48
-	bnez	s5, .LBB0_17
-# %bb.18:                               # %"end for idw.s1.r4$x.r4$x"
+	lw	a2, 0(s1)
+	lw	a4, -4(s1)
+	subw	a2, a5, a2
+	subw	a4, a3, a4
+	mulw	a2, a2, a2
+	mulw	a4, a4, a4
+	flw	ft1, 0(s0)
+	add	a2, a2, a4
+	fcvt.s.w	ft2, a2
+	fsqrt.s	ft2, ft2
+	fmadd.s	ft0, ft2, ft1, ft0
+	addi	a0, a0, -1
+	addi	s0, s0, 4
+	addi	s1, s1, 12
+	bnez	a0, .LBB0_17
+# %bb.18:                               # %"end for idw.s1.r4$x.us"
                                         #   in Loop: Header=BB0_16 Depth=2
-	addi	t4, t4, 1
-	fsw	ft0, 0(s2)
-	bne	t4, s8, .LBB0_16
-.LBB0_19:                               # %"end for idw.s1.x.rebased"
-                                        #   in Loop: Header=BB0_14 Depth=1
-	ld	a4, 40(sp)                      # 8-byte Folded Reload
-	addi	a4, a4, 1
-	ld	a0, 32(sp)                      # 8-byte Folded Reload
-	sd	a4, 40(sp)                      # 8-byte Folded Spill
-	bne	a4, a0, .LBB0_14
+	addi	a1, a1, 1
+	fsw	ft0, 0(t0)
+	bne	a1, s7, .LBB0_16
+# %bb.19:                               # %"end for idw.s1.x.rebased.loopexit.us"
+                                        #   in Loop: Header=BB0_15 Depth=1
+	addi	a6, a6, 1
+	bne	a6, s5, .LBB0_15
 .LBB0_20:                               # %destructor_block
 	li	a0, 0
-	ld	ra, 184(sp)                     # 8-byte Folded Reload
-	ld	s0, 176(sp)                     # 8-byte Folded Reload
-	ld	s1, 168(sp)                     # 8-byte Folded Reload
-	ld	s2, 160(sp)                     # 8-byte Folded Reload
-	ld	s3, 152(sp)                     # 8-byte Folded Reload
-	ld	s4, 144(sp)                     # 8-byte Folded Reload
-	ld	s5, 136(sp)                     # 8-byte Folded Reload
-	ld	s6, 128(sp)                     # 8-byte Folded Reload
-	ld	s7, 120(sp)                     # 8-byte Folded Reload
-	ld	s8, 112(sp)                     # 8-byte Folded Reload
-	ld	s9, 104(sp)                     # 8-byte Folded Reload
-	ld	s10, 96(sp)                     # 8-byte Folded Reload
-	ld	s11, 88(sp)                     # 8-byte Folded Reload
-	addi	sp, sp, 192
+	ld	ra, 88(sp)                      # 8-byte Folded Reload
+	ld	s0, 80(sp)                      # 8-byte Folded Reload
+	ld	s1, 72(sp)                      # 8-byte Folded Reload
+	ld	s2, 64(sp)                      # 8-byte Folded Reload
+	ld	s3, 56(sp)                      # 8-byte Folded Reload
+	ld	s4, 48(sp)                      # 8-byte Folded Reload
+	ld	s5, 40(sp)                      # 8-byte Folded Reload
+	ld	s6, 32(sp)                      # 8-byte Folded Reload
+	ld	s7, 24(sp)                      # 8-byte Folded Reload
+	ld	s8, 16(sp)                      # 8-byte Folded Reload
+	ld	s9, 8(sp)                       # 8-byte Folded Reload
+	ld	s10, 0(sp)                      # 8-byte Folded Reload
+	addi	sp, sp, 96
 	ret
 .Lfunc_end0:
 	.size	idw, .Lfunc_end0-idw
diff --git a/aot/idw_halide.h b/aot/idw_halide.h
new file mode 100644
index 0000000..e06605e
--- /dev/null
+++ b/aot/idw_halide.h
@@ -0,0 +1,56 @@
+#ifndef HALIDE__idw_halide_h
+#define HALIDE__idw_halide_h
+#include <stdint.h>
+
+// Forward declarations of the types used in the interface
+// to the Halide pipeline.
+//
+// For the definitions of these structs, include HalideRuntime.h
+
+// Halide's representation of a multi-dimensional array.
+// Halide::Runtime::Buffer is a more user-friendly wrapper
+// around this. Its declaration is in HalideBuffer.h
+struct halide_buffer_t;
+
+// Metadata describing the arguments to the generated function.
+// Used to construct calls to the _argv version of the function.
+struct halide_filter_metadata_t;
+
+#ifndef HALIDE_MUST_USE_RESULT
+#ifdef __has_attribute
+#if __has_attribute(nodiscard)
+#define HALIDE_MUST_USE_RESULT [[nodiscard]]
+#elif __has_attribute(warn_unused_result)
+#define HALIDE_MUST_USE_RESULT __attribute__((warn_unused_result))
+#else
+#define HALIDE_MUST_USE_RESULT
+#endif
+#else
+#define HALIDE_MUST_USE_RESULT
+#endif
+#endif
+
+#ifndef HALIDE_FUNCTION_ATTRS
+#define HALIDE_FUNCTION_ATTRS
+#endif
+
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+HALIDE_FUNCTION_ATTRS
+int idw_halide_(struct halide_buffer_t *_idw_halide__buffer);
+
+HALIDE_FUNCTION_ATTRS
+int idw_halide__argv(void **args);
+
+HALIDE_FUNCTION_ATTRS
+const struct halide_filter_metadata_t *idw_halide__metadata();
+
+#ifdef __cplusplus
+}  // extern "C"
+#endif
+
+#endif
diff --git a/aot/idw_halide.s b/aot/idw_halide.s
new file mode 100644
index 0000000..74b4a7e
--- /dev/null
+++ b/aot/idw_halide.s
@@ -0,0 +1,362 @@
+	.text
+	.attribute	4, 16
+	.attribute	5, "rv64gcv0p7"
+	.file	"halide_buffer_t.cpp"
+	.section	.text.idw_halide_,"ax",@progbits
+	.globl	idw_halide_                     # -- Begin function idw_halide_
+	.p2align	1
+	.type	idw_halide_,@function
+idw_halide_:                            # @idw_halide_
+# %bb.0:                                # %entry
+	addi	sp, sp, -96
+	sd	ra, 88(sp)                      # 8-byte Folded Spill
+	sd	s0, 80(sp)                      # 8-byte Folded Spill
+	sd	s1, 72(sp)                      # 8-byte Folded Spill
+	sd	s2, 64(sp)                      # 8-byte Folded Spill
+	sd	s3, 56(sp)                      # 8-byte Folded Spill
+	sd	s4, 48(sp)                      # 8-byte Folded Spill
+	sd	s5, 40(sp)                      # 8-byte Folded Spill
+	sd	s6, 32(sp)                      # 8-byte Folded Spill
+	sd	s7, 24(sp)                      # 8-byte Folded Spill
+	sd	s8, 16(sp)                      # 8-byte Folded Spill
+	sd	s9, 8(sp)                       # 8-byte Folded Spill
+	sd	s10, 0(sp)                      # 8-byte Folded Spill
+.Lpcrel_hi0:
+	auipc	a1, %pcrel_hi(.Lb2.buffer)
+	addi	a1, a1, %pcrel_lo(.Lpcrel_hi0)
+	ld	s8, 16(a1)
+	ld	a3, 40(a0)
+.Lpcrel_hi1:
+	auipc	a2, %pcrel_hi(.Lb3.buffer)
+	addi	a2, a2, %pcrel_lo(.Lpcrel_hi1)
+	ld	s9, 16(a2)
+	lw	s3, 0(a3)
+	lw	s7, 4(a3)
+	lw	s4, 16(a3)
+	ld	a4, 0(a1)
+	lwu	s5, 20(a3)
+	lw	s6, 24(a3)
+	or	a5, s8, a4
+	mv	a4, s9
+	beqz	a5, .LBB0_4
+# %bb.1:                                # %after_bb
+	ld	a5, 0(a2)
+	ld	s10, 16(a0)
+	or	a4, a4, a5
+	beqz	a4, .LBB0_5
+.LBB0_2:                                # %after_bb1
+	sext.w	a4, s5
+	beqz	s10, .LBB0_6
+.LBB0_3:
+	li	a0, 0
+	j	.LBB0_9
+.LBB0_4:                                # %then_bb
+	lui	a4, 128
+	addiw	a4, a4, 9
+	slli	a4, a4, 13
+	sd	a4, 32(a1)
+	ld	a4, 40(a1)
+	sd	zero, 0(a1)
+	sd	zero, 8(a1)
+	sd	zero, 16(a1)
+	sw	zero, 0(a4)
+	li	a5, 300
+	sw	a5, 4(a4)
+	li	a5, 1
+	sw	a5, 8(a4)
+	sw	zero, 12(a4)
+	ld	a4, 16(a2)
+	sd	zero, 24(a1)
+	ld	a5, 0(a2)
+	ld	s10, 16(a0)
+	or	a4, a4, a5
+	bnez	a4, .LBB0_2
+.LBB0_5:                                # %then_bb2
+	sd	zero, 16(a2)
+	sd	zero, 8(a2)
+	sd	zero, 0(a2)
+	lui	a4, 128
+	addiw	a4, a4, 9
+	ld	a5, 40(a2)
+	slli	a4, a4, 13
+	addi	a4, a4, 2
+	sd	a4, 32(a2)
+	sw	zero, 0(a5)
+	li	a4, 100
+	sw	a4, 4(a5)
+	li	a4, 1
+	sw	a4, 8(a5)
+	sw	zero, 12(a5)
+	sd	zero, 24(a2)
+	sext.w	a4, s5
+	bnez	s10, .LBB0_3
+.LBB0_6:                                # %_halide_buffer_is_bounds_query.exit36
+	ld	a5, 0(a0)
+	bnez	a5, .LBB0_8
+# %bb.7:                                # %then_bb5
+	sd	zero, 16(a0)
+	sd	zero, 8(a0)
+	sd	zero, 0(a0)
+	lui	a5, 256
+	addiw	a5, a5, 9
+	slli	a5, a5, 13
+	addi	a5, a5, 2
+	sd	a5, 32(a0)
+	sw	s3, 0(a3)
+	sw	s7, 4(a3)
+	li	a5, 1
+	sw	a5, 8(a3)
+	sw	zero, 12(a3)
+	sw	s4, 16(a3)
+	sw	a4, 20(a3)
+	sw	s7, 24(a3)
+	sw	zero, 28(a3)
+	sd	zero, 24(a0)
+.LBB0_8:                                # %land.rhs.i49
+	ld	a0, 0(a0)
+	seqz	a0, a0
+.LBB0_9:                                # %_halide_buffer_is_bounds_query.exit50
+	ld	a3, 16(a1)
+	ld	a1, 0(a1)
+	ld	a5, 16(a2)
+	ld	a2, 0(a2)
+	or	a1, a1, a3
+	seqz	a1, a1
+	or	a2, a2, a5
+	seqz	a2, a2
+	or	a1, a1, a2
+	or	a0, a0, a1
+	slti	a1, a4, 1
+	or	a0, a0, a1
+	bnez	a0, .LBB0_20
+# %bb.10:                               # %"for idw_halide_.s0.y.rebased.preheader"
+	blez	s7, .LBB0_20
+# %bb.11:                               # %"for idw_halide_.s0.y.rebased.us.preheader"
+	li	s0, 0
+	slli	s2, s7, 2
+	mv	s1, s5
+.LBB0_12:                               # %"for idw_halide_.s0.y.rebased.us"
+                                        # =>This Inner Loop Header: Depth=1
+	slli	a0, s0, 2
+	add	a0, a0, s10
+	li	a1, 0
+	mv	a2, s2
+	call	memset@plt
+	addi	s1, s1, -1
+	addw	s0, s0, s6
+	bnez	s1, .LBB0_12
+# %bb.13:                               # %"for idw_halide_.s1.y.rebased.preheader"
+	blez	s7, .LBB0_20
+# %bb.14:                               # %"for idw_halide_.s1.y.rebased.us.preheader"
+	li	a6, 0
+	addi	s8, s8, 4
+.LBB0_15:                               # %"for idw_halide_.s1.y.rebased.us"
+                                        # =>This Loop Header: Depth=1
+                                        #     Child Loop BB0_16 Depth 2
+                                        #       Child Loop BB0_17 Depth 3
+	li	a1, 0
+	mulw	a7, s6, a6
+	add	a3, s4, a6
+.LBB0_16:                               # %"for idw_halide_.s1.x.rebased.us"
+                                        #   Parent Loop BB0_15 Depth=1
+                                        # =>  This Loop Header: Depth=2
+                                        #       Child Loop BB0_17 Depth 3
+	add	a0, a1, a7
+	slli	a0, a0, 2
+	add	t0, s10, a0
+	flw	ft0, 0(t0)
+	add	a5, s3, a1
+	li	a0, 100
+	mv	s1, s8
+	mv	s0, s9
+.LBB0_17:                               # %"for idw_halide_.s1.r4$x.us"
+                                        #   Parent Loop BB0_15 Depth=1
+                                        #     Parent Loop BB0_16 Depth=2
+                                        # =>    This Inner Loop Header: Depth=3
+	lw	a2, 0(s1)
+	lw	a4, -4(s1)
+	subw	a2, a5, a2
+	subw	a4, a3, a4
+	mulw	a2, a2, a2
+	mulw	a4, a4, a4
+	flw	ft1, 0(s0)
+	add	a2, a2, a4
+	fcvt.s.w	ft2, a2
+	fsqrt.s	ft2, ft2
+	fmadd.s	ft0, ft2, ft1, ft0
+	addi	a0, a0, -1
+	addi	s0, s0, 4
+	addi	s1, s1, 12
+	bnez	a0, .LBB0_17
+# %bb.18:                               # %"end for idw_halide_.s1.r4$x.us"
+                                        #   in Loop: Header=BB0_16 Depth=2
+	addi	a1, a1, 1
+	fsw	ft0, 0(t0)
+	bne	a1, s7, .LBB0_16
+# %bb.19:                               # %"end for idw_halide_.s1.x.rebased.loopexit.us"
+                                        #   in Loop: Header=BB0_15 Depth=1
+	addi	a6, a6, 1
+	bne	a6, s5, .LBB0_15
+.LBB0_20:                               # %destructor_block
+	li	a0, 0
+	ld	ra, 88(sp)                      # 8-byte Folded Reload
+	ld	s0, 80(sp)                      # 8-byte Folded Reload
+	ld	s1, 72(sp)                      # 8-byte Folded Reload
+	ld	s2, 64(sp)                      # 8-byte Folded Reload
+	ld	s3, 56(sp)                      # 8-byte Folded Reload
+	ld	s4, 48(sp)                      # 8-byte Folded Reload
+	ld	s5, 40(sp)                      # 8-byte Folded Reload
+	ld	s6, 32(sp)                      # 8-byte Folded Reload
+	ld	s7, 24(sp)                      # 8-byte Folded Reload
+	ld	s8, 16(sp)                      # 8-byte Folded Reload
+	ld	s9, 8(sp)                       # 8-byte Folded Reload
+	ld	s10, 0(sp)                      # 8-byte Folded Reload
+	addi	sp, sp, 96
+	ret
+.Lfunc_end0:
+	.size	idw_halide_, .Lfunc_end0-idw_halide_
+                                        # -- End function
+	.section	.text.idw_halide__argv,"ax",@progbits
+	.globl	idw_halide__argv                # -- Begin function idw_halide__argv
+	.p2align	1
+	.type	idw_halide__argv,@function
+idw_halide__argv:                       # @idw_halide__argv
+# %bb.0:                                # %entry
+	addi	sp, sp, -16
+	sd	ra, 8(sp)                       # 8-byte Folded Spill
+	ld	a0, 0(a0)
+	call	idw_halide_@plt
+	li	a0, 0
+	ld	ra, 8(sp)                       # 8-byte Folded Reload
+	addi	sp, sp, 16
+	ret
+.Lfunc_end1:
+	.size	idw_halide__argv, .Lfunc_end1-idw_halide__argv
+                                        # -- End function
+	.section	.text.idw_halide__metadata,"ax",@progbits
+	.globl	idw_halide__metadata            # -- Begin function idw_halide__metadata
+	.p2align	1
+	.type	idw_halide__metadata,@function
+idw_halide__metadata:                   # @idw_halide__metadata
+# %bb.0:                                # %entry
+.Lpcrel_hi2:
+	auipc	a0, %pcrel_hi(.Lidw_halide__metadata_storage)
+	addi	a0, a0, %pcrel_lo(.Lpcrel_hi2)
+	ret
+.Lfunc_end2:
+	.size	idw_halide__metadata, .Lfunc_end2-idw_halide__metadata
+                                        # -- End function
+	.type	.Lb2.shape,@object              # @b2.shape
+	.section	.rodata,"a",@progbits
+	.p2align	5, 0x0
+.Lb2.shape:
+	.asciz	"\000\000\000\000,\001\000\000\001\000\000\000\000\000\000"
+	.size	.Lb2.shape, 16
+
+	.type	.Lb2.data,@object               # @b2.data
+	.p2align	5, 0x0
+.Lb2.data:
+	.asciz	"\000\000\000\000\000\000\000\000H\000\000\000\000\000\000\000\325\000\000\000O\000\000\000\000\000\000\000\252\001\000\000<\000\000\000\000\000\000\000\200\002\000\000L\000\000\000\000\000\000\000U\003\000\000\200\000\000\000\000\000\000\000*\004\000\000C\000\000\000\000\000\000\000\000\005\000\000A\000\000\000\000\000\000\000\325\005\000\000@\000\000\000\000\000\000\000\252\006\000\000<\000\000\000\000\000\000\000\200\007\000\000=\000\000\000x\000\000\000\000\000\000\000Q\000\000\000x\000\000\000\325\000\000\000O\000\000\000x\000\000\000\252\001\000\000:\000\000\000x\000\000\000\200\002\000\000\204\000\000\000x\000\000\000U\003\000\000\225\000\000\000x\000\000\000*\004\000\000\216\000\000\000x\000\000\000\000\005\000\000@\000\000\000x\000\000\000\325\005\000\000E\000\000\000x\000\000\000\252\006\000\000A\000\000\000x\000\000\000\200\007\000\000@\000\000\000\360\000\000\000\000\000\000\000K\000\000\000\360\000\000\000\325\000\000\000D\000\000\000\360\000\000\000\252\001\000\000\214\000\000\000\360\000\000\000\200\002\000\000\231\000\000\000\360\000\000\000U\003\000\000\221\000\000\000\360\000\000\000*\004\000\000\204\000\000\000\360\000\000\000\000\005\000\000\230\000\000\000\360\000\000\000\325\005\000\000}\000\000\000\360\000\000\000\252\006\000\000B\000\000\000\360\000\000\000\200\007\000\000:\000\000\000h\001\000\000\000\000\000\000N\000\000\000h\001\000\000\325\000\000\000<\000\000\000h\001\000\000\252\001\000\000\213\000\000\000h\001\000\000\200\002\000\000\250\000\000\000h\001\000\000U\003\000\000\232\000\000\000h\001\000\000*\004\000\000\212\000\000\000h\001\000\000\000\005\000\000\221\000\000\000h\001\000\000\325\005\000\000\240\000\000\000h\001\000\000\252\006\000\000D\000\000\000h\001\000\000\200\007\000\000<\000\000\000\340\001\000\000\000\000\000\000M\000\000\000\340\001\000\000\325\000\000\000;\000\000\000\340\001\000\000\252\001\000\000\245\000\000\000\340\001\000\000\200\002\000\000\267\000\000\000\340\001\000\000U\003\000\000\246\000\000\000\340\001\000\000*\004\000\000\216\000\000\000\340\001\000\000\000\005\000\000{\000\000\000\340\001\000\000\325\005\000\000\233\000\000\000\340\001\000\000\252\006\000\000\220\000\000\000\340\001\000\000\200\007\000\000<\000\000\000X\002\000\000\000\000\000\000S\000\000\000X\002\000\000\325\000\000\000A\000\000\000X\002\000\000\252\001\000\000\262\000\000\000X\002\000\000\200\002\000\000\270\000\000\000X\002\000\000U\003\000\000\212\000\000\000X\002\000\000*\004\000\000|\000\000\000X\002\000\000\000\005\000\000\204\000\000\000X\002\000\000\325\005\000\000\257\000\000\000X\002\000\000\252\006\000\000\257\000\000\000X\002\000\000\200\007\000\000<\000\000\000\320\002\000\000\000\000\000\000V\000\000\000\320\002\000\000\325\000\000\000<\000\000\000\320\002\000\000\252\001\000\000\266\000\000\000\320\002\000\000\200\002\000\000\263\000\000\000\320\002\000\000U\003\000\000\230\000\000\000\320\002\000\000*\004\000\000\214\000\000\000\320\002\000\000\000\005\000\000s\000\000\000\320\002\000\000\325\005\000\000\234\000\000\000\320\002\000\000\252\006\000\000\254\000\000\000\320\002\000\000\200\007\000\000A\000\000\000H\003\000\000\000\000\000\000Z\000\000\000H\003\000\000\325\000\000\000E\000\000\000H\003\000\000\252\001\000\000\244\000\000\000H\003\000\000\200\002\000\000\272\000\000\000H\003\000\000U\003\000\000\222\000\000\000H\003\000\000*\004\000\000\223\000\000\000H\003\000\000\000\005\000\000z\000\000\000H\003\000\000\325\005\000\000\230\000\000\000H\003\000\000\252\006\000\000\246\000\000\000H\003\000\000\200\007\000\000E\000\000\000\300\003\000\000\000\000\000\000U\000\000\000\300\003\000\000\325\000\000\000C\000\000\000\300\003\000\000\252\001\000\000\272\000\000\000\300\003\000\000\200\002\000\000\251\000\000\000\300\003\000\000U\003\000\000\242\000\000\000\300\003\000\000*\004\000\000\223\000\000\000\300\003\000\000\000\005\000\000|\000\000\000\300\003\000\000\325\005\000\000\225\000\000\000\300\003\000\000\252\006\000\000\235\000\000\000\300\003\000\000\200\007\000\000E\000\000\0008\004\000\000\000\000\000\000U\000\000\0008\004\000\000\325\000\000\000F\000\000\0008\004\000\000\252\001\000\000\301\000\000\0008\004\000\000\200\002\000\000\300\000\000\0008\004\000\000U\003\000\000\233\000\000\0008\004\000\000*\004\000\000\207\000\000\0008\004\000\000\000\005\000\000x\000\000\0008\004\000\000\325\005\000\000\212\000\000\0008\004\000\000\252\006\000\000\200\000\000\0008\004\000\000\200\007\000\000F\000\000"
+	.size	.Lb2.data, 1200
+
+	.type	.Lb2.buffer,@object             # @b2.buffer
+	.data
+	.p2align	4, 0x0
+.Lb2.buffer:
+	.quad	0                               # 0x0
+	.quad	0
+	.quad	.Lb2.data
+	.quad	1                               # 0x1
+	.byte	0                               # 0x0
+	.byte	32                              # 0x20
+	.half	1                               # 0x1
+	.word	1                               # 0x1
+	.quad	.Lb2.shape
+	.quad	0
+	.size	.Lb2.buffer, 56
+
+	.type	.Lb3.shape,@object              # @b3.shape
+	.section	.rodata,"a",@progbits
+	.p2align	5, 0x0
+.Lb3.shape:
+	.asciz	"\000\000\000\000d\000\000\000\001\000\000\000\000\000\000"
+	.size	.Lb3.shape, 16
+
+	.type	.Lb3.data,@object               # @b3.data
+	.p2align	5, 0x0
+.Lb3.data:
+	.ascii	"\211\004&=\244\022\233\275\217U%;\272)J>#\026\024\276@_\234>\340\243|\275\334\034K\275\232\300\004:A\313\245\274\303\024r\275\345\313a\275x\231\300>\201\334\"\276\236}\321\275H\346\257\276\202V\310>\225\375(>},\026\274E\235W\275\241\236\236<\245\220\222=\004\361\263\276\2620\240\274\265\036\240=\\:\336=\240\016\263\276\260\224\371\275p\235(=\311\\x;\226y\\\275\360\227\377=\275\364\200=rw/<@U\017=c\327\003=nV\267\273\354<Y\276\257$\260>3\207\311\274<&{\274h\b\031>\203\222\003\276\213\3649\275d\177\017\276\330\306\246\275'\314:>\203\024\253=S\331M\276\257\263S=b\221q\275\336\t\306=#^\005\2763)\256\274\344\255G>\224!\374=\374\360o\275\215#\031\276\322`J\276y6\360=\273\303;\275N\"N>M\022K\276\016\035C=1Fb\275u\017\220\275\0040\033>,\037\223=t\335\304\275\363\344\271=\320\341\260\275\2265\255=\310\"b=&\031\330\275;\314\000>>\347o\275\255\222\020=C\rU<\370d\307\275\373\306\206=I\345\373\274Y|\034>WY;\276O\240?>B\347\226\275\203\271\203\275W\243\313<\214\364\343\274Ey\033\276J4\214=\261\304\241\274-\271p>X\313l\276*_\366\275\260^\\=\237l1=\371\272\200=\002\373?\274*\3578<]Gv="
+	.size	.Lb3.data, 400
+
+	.type	.Lb3.buffer,@object             # @b3.buffer
+	.data
+	.p2align	4, 0x0
+.Lb3.buffer:
+	.quad	0                               # 0x0
+	.quad	0
+	.quad	.Lb3.data
+	.quad	1                               # 0x1
+	.byte	2                               # 0x2
+	.byte	32                              # 0x20
+	.half	1                               # 0x1
+	.word	1                               # 0x1
+	.quad	.Lb3.shape
+	.quad	0
+	.size	.Lb3.buffer, 56
+
+	.type	.L__unnamed_1,@object           # @0
+	.section	.rodata,"a",@progbits
+	.p2align	4, 0x0
+.L__unnamed_1:
+	.zero	32
+	.size	.L__unnamed_1, 32
+
+	.type	.Lstr,@object                   # @str
+	.p2align	5, 0x0
+.Lstr:
+	.asciz	"idw_halide_"
+	.size	.Lstr, 12
+
+	.type	.L__unnamed_2,@object           # @1
+	.section	.data.rel.ro,"aw",@progbits
+	.p2align	4, 0x0
+.L__unnamed_2:
+	.quad	.Lstr
+	.word	2                               # 0x2
+	.word	2                               # 0x2
+	.byte	2                               # 0x2
+	.byte	32                              # 0x20
+	.half	1                               # 0x1
+	.zero	4
+	.quad	0
+	.quad	0
+	.quad	0
+	.quad	0
+	.quad	.L__unnamed_1
+	.size	.L__unnamed_2, 64
+
+	.type	.Lstr.4,@object                 # @str.4
+	.section	.rodata,"a",@progbits
+	.p2align	5, 0x0
+.Lstr.4:
+	.asciz	"riscv-64-linux-no_asserts-no_runtime-rvv-vector_bits_128"
+	.size	.Lstr.4, 57
+
+	.type	.Lidw_halide__metadata_storage,@object # @idw_halide__metadata_storage
+	.section	.data.rel.ro,"aw",@progbits
+	.p2align	4, 0x0
+.Lidw_halide__metadata_storage:
+	.word	1                               # 0x1
+	.word	1                               # 0x1
+	.quad	.L__unnamed_2
+	.quad	.Lstr.4
+	.quad	.Lstr
+	.size	.Lidw_halide__metadata_storage, 32
+
+	.ident	"clang version 16.0.0 (https://github.com/dkurt/llvm-rvv-071 b027aa1b59c9f53240bdc836f39656723fdf9df0)"
+	.ident	"clang version 16.0.0 (https://github.com/dkurt/llvm-rvv-071 b027aa1b59c9f53240bdc836f39656723fdf9df0)"
+	.ident	"clang version 16.0.0 (https://github.com/dkurt/llvm-rvv-071 b027aa1b59c9f53240bdc836f39656723fdf9df0)"
+	.ident	"clang version 16.0.0 (https://github.com/dkurt/llvm-rvv-071 b027aa1b59c9f53240bdc836f39656723fdf9df0)"
+	.ident	"clang version 16.0.0 (https://github.com/dkurt/llvm-rvv-071 b027aa1b59c9f53240bdc836f39656723fdf9df0)"
+	.ident	"clang version 16.0.0 (https://github.com/dkurt/llvm-rvv-071 b027aa1b59c9f53240bdc836f39656723fdf9df0)"
+	.section	".note.GNU-stack","",@progbits
diff --git a/aot/idw_halide_parallel.h b/aot/idw_halide_parallel.h
new file mode 100644
index 0000000..6f272cb
--- /dev/null
+++ b/aot/idw_halide_parallel.h
@@ -0,0 +1,56 @@
+#ifndef HALIDE__idw_halide_parallel_h
+#define HALIDE__idw_halide_parallel_h
+#include <stdint.h>
+
+// Forward declarations of the types used in the interface
+// to the Halide pipeline.
+//
+// For the definitions of these structs, include HalideRuntime.h
+
+// Halide's representation of a multi-dimensional array.
+// Halide::Runtime::Buffer is a more user-friendly wrapper
+// around this. Its declaration is in HalideBuffer.h
+struct halide_buffer_t;
+
+// Metadata describing the arguments to the generated function.
+// Used to construct calls to the _argv version of the function.
+struct halide_filter_metadata_t;
+
+#ifndef HALIDE_MUST_USE_RESULT
+#ifdef __has_attribute
+#if __has_attribute(nodiscard)
+#define HALIDE_MUST_USE_RESULT [[nodiscard]]
+#elif __has_attribute(warn_unused_result)
+#define HALIDE_MUST_USE_RESULT __attribute__((warn_unused_result))
+#else
+#define HALIDE_MUST_USE_RESULT
+#endif
+#else
+#define HALIDE_MUST_USE_RESULT
+#endif
+#endif
+
+#ifndef HALIDE_FUNCTION_ATTRS
+#define HALIDE_FUNCTION_ATTRS
+#endif
+
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+HALIDE_FUNCTION_ATTRS
+int idw_halide_parallel_(struct halide_buffer_t *_idw_halide_parallel__buffer);
+
+HALIDE_FUNCTION_ATTRS
+int idw_halide_parallel__argv(void **args);
+
+HALIDE_FUNCTION_ATTRS
+const struct halide_filter_metadata_t *idw_halide_parallel__metadata();
+
+#ifdef __cplusplus
+}  // extern "C"
+#endif
+
+#endif
diff --git a/aot/idw_halide_parallel.s b/aot/idw_halide_parallel.s
new file mode 100644
index 0000000..7578098
--- /dev/null
+++ b/aot/idw_halide_parallel.s
@@ -0,0 +1,598 @@
+	.text
+	.attribute	4, 16
+	.attribute	5, "rv64gcv0p7"
+	.file	"halide_buffer_t.cpp"
+	.section	.text.idw_halide_parallel__par_for_idw_halide_parallel__s0_x_x_outer_tile_index__1,"ax",@progbits
+	.globl	idw_halide_parallel__par_for_idw_halide_parallel__s0_x_x_outer_tile_index__1 # -- Begin function idw_halide_parallel__par_for_idw_halide_parallel__s0_x_x_outer_tile_index__1
+	.p2align	1
+	.type	idw_halide_parallel__par_for_idw_halide_parallel__s0_x_x_outer_tile_index__1,@function
+idw_halide_parallel__par_for_idw_halide_parallel__s0_x_x_outer_tile_index__1: # @idw_halide_parallel__par_for_idw_halide_parallel__s0_x_x_outer_tile_index__1
+# %bb.0:                                # %entry
+	addi	sp, sp, -48
+	sd	ra, 40(sp)                      # 8-byte Folded Spill
+	sd	s0, 32(sp)                      # 8-byte Folded Spill
+	sd	s1, 24(sp)                      # 8-byte Folded Spill
+	sd	s2, 16(sp)                      # 8-byte Folded Spill
+	sd	s3, 8(sp)                       # 8-byte Folded Spill
+	lw	a4, 24(a2)
+	lw	a7, 8(a2)
+	lw	a6, 12(a2)
+	sraiw	a3, a1, 31
+	seqz	a5, a4
+	negw	a0, a5
+	srai	s1, a4, 31
+	subw	t0, a1, a3
+	or	a0, a0, a4
+	remw	a1, t0, a0
+	xor	s0, s1, a4
+	not	a0, s1
+	add	s0, s0, a0
+	and	s0, s0, a3
+	add	a1, a1, s0
+	li	s0, 480
+	mulw	a1, a1, s0
+	addi	s0, a5, -1
+	and	t2, s0, a1
+	addiw	a1, a7, -480
+	blt	t2, a1, .LBB0_2
+# %bb.1:                                # %entry
+	mv	t2, a1
+.LBB0_2:                                # %entry
+	lw	a7, 16(a2)
+	lw	s2, 20(a2)
+	lw	t1, 28(a2)
+	add	a4, a4, a5
+	divw	a1, t0, a4
+	subw	a0, a0, s1
+	and	a0, a0, a3
+	add	a0, a0, a1
+	li	s1, 270
+	mulw	a0, a0, s1
+	and	a0, a0, s0
+	addiw	a1, a6, -270
+	blt	a0, a1, .LBB0_4
+# %bb.3:                                # %entry
+	mv	a0, a1
+.LBB0_4:                                # %entry
+	ld	s3, 0(a2)
+	add	a0, a0, a7
+	add	t1, t1, t2
+	mulw	a0, s2, a0
+	addw	s0, t1, a0
+.LBB0_5:                                # %"for idw_halide_parallel_.s0.y.y_inner"
+                                        # =>This Inner Loop Header: Depth=1
+	slli	a0, s0, 2
+	add	a0, a0, s3
+	li	a2, 1920
+	li	a1, 0
+	call	memset@plt
+	addi	s1, s1, -1
+	addw	s0, s0, s2
+	bnez	s1, .LBB0_5
+# %bb.6:                                # %destructor_block
+	li	a0, 0
+	ld	ra, 40(sp)                      # 8-byte Folded Reload
+	ld	s0, 32(sp)                      # 8-byte Folded Reload
+	ld	s1, 24(sp)                      # 8-byte Folded Reload
+	ld	s2, 16(sp)                      # 8-byte Folded Reload
+	ld	s3, 8(sp)                       # 8-byte Folded Reload
+	addi	sp, sp, 48
+	ret
+.Lfunc_end0:
+	.size	idw_halide_parallel__par_for_idw_halide_parallel__s0_x_x_outer_tile_index__1, .Lfunc_end0-idw_halide_parallel__par_for_idw_halide_parallel__s0_x_x_outer_tile_index__1
+                                        # -- End function
+	.section	.text.idw_halide_parallel__par_for_idw_halide_parallel__s1_x_x_outer_tile_index__1,"ax",@progbits
+	.globl	idw_halide_parallel__par_for_idw_halide_parallel__s1_x_x_outer_tile_index__1 # -- Begin function idw_halide_parallel__par_for_idw_halide_parallel__s1_x_x_outer_tile_index__1
+	.p2align	1
+	.type	idw_halide_parallel__par_for_idw_halide_parallel__s1_x_x_outer_tile_index__1,@function
+idw_halide_parallel__par_for_idw_halide_parallel__s1_x_x_outer_tile_index__1: # @idw_halide_parallel__par_for_idw_halide_parallel__s1_x_x_outer_tile_index__1
+# %bb.0:                                # %entry
+	addi	sp, sp, -48
+	sd	s0, 40(sp)                      # 8-byte Folded Spill
+	sd	s1, 32(sp)                      # 8-byte Folded Spill
+	sd	s2, 24(sp)                      # 8-byte Folded Spill
+	sd	s3, 16(sp)                      # 8-byte Folded Spill
+	sd	s4, 8(sp)                       # 8-byte Folded Spill
+	sd	s5, 0(sp)                       # 8-byte Folded Spill
+	li	a7, 0
+	ld	s3, 0(a2)
+	ld	t4, 8(a2)
+	ld	t3, 16(a2)
+	lw	a0, 36(a2)
+	lw	t5, 24(a2)
+	lw	t1, 28(a2)
+	sraiw	a3, a1, 31
+	seqz	a4, a0
+	negw	a5, a4
+	srai	s1, a0, 31
+	subw	a1, a1, a3
+	or	a5, a5, a0
+	remw	a6, a1, a5
+	xor	s0, s1, a0
+	not	a5, s1
+	add	s0, s0, a5
+	and	s0, s0, a3
+	add	s0, s0, a6
+	lw	a6, 32(a2)
+	lw	t6, 40(a2)
+	add	a0, a0, a4
+	divw	a0, a1, a0
+	subw	a5, a5, s1
+	and	a3, a3, a5
+	add	a0, a0, a3
+	li	t2, 480
+	mulw	a1, s0, t2
+	addi	a4, a4, -1
+	and	s2, a4, a1
+	li	t0, 270
+	mulw	a0, a0, t0
+	and	a0, a0, a4
+	add	t1, t1, a0
+	addi	s3, s3, 4
+.LBB1_1:                                # %"for idw_halide_parallel_.s1.y.y_inner"
+                                        # =>This Loop Header: Depth=1
+                                        #     Child Loop BB1_2 Depth 2
+                                        #       Child Loop BB1_3 Depth 3
+	li	a4, 0
+	add	a3, t1, a7
+	mulw	s4, a3, a6
+.LBB1_2:                                # %"for idw_halide_parallel_.s1.x.x_inner"
+                                        #   Parent Loop BB1_1 Depth=1
+                                        # =>  This Loop Header: Depth=2
+                                        #       Child Loop BB1_3 Depth 3
+	add	s0, s2, a4
+	add	a0, s0, t6
+	addw	a0, a0, s4
+	slli	a0, a0, 2
+	add	s5, t3, a0
+	flw	ft0, 0(s5)
+	add	s0, s0, t5
+	li	s1, 100
+	mv	a1, s3
+	mv	a0, t4
+.LBB1_3:                                # %"for idw_halide_parallel_.s1.r13$x"
+                                        #   Parent Loop BB1_1 Depth=1
+                                        #     Parent Loop BB1_2 Depth=2
+                                        # =>    This Inner Loop Header: Depth=3
+	lw	a5, 0(a1)
+	lw	a2, -4(a1)
+	subw	a5, s0, a5
+	subw	a2, a3, a2
+	mulw	a5, a5, a5
+	mulw	a2, a2, a2
+	flw	ft1, 0(a0)
+	add	a2, a2, a5
+	fcvt.s.w	ft2, a2
+	fsqrt.s	ft2, ft2
+	fmadd.s	ft0, ft2, ft1, ft0
+	addi	s1, s1, -1
+	addi	a0, a0, 4
+	addi	a1, a1, 12
+	bnez	s1, .LBB1_3
+# %bb.4:                                # %"end for idw_halide_parallel_.s1.r13$x"
+                                        #   in Loop: Header=BB1_2 Depth=2
+	addi	a4, a4, 1
+	fsw	ft0, 0(s5)
+	bne	a4, t2, .LBB1_2
+# %bb.5:                                # %"end for idw_halide_parallel_.s1.x.x_inner"
+                                        #   in Loop: Header=BB1_1 Depth=1
+	addiw	a7, a7, 1
+	bne	a7, t0, .LBB1_1
+# %bb.6:                                # %destructor_block
+	li	a0, 0
+	ld	s0, 40(sp)                      # 8-byte Folded Reload
+	ld	s1, 32(sp)                      # 8-byte Folded Reload
+	ld	s2, 24(sp)                      # 8-byte Folded Reload
+	ld	s3, 16(sp)                      # 8-byte Folded Reload
+	ld	s4, 8(sp)                       # 8-byte Folded Reload
+	ld	s5, 0(sp)                       # 8-byte Folded Reload
+	addi	sp, sp, 48
+	ret
+.Lfunc_end1:
+	.size	idw_halide_parallel__par_for_idw_halide_parallel__s1_x_x_outer_tile_index__1, .Lfunc_end1-idw_halide_parallel__par_for_idw_halide_parallel__s1_x_x_outer_tile_index__1
+                                        # -- End function
+	.section	.text.idw_halide_parallel_,"ax",@progbits
+	.globl	idw_halide_parallel_            # -- Begin function idw_halide_parallel_
+	.p2align	1
+	.type	idw_halide_parallel_,@function
+idw_halide_parallel_:                   # @idw_halide_parallel_
+# %bb.0:                                # %entry
+	addi	sp, sp, -176
+	sd	ra, 168(sp)                     # 8-byte Folded Spill
+	sd	s0, 160(sp)                     # 8-byte Folded Spill
+	sd	s1, 152(sp)                     # 8-byte Folded Spill
+	sd	s2, 144(sp)                     # 8-byte Folded Spill
+	sd	s3, 136(sp)                     # 8-byte Folded Spill
+	sd	s4, 128(sp)                     # 8-byte Folded Spill
+	sd	s5, 120(sp)                     # 8-byte Folded Spill
+	sd	s6, 112(sp)                     # 8-byte Folded Spill
+	sd	s7, 104(sp)                     # 8-byte Folded Spill
+	sd	s8, 96(sp)                      # 8-byte Folded Spill
+	sd	s9, 88(sp)                      # 8-byte Folded Spill
+	ld	t5, 40(a0)
+	lw	a4, 4(t5)
+	lw	t2, 20(t5)
+	li	a2, -480
+	subw	a6, a2, a4
+	addiw	a2, a4, -1
+	blt	a2, a6, .LBB2_2
+# %bb.1:                                # %entry
+	mv	a6, a2
+.LBB2_2:                                # %entry
+.Lpcrel_hi0:
+	auipc	a2, %pcrel_hi(.Lb24.buffer)
+	addi	a3, a2, %pcrel_lo(.Lpcrel_hi0)
+	sgtz	t4, a4
+	slti	a5, a4, -479
+	slti	s0, t2, -269
+	lui	s1, 559241
+	li	a7, 480
+	addiw	t1, s1, -1911
+	mv	t0, a4
+	blt	a4, a7, .LBB2_4
+# %bb.3:                                # %entry
+	li	t0, 480
+.LBB2_4:                                # %entry
+.Lpcrel_hi1:
+	auipc	t6, %pcrel_hi(.Lb25.buffer)
+	ld	s2, 16(a3)
+	add	a1, a4, a5
+	addiw	a1, a1, 479
+	mul	a2, a1, t1
+	srli	a2, a2, 32
+	add	a1, a1, a2
+	srliw	a2, a1, 31
+	sraiw	a1, a1, 8
+	add	a1, a1, a2
+	subw	s9, a1, a5
+	add	a1, t2, s0
+	addiw	a1, a1, 269
+	lui	a2, 994205
+	addiw	a2, a2, 1609
+	mul	a2, a1, a2
+	srli	a2, a2, 32
+	add	a1, a1, a2
+	srliw	a2, a1, 31
+	sraiw	a1, a1, 8
+	add	a1, a1, a2
+	subw	a1, a1, s0
+	mulw	s7, a1, s9
+	slti	a1, s7, 1
+	negw	a2, a1
+	seqz	s1, s9
+	add	a1, a1, s7
+	addiw	a1, a1, -1
+	add	s0, s9, s1
+	divw	a1, a1, s0
+	add	a5, a5, a5
+	addiw	a5, a5, -1
+	and	a2, a2, a5
+	add	a1, a1, a2
+	li	a2, 270
+	mulw	a1, a1, a2
+	addi	s1, s1, -1
+	and	t3, s1, a1
+	addi	t4, t4, -1
+	and	t4, t4, t3
+	addiw	a2, t2, -270
+	addi	a5, t6, %pcrel_lo(.Lpcrel_hi1)
+	blt	t4, a2, .LBB2_6
+# %bb.5:                                # %entry
+	mv	t4, a2
+.LBB2_6:                                # %entry
+	ld	s4, 16(a5)
+	lw	s3, 0(t5)
+	ld	a1, 0(a3)
+	lw	s6, 16(t5)
+	lw	s5, 24(t5)
+	or	a1, s2, a1
+	mv	s0, s4
+	beqz	a1, .LBB2_10
+# %bb.7:                                # %after_bb
+	ld	a1, 0(a5)
+	ld	s8, 16(a0)
+	or	a1, a1, s0
+	beqz	a1, .LBB2_11
+.LBB2_8:                                # %after_bb1
+	beqz	s8, .LBB2_12
+.LBB2_9:
+	li	a0, 0
+	j	.LBB2_18
+.LBB2_10:                               # %then_bb
+	lui	a1, 128
+	addiw	a1, a1, 9
+	slli	a1, a1, 13
+	sd	a1, 32(a3)
+	ld	a1, 40(a3)
+	sd	zero, 0(a3)
+	sd	zero, 8(a3)
+	sd	zero, 16(a3)
+	sw	zero, 0(a1)
+	li	a2, 300
+	sw	a2, 4(a1)
+	li	a2, 1
+	sw	a2, 8(a1)
+	sw	zero, 12(a1)
+	ld	s0, 16(a5)
+	sd	zero, 24(a3)
+	ld	a1, 0(a5)
+	ld	s8, 16(a0)
+	or	a1, a1, s0
+	bnez	a1, .LBB2_8
+.LBB2_11:                               # %then_bb2
+	sd	zero, 16(a5)
+	sd	zero, 8(a5)
+	sd	zero, 0(a5)
+	lui	a1, 128
+	addiw	a1, a1, 9
+	ld	a2, 40(a5)
+	slli	a1, a1, 13
+	addi	a1, a1, 2
+	sd	a1, 32(a5)
+	sw	zero, 0(a2)
+	li	a1, 100
+	sw	a1, 4(a2)
+	li	a1, 1
+	sw	a1, 8(a2)
+	sw	zero, 12(a2)
+	sd	zero, 24(a5)
+	bnez	s8, .LBB2_9
+.LBB2_12:                               # %_halide_buffer_is_bounds_query.exit31
+	ld	a1, 0(a0)
+	bnez	a1, .LBB2_17
+# %bb.13:                               # %then_bb5
+	srai	a1, a6, 31
+	subw	a2, a6, a1
+	mul	s1, a2, t1
+	srli	s1, s1, 32
+	add	a2, a2, s1
+	srliw	s1, a2, 31
+	sraiw	a2, a2, 8
+	addw	a2, a2, s1
+	add	a1, a1, a2
+	sgtz	a2, a1
+	negw	a2, a2
+	and	a1, a1, a2
+	mulw	a1, a1, a7
+	subw	a1, a1, t0
+	add	t0, t0, s3
+	addiw	t0, t0, -480
+	addiw	a6, a1, 960
+	add	a7, t4, s6
+	bgtz	a4, .LBB2_15
+# %bb.14:                               # %then_bb5
+	li	s0, 270
+	j	.LBB2_16
+.LBB2_15:
+	addi	s0, t3, 270
+.LBB2_16:                               # %then_bb5
+	subw	a1, s0, t4
+	sd	zero, 16(a0)
+	sd	zero, 8(a0)
+	sd	zero, 0(a0)
+	lui	a2, 256
+	addiw	a2, a2, 9
+	slli	a2, a2, 13
+	addi	a2, a2, 2
+	sd	a2, 32(a0)
+	sw	t0, 0(t5)
+	sw	a6, 4(t5)
+	li	a2, 1
+	sw	a2, 8(t5)
+	sw	zero, 12(t5)
+	sw	a7, 16(t5)
+	sw	a1, 20(t5)
+	sw	a6, 24(t5)
+	sw	zero, 28(t5)
+	sd	zero, 24(a0)
+.LBB2_17:                               # %land.rhs.i44
+	ld	a0, 0(a0)
+	seqz	a0, a0
+.LBB2_18:                               # %_halide_buffer_is_bounds_query.exit45
+	ld	a1, 16(a3)
+	ld	a2, 0(a3)
+	ld	a3, 16(a5)
+	ld	a5, 0(a5)
+	or	a1, a1, a2
+	seqz	a1, a1
+	or	a3, a3, a5
+	seqz	a2, a3
+	or	a1, a1, a2
+	or	a0, a0, a1
+	bnez	a0, .LBB2_20
+# %bb.19:                               # %then_bb8
+	mulw	a0, s6, s5
+	negw	s0, a0
+	sd	s8, 8(sp)
+	sw	a4, 16(sp)
+	sw	t2, 20(sp)
+	sw	s6, 24(sp)
+	sw	s5, 28(sp)
+	sw	s9, 32(sp)
+	sw	s0, 36(sp)
+.Lpcrel_hi2:
+	auipc	a0, %got_pcrel_hi(idw_halide_parallel__par_for_idw_halide_parallel__s0_x_x_outer_tile_index__1)
+	ld	a1, %pcrel_lo(.Lpcrel_hi2)(a0)
+	addi	a4, sp, 8
+	li	a0, 0
+	li	a2, 0
+	mv	a3, s7
+	call	halide_do_par_for@plt
+	sd	s2, 40(sp)
+	sd	s4, 48(sp)
+	sd	s8, 56(sp)
+	sw	s3, 64(sp)
+	sw	s6, 68(sp)
+	sw	s5, 72(sp)
+	sw	s9, 76(sp)
+	sw	s0, 80(sp)
+.Lpcrel_hi3:
+	auipc	a0, %got_pcrel_hi(idw_halide_parallel__par_for_idw_halide_parallel__s1_x_x_outer_tile_index__1)
+	ld	a1, %pcrel_lo(.Lpcrel_hi3)(a0)
+	addi	a4, sp, 40
+	li	a0, 0
+	li	a2, 0
+	mv	a3, s7
+	call	halide_do_par_for@plt
+.LBB2_20:                               # %destructor_block
+	li	a0, 0
+	ld	ra, 168(sp)                     # 8-byte Folded Reload
+	ld	s0, 160(sp)                     # 8-byte Folded Reload
+	ld	s1, 152(sp)                     # 8-byte Folded Reload
+	ld	s2, 144(sp)                     # 8-byte Folded Reload
+	ld	s3, 136(sp)                     # 8-byte Folded Reload
+	ld	s4, 128(sp)                     # 8-byte Folded Reload
+	ld	s5, 120(sp)                     # 8-byte Folded Reload
+	ld	s6, 112(sp)                     # 8-byte Folded Reload
+	ld	s7, 104(sp)                     # 8-byte Folded Reload
+	ld	s8, 96(sp)                      # 8-byte Folded Reload
+	ld	s9, 88(sp)                      # 8-byte Folded Reload
+	addi	sp, sp, 176
+	ret
+.Lfunc_end2:
+	.size	idw_halide_parallel_, .Lfunc_end2-idw_halide_parallel_
+                                        # -- End function
+	.section	.text.idw_halide_parallel__argv,"ax",@progbits
+	.globl	idw_halide_parallel__argv       # -- Begin function idw_halide_parallel__argv
+	.p2align	1
+	.type	idw_halide_parallel__argv,@function
+idw_halide_parallel__argv:              # @idw_halide_parallel__argv
+# %bb.0:                                # %entry
+	addi	sp, sp, -16
+	sd	ra, 8(sp)                       # 8-byte Folded Spill
+	ld	a0, 0(a0)
+	call	idw_halide_parallel_@plt
+	li	a0, 0
+	ld	ra, 8(sp)                       # 8-byte Folded Reload
+	addi	sp, sp, 16
+	ret
+.Lfunc_end3:
+	.size	idw_halide_parallel__argv, .Lfunc_end3-idw_halide_parallel__argv
+                                        # -- End function
+	.section	.text.idw_halide_parallel__metadata,"ax",@progbits
+	.globl	idw_halide_parallel__metadata   # -- Begin function idw_halide_parallel__metadata
+	.p2align	1
+	.type	idw_halide_parallel__metadata,@function
+idw_halide_parallel__metadata:          # @idw_halide_parallel__metadata
+# %bb.0:                                # %entry
+.Lpcrel_hi4:
+	auipc	a0, %pcrel_hi(.Lidw_halide_parallel__metadata_storage)
+	addi	a0, a0, %pcrel_lo(.Lpcrel_hi4)
+	ret
+.Lfunc_end4:
+	.size	idw_halide_parallel__metadata, .Lfunc_end4-idw_halide_parallel__metadata
+                                        # -- End function
+	.type	.Lb24.shape,@object             # @b24.shape
+	.section	.rodata,"a",@progbits
+	.p2align	5, 0x0
+.Lb24.shape:
+	.asciz	"\000\000\000\000,\001\000\000\001\000\000\000\000\000\000"
+	.size	.Lb24.shape, 16
+
+	.type	.Lb24.data,@object              # @b24.data
+	.p2align	5, 0x0
+.Lb24.data:
+	.asciz	"\000\000\000\000\000\000\000\000H\000\000\000\000\000\000\000\325\000\000\000O\000\000\000\000\000\000\000\252\001\000\000<\000\000\000\000\000\000\000\200\002\000\000L\000\000\000\000\000\000\000U\003\000\000\200\000\000\000\000\000\000\000*\004\000\000C\000\000\000\000\000\000\000\000\005\000\000A\000\000\000\000\000\000\000\325\005\000\000@\000\000\000\000\000\000\000\252\006\000\000<\000\000\000\000\000\000\000\200\007\000\000=\000\000\000x\000\000\000\000\000\000\000Q\000\000\000x\000\000\000\325\000\000\000O\000\000\000x\000\000\000\252\001\000\000:\000\000\000x\000\000\000\200\002\000\000\204\000\000\000x\000\000\000U\003\000\000\225\000\000\000x\000\000\000*\004\000\000\216\000\000\000x\000\000\000\000\005\000\000@\000\000\000x\000\000\000\325\005\000\000E\000\000\000x\000\000\000\252\006\000\000A\000\000\000x\000\000\000\200\007\000\000@\000\000\000\360\000\000\000\000\000\000\000K\000\000\000\360\000\000\000\325\000\000\000D\000\000\000\360\000\000\000\252\001\000\000\214\000\000\000\360\000\000\000\200\002\000\000\231\000\000\000\360\000\000\000U\003\000\000\221\000\000\000\360\000\000\000*\004\000\000\204\000\000\000\360\000\000\000\000\005\000\000\230\000\000\000\360\000\000\000\325\005\000\000}\000\000\000\360\000\000\000\252\006\000\000B\000\000\000\360\000\000\000\200\007\000\000:\000\000\000h\001\000\000\000\000\000\000N\000\000\000h\001\000\000\325\000\000\000<\000\000\000h\001\000\000\252\001\000\000\213\000\000\000h\001\000\000\200\002\000\000\250\000\000\000h\001\000\000U\003\000\000\232\000\000\000h\001\000\000*\004\000\000\212\000\000\000h\001\000\000\000\005\000\000\221\000\000\000h\001\000\000\325\005\000\000\240\000\000\000h\001\000\000\252\006\000\000D\000\000\000h\001\000\000\200\007\000\000<\000\000\000\340\001\000\000\000\000\000\000M\000\000\000\340\001\000\000\325\000\000\000;\000\000\000\340\001\000\000\252\001\000\000\245\000\000\000\340\001\000\000\200\002\000\000\267\000\000\000\340\001\000\000U\003\000\000\246\000\000\000\340\001\000\000*\004\000\000\216\000\000\000\340\001\000\000\000\005\000\000{\000\000\000\340\001\000\000\325\005\000\000\233\000\000\000\340\001\000\000\252\006\000\000\220\000\000\000\340\001\000\000\200\007\000\000<\000\000\000X\002\000\000\000\000\000\000S\000\000\000X\002\000\000\325\000\000\000A\000\000\000X\002\000\000\252\001\000\000\262\000\000\000X\002\000\000\200\002\000\000\270\000\000\000X\002\000\000U\003\000\000\212\000\000\000X\002\000\000*\004\000\000|\000\000\000X\002\000\000\000\005\000\000\204\000\000\000X\002\000\000\325\005\000\000\257\000\000\000X\002\000\000\252\006\000\000\257\000\000\000X\002\000\000\200\007\000\000<\000\000\000\320\002\000\000\000\000\000\000V\000\000\000\320\002\000\000\325\000\000\000<\000\000\000\320\002\000\000\252\001\000\000\266\000\000\000\320\002\000\000\200\002\000\000\263\000\000\000\320\002\000\000U\003\000\000\230\000\000\000\320\002\000\000*\004\000\000\214\000\000\000\320\002\000\000\000\005\000\000s\000\000\000\320\002\000\000\325\005\000\000\234\000\000\000\320\002\000\000\252\006\000\000\254\000\000\000\320\002\000\000\200\007\000\000A\000\000\000H\003\000\000\000\000\000\000Z\000\000\000H\003\000\000\325\000\000\000E\000\000\000H\003\000\000\252\001\000\000\244\000\000\000H\003\000\000\200\002\000\000\272\000\000\000H\003\000\000U\003\000\000\222\000\000\000H\003\000\000*\004\000\000\223\000\000\000H\003\000\000\000\005\000\000z\000\000\000H\003\000\000\325\005\000\000\230\000\000\000H\003\000\000\252\006\000\000\246\000\000\000H\003\000\000\200\007\000\000E\000\000\000\300\003\000\000\000\000\000\000U\000\000\000\300\003\000\000\325\000\000\000C\000\000\000\300\003\000\000\252\001\000\000\272\000\000\000\300\003\000\000\200\002\000\000\251\000\000\000\300\003\000\000U\003\000\000\242\000\000\000\300\003\000\000*\004\000\000\223\000\000\000\300\003\000\000\000\005\000\000|\000\000\000\300\003\000\000\325\005\000\000\225\000\000\000\300\003\000\000\252\006\000\000\235\000\000\000\300\003\000\000\200\007\000\000E\000\000\0008\004\000\000\000\000\000\000U\000\000\0008\004\000\000\325\000\000\000F\000\000\0008\004\000\000\252\001\000\000\301\000\000\0008\004\000\000\200\002\000\000\300\000\000\0008\004\000\000U\003\000\000\233\000\000\0008\004\000\000*\004\000\000\207\000\000\0008\004\000\000\000\005\000\000x\000\000\0008\004\000\000\325\005\000\000\212\000\000\0008\004\000\000\252\006\000\000\200\000\000\0008\004\000\000\200\007\000\000F\000\000"
+	.size	.Lb24.data, 1200
+
+	.type	.Lb24.buffer,@object            # @b24.buffer
+	.data
+	.p2align	4, 0x0
+.Lb24.buffer:
+	.quad	0                               # 0x0
+	.quad	0
+	.quad	.Lb24.data
+	.quad	1                               # 0x1
+	.byte	0                               # 0x0
+	.byte	32                              # 0x20
+	.half	1                               # 0x1
+	.word	1                               # 0x1
+	.quad	.Lb24.shape
+	.quad	0
+	.size	.Lb24.buffer, 56
+
+	.type	.Lb25.shape,@object             # @b25.shape
+	.section	.rodata,"a",@progbits
+	.p2align	5, 0x0
+.Lb25.shape:
+	.asciz	"\000\000\000\000d\000\000\000\001\000\000\000\000\000\000"
+	.size	.Lb25.shape, 16
+
+	.type	.Lb25.data,@object              # @b25.data
+	.p2align	5, 0x0
+.Lb25.data:
+	.ascii	"\211\004&=\244\022\233\275\217U%;\272)J>#\026\024\276@_\234>\340\243|\275\334\034K\275\232\300\004:A\313\245\274\303\024r\275\345\313a\275x\231\300>\201\334\"\276\236}\321\275H\346\257\276\202V\310>\225\375(>},\026\274E\235W\275\241\236\236<\245\220\222=\004\361\263\276\2620\240\274\265\036\240=\\:\336=\240\016\263\276\260\224\371\275p\235(=\311\\x;\226y\\\275\360\227\377=\275\364\200=rw/<@U\017=c\327\003=nV\267\273\354<Y\276\257$\260>3\207\311\274<&{\274h\b\031>\203\222\003\276\213\3649\275d\177\017\276\330\306\246\275'\314:>\203\024\253=S\331M\276\257\263S=b\221q\275\336\t\306=#^\005\2763)\256\274\344\255G>\224!\374=\374\360o\275\215#\031\276\322`J\276y6\360=\273\303;\275N\"N>M\022K\276\016\035C=1Fb\275u\017\220\275\0040\033>,\037\223=t\335\304\275\363\344\271=\320\341\260\275\2265\255=\310\"b=&\031\330\275;\314\000>>\347o\275\255\222\020=C\rU<\370d\307\275\373\306\206=I\345\373\274Y|\034>WY;\276O\240?>B\347\226\275\203\271\203\275W\243\313<\214\364\343\274Ey\033\276J4\214=\261\304\241\274-\271p>X\313l\276*_\366\275\260^\\=\237l1=\371\272\200=\002\373?\274*\3578<]Gv="
+	.size	.Lb25.data, 400
+
+	.type	.Lb25.buffer,@object            # @b25.buffer
+	.data
+	.p2align	4, 0x0
+.Lb25.buffer:
+	.quad	0                               # 0x0
+	.quad	0
+	.quad	.Lb25.data
+	.quad	1                               # 0x1
+	.byte	2                               # 0x2
+	.byte	32                              # 0x20
+	.half	1                               # 0x1
+	.word	1                               # 0x1
+	.quad	.Lb25.shape
+	.quad	0
+	.size	.Lb25.buffer, 56
+
+	.type	.L__unnamed_1,@object           # @0
+	.section	.rodata,"a",@progbits
+	.p2align	4, 0x0
+.L__unnamed_1:
+	.zero	32
+	.size	.L__unnamed_1, 32
+
+	.type	.Lstr,@object                   # @str
+	.p2align	5, 0x0
+.Lstr:
+	.asciz	"idw_halide_parallel_"
+	.size	.Lstr, 21
+
+	.type	.L__unnamed_2,@object           # @1
+	.section	.data.rel.ro,"aw",@progbits
+	.p2align	4, 0x0
+.L__unnamed_2:
+	.quad	.Lstr
+	.word	2                               # 0x2
+	.word	2                               # 0x2
+	.byte	2                               # 0x2
+	.byte	32                              # 0x20
+	.half	1                               # 0x1
+	.zero	4
+	.quad	0
+	.quad	0
+	.quad	0
+	.quad	0
+	.quad	.L__unnamed_1
+	.size	.L__unnamed_2, 64
+
+	.type	.Lstr.4,@object                 # @str.4
+	.section	.rodata,"a",@progbits
+	.p2align	5, 0x0
+.Lstr.4:
+	.asciz	"riscv-64-linux-no_asserts-no_runtime-rvv-vector_bits_128"
+	.size	.Lstr.4, 57
+
+	.type	.Lidw_halide_parallel__metadata_storage,@object # @idw_halide_parallel__metadata_storage
+	.section	.data.rel.ro,"aw",@progbits
+	.p2align	4, 0x0
+.Lidw_halide_parallel__metadata_storage:
+	.word	1                               # 0x1
+	.word	1                               # 0x1
+	.quad	.L__unnamed_2
+	.quad	.Lstr.4
+	.quad	.Lstr
+	.size	.Lidw_halide_parallel__metadata_storage, 32
+
+	.ident	"clang version 16.0.0 (https://github.com/dkurt/llvm-rvv-071 b027aa1b59c9f53240bdc836f39656723fdf9df0)"
+	.ident	"clang version 16.0.0 (https://github.com/dkurt/llvm-rvv-071 b027aa1b59c9f53240bdc836f39656723fdf9df0)"
+	.ident	"clang version 16.0.0 (https://github.com/dkurt/llvm-rvv-071 b027aa1b59c9f53240bdc836f39656723fdf9df0)"
+	.ident	"clang version 16.0.0 (https://github.com/dkurt/llvm-rvv-071 b027aa1b59c9f53240bdc836f39656723fdf9df0)"
+	.ident	"clang version 16.0.0 (https://github.com/dkurt/llvm-rvv-071 b027aa1b59c9f53240bdc836f39656723fdf9df0)"
+	.ident	"clang version 16.0.0 (https://github.com/dkurt/llvm-rvv-071 b027aa1b59c9f53240bdc836f39656723fdf9df0)"
+	.section	".note.GNU-stack","",@progbits
diff --git a/aot/idw_halide_parallel_vec.h b/aot/idw_halide_parallel_vec.h
new file mode 100644
index 0000000..156229d
--- /dev/null
+++ b/aot/idw_halide_parallel_vec.h
@@ -0,0 +1,56 @@
+#ifndef HALIDE__idw_halide_parallel_vec_h
+#define HALIDE__idw_halide_parallel_vec_h
+#include <stdint.h>
+
+// Forward declarations of the types used in the interface
+// to the Halide pipeline.
+//
+// For the definitions of these structs, include HalideRuntime.h
+
+// Halide's representation of a multi-dimensional array.
+// Halide::Runtime::Buffer is a more user-friendly wrapper
+// around this. Its declaration is in HalideBuffer.h
+struct halide_buffer_t;
+
+// Metadata describing the arguments to the generated function.
+// Used to construct calls to the _argv version of the function.
+struct halide_filter_metadata_t;
+
+#ifndef HALIDE_MUST_USE_RESULT
+#ifdef __has_attribute
+#if __has_attribute(nodiscard)
+#define HALIDE_MUST_USE_RESULT [[nodiscard]]
+#elif __has_attribute(warn_unused_result)
+#define HALIDE_MUST_USE_RESULT __attribute__((warn_unused_result))
+#else
+#define HALIDE_MUST_USE_RESULT
+#endif
+#else
+#define HALIDE_MUST_USE_RESULT
+#endif
+#endif
+
+#ifndef HALIDE_FUNCTION_ATTRS
+#define HALIDE_FUNCTION_ATTRS
+#endif
+
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+HALIDE_FUNCTION_ATTRS
+int idw_halide_parallel_vec_(struct halide_buffer_t *_idw_halide_parallel_vec__buffer);
+
+HALIDE_FUNCTION_ATTRS
+int idw_halide_parallel_vec__argv(void **args);
+
+HALIDE_FUNCTION_ATTRS
+const struct halide_filter_metadata_t *idw_halide_parallel_vec__metadata();
+
+#ifdef __cplusplus
+}  // extern "C"
+#endif
+
+#endif
diff --git a/aot/idw_halide_parallel_vec.s b/aot/idw_halide_parallel_vec.s
new file mode 100644
index 0000000..484d582
--- /dev/null
+++ b/aot/idw_halide_parallel_vec.s
@@ -0,0 +1,739 @@
+	.text
+	.attribute	4, 16
+	.attribute	5, "rv64gcv0p7"
+	.file	"halide_buffer_t.cpp"
+	.section	.text.idw_halide_parallel_vec__par_for_idw_halide_parallel_vec__s0_x_x_outer_tile_index__1,"ax",@progbits
+	.globl	idw_halide_parallel_vec__par_for_idw_halide_parallel_vec__s0_x_x_outer_tile_index__1 # -- Begin function idw_halide_parallel_vec__par_for_idw_halide_parallel_vec__s0_x_x_outer_tile_index__1
+	.p2align	1
+	.type	idw_halide_parallel_vec__par_for_idw_halide_parallel_vec__s0_x_x_outer_tile_index__1,@function
+idw_halide_parallel_vec__par_for_idw_halide_parallel_vec__s0_x_x_outer_tile_index__1: # @idw_halide_parallel_vec__par_for_idw_halide_parallel_vec__s0_x_x_outer_tile_index__1
+# %bb.0:                                # %entry
+	addi	sp, sp, -48
+	sd	ra, 40(sp)                      # 8-byte Folded Spill
+	sd	s0, 32(sp)                      # 8-byte Folded Spill
+	sd	s1, 24(sp)                      # 8-byte Folded Spill
+	sd	s2, 16(sp)                      # 8-byte Folded Spill
+	sd	s3, 8(sp)                       # 8-byte Folded Spill
+	lw	a4, 24(a2)
+	lw	a7, 8(a2)
+	lw	a6, 12(a2)
+	sraiw	a3, a1, 31
+	seqz	a5, a4
+	negw	a0, a5
+	srai	s1, a4, 31
+	subw	t0, a1, a3
+	or	a0, a0, a4
+	remw	a1, t0, a0
+	xor	s0, s1, a4
+	not	a0, s1
+	add	s0, s0, a0
+	and	s0, s0, a3
+	add	a1, a1, s0
+	li	s0, 480
+	mulw	a1, a1, s0
+	addi	s0, a5, -1
+	and	t2, s0, a1
+	addiw	a1, a7, -480
+	blt	t2, a1, .LBB0_2
+# %bb.1:                                # %entry
+	mv	t2, a1
+.LBB0_2:                                # %entry
+	lw	a7, 16(a2)
+	lw	s2, 20(a2)
+	lw	t1, 28(a2)
+	add	a4, a4, a5
+	divw	a1, t0, a4
+	subw	a0, a0, s1
+	and	a0, a0, a3
+	add	a0, a0, a1
+	li	s1, 270
+	mulw	a0, a0, s1
+	and	a0, a0, s0
+	addiw	a1, a6, -270
+	blt	a0, a1, .LBB0_4
+# %bb.3:                                # %entry
+	mv	a0, a1
+.LBB0_4:                                # %entry
+	ld	s3, 0(a2)
+	add	a0, a0, a7
+	add	t1, t1, t2
+	mulw	a0, s2, a0
+	addw	s0, t1, a0
+.LBB0_5:                                # %"for idw_halide_parallel_vec_.s0.y.y_inner"
+                                        # =>This Inner Loop Header: Depth=1
+	slli	a0, s0, 2
+	add	a0, a0, s3
+	li	a2, 1920
+	li	a1, 0
+	call	memset@plt
+	addi	s1, s1, -1
+	addw	s0, s0, s2
+	bnez	s1, .LBB0_5
+# %bb.6:                                # %destructor_block
+	li	a0, 0
+	ld	ra, 40(sp)                      # 8-byte Folded Reload
+	ld	s0, 32(sp)                      # 8-byte Folded Reload
+	ld	s1, 24(sp)                      # 8-byte Folded Reload
+	ld	s2, 16(sp)                      # 8-byte Folded Reload
+	ld	s3, 8(sp)                       # 8-byte Folded Reload
+	addi	sp, sp, 48
+	ret
+.Lfunc_end0:
+	.size	idw_halide_parallel_vec__par_for_idw_halide_parallel_vec__s0_x_x_outer_tile_index__1, .Lfunc_end0-idw_halide_parallel_vec__par_for_idw_halide_parallel_vec__s0_x_x_outer_tile_index__1
+                                        # -- End function
+	.section	.text.idw_halide_parallel_vec__par_for_idw_halide_parallel_vec__s1_x_x_outer_tile_index__1,"ax",@progbits
+	.globl	idw_halide_parallel_vec__par_for_idw_halide_parallel_vec__s1_x_x_outer_tile_index__1 # -- Begin function idw_halide_parallel_vec__par_for_idw_halide_parallel_vec__s1_x_x_outer_tile_index__1
+	.p2align	1
+	.type	idw_halide_parallel_vec__par_for_idw_halide_parallel_vec__s1_x_x_outer_tile_index__1,@function
+idw_halide_parallel_vec__par_for_idw_halide_parallel_vec__s1_x_x_outer_tile_index__1: # @idw_halide_parallel_vec__par_for_idw_halide_parallel_vec__s1_x_x_outer_tile_index__1
+# %bb.0:                                # %entry
+	addi	sp, sp, -208
+	sd	ra, 200(sp)                     # 8-byte Folded Spill
+	sd	s0, 192(sp)                     # 8-byte Folded Spill
+	sd	s1, 184(sp)                     # 8-byte Folded Spill
+	sd	s2, 176(sp)                     # 8-byte Folded Spill
+	sd	s3, 168(sp)                     # 8-byte Folded Spill
+	sd	s4, 160(sp)                     # 8-byte Folded Spill
+	sd	s5, 152(sp)                     # 8-byte Folded Spill
+	sd	s6, 144(sp)                     # 8-byte Folded Spill
+	sd	s7, 136(sp)                     # 8-byte Folded Spill
+	sd	s8, 128(sp)                     # 8-byte Folded Spill
+	sd	s9, 120(sp)                     # 8-byte Folded Spill
+	sd	s10, 112(sp)                    # 8-byte Folded Spill
+	sd	s11, 104(sp)                    # 8-byte Folded Spill
+	li	a3, 0
+	ld	t4, 0(a2)
+	ld	t0, 8(a2)
+	ld	a0, 16(a2)
+	sd	a0, 48(sp)                      # 8-byte Folded Spill
+	lw	a0, 24(a2)
+	sd	a0, 40(sp)                      # 8-byte Folded Spill
+	lw	a7, 28(a2)
+	lw	a4, 36(a2)
+	lw	a0, 32(a2)
+	sd	a0, 8(sp)                       # 8-byte Folded Spill
+	lw	a0, 40(a2)
+	sd	a0, 32(sp)                      # 8-byte Folded Spill
+	sraiw	a2, a1, 31
+	seqz	a5, a4
+	negw	s1, a5
+	srai	s0, a4, 31
+	subw	a1, a1, a2
+	or	s1, s1, a4
+	remw	t1, a1, s1
+	xor	a0, s0, a4
+	not	s1, s0
+	add	a0, a0, s1
+	and	a0, a0, a2
+	add	a0, a0, t1
+	add	a4, a4, a5
+	divw	a1, a1, a4
+	subw	s1, s1, s0
+	and	a2, a2, s1
+	add	a1, a1, a2
+	li	a2, 480
+	mulw	a0, a0, a2
+	addi	a5, a5, -1
+	and	a0, a0, a5
+	sd	a0, 24(sp)                      # 8-byte Folded Spill
+	li	a0, 270
+	mulw	a0, a1, a0
+	and	a0, a0, a5
+	add	a0, a0, a7
+	sd	a0, 0(sp)                       # 8-byte Folded Spill
+	li	s2, 48
+	li	a0, 6
+	li	a2, 1
+	addi	t5, sp, 80
+	addi	s3, sp, 92
+	addi	s4, sp, 88
+	addi	s5, sp, 84
+	li	s0, 4
+	addi	t6, sp, 64
+	addi	s6, sp, 68
+	addi	s7, sp, 76
+	addi	s8, sp, 72
+	li	s9, 2
+	li	s10, 3
+	li	ra, 25
+.LBB1_1:                                # %"for idw_halide_parallel_vec_.s1.y.y_inner"
+                                        # =>This Loop Header: Depth=1
+                                        #     Child Loop BB1_2 Depth 2
+                                        #       Child Loop BB1_3 Depth 3
+                                        #         Child Loop BB1_4 Depth 4
+                                        #           Child Loop BB1_9 Depth 5
+	li	s11, 0
+	sd	a3, 16(sp)                      # 8-byte Folded Spill
+	ld	a1, 0(sp)                       # 8-byte Folded Reload
+	addw	t1, a1, a3
+	ld	a1, 8(sp)                       # 8-byte Folded Reload
+	mulw	a1, t1, a1
+	sd	a1, 56(sp)                      # 8-byte Folded Spill
+.LBB1_2:                                # %"for idw_halide_parallel_vec_.s1.x.x_inner"
+                                        #   Parent Loop BB1_1 Depth=1
+                                        # =>  This Loop Header: Depth=2
+                                        #       Child Loop BB1_3 Depth 3
+                                        #         Child Loop BB1_4 Depth 4
+                                        #           Child Loop BB1_9 Depth 5
+	li	a7, 0
+	ld	t2, 24(sp)                      # 8-byte Folded Reload
+	add	t2, t2, s11
+	ld	a4, 32(sp)                      # 8-byte Folded Reload
+	add	a4, a4, t2
+	ld	a1, 56(sp)                      # 8-byte Folded Reload
+	addw	a4, a4, a1
+	ld	a1, 40(sp)                      # 8-byte Folded Reload
+	addw	t2, t2, a1
+	slli	a4, a4, 2
+	ld	t3, 48(sp)                      # 8-byte Folded Reload
+	add	t3, t3, a4
+.LBB1_3:                                # %"for idw_halide_parallel_vec_.s1.r52$x.r52$x"
+                                        #   Parent Loop BB1_1 Depth=1
+                                        #     Parent Loop BB1_2 Depth=2
+                                        # =>    This Loop Header: Depth=3
+                                        #         Child Loop BB1_4 Depth 4
+                                        #           Child Loop BB1_9 Depth 5
+	mul	a4, a7, s2
+	add	a4, a4, t4
+	addi	a6, a4, 4
+	vsetvli	zero, a0, e32, m2
+	vlwu.v	v8, (a6)
+	vsetvli	zero, a2, e32, m2
+	vslidedown.vi	v10, v8, 3
+	addi	a3, a4, 20
+	vsetvli	zero, a0, e32, m2
+	vlwu.v	v12, (a3)
+	vsetvli	zero, a2, e32, m2
+	vslidedown.vi	v14, v12, 2
+	vslidedown.vi	v12, v12, 5
+	vsw.v	v8, (t5)
+	vsw.v	v12, (s3)
+	vsw.v	v14, (s4)
+	vsw.v	v10, (s5)
+	vsetvli	zero, s0, e32, m1
+	vlwu.v	v8, (t5)
+	vsetvli	a3, zero, e32, m1
+	vrsub.vx	v8, v8, t2
+	vsetvli	zero, a0, e32, m2
+	vlwu.v	v10, (a4)
+	vsetvli	zero, a2, e32, m2
+	vslidedown.vi	v12, v10, 3
+	addi	a4, a4, 16
+	vsetvli	zero, a0, e32, m2
+	vlwu.v	v14, (a4)
+	vsetvli	zero, a2, e32, m2
+	vslidedown.vi	v16, v14, 2
+	vslidedown.vi	v14, v14, 5
+	vsw.v	v10, (t6)
+	vsw.v	v12, (s6)
+	vsw.v	v14, (s7)
+	vsw.v	v16, (s8)
+	vsetvli	zero, s0, e32, m1
+	vlwu.v	v9, (t6)
+	vsetvli	a3, zero, e32, m1
+	lw	a3, 0(t3)
+	vrsub.vx	v9, v9, t1
+	fmv.w.x	ft0, a3
+	vmul.vv	v9, v9, v9
+	vmacc.vv	v9, v8, v8
+	vfcvt.f.x.v	v9, v9
+	vfmv.f.s	ft1, v9
+	fsqrt.s	ft1, ft1
+	vfmv.s.f	v8, ft1
+	vsetvli	zero, a2, e32, m1
+	vslidedown.vi	v10, v9, 1
+	vfmv.f.s	ft1, v10
+	fsqrt.s	ft1, ft1
+	vsetvli	a3, zero, e32, m1
+	vfmv.s.f	v10, ft1
+	vsetvli	zero, s9, e32, m1
+	vslideup.vi	v8, v10, 1
+	vsetvli	zero, a2, e32, m1
+	vslidedown.vi	v10, v9, 2
+	vfmv.f.s	ft1, v10
+	fsqrt.s	ft1, ft1
+	vsetvli	a3, zero, e32, m1
+	vfmv.s.f	v10, ft1
+	vsetvli	zero, s10, e32, m1
+	vslideup.vi	v8, v10, 2
+	vsetvli	zero, a2, e32, m1
+	vslidedown.vi	v9, v9, 3
+	vfmv.f.s	ft1, v9
+	fsqrt.s	ft1, ft1
+	vsetvli	a3, zero, e32, m1
+	vfmv.s.f	v9, ft1
+	vsetvli	zero, s0, e32, m1
+	vslideup.vi	v8, v9, 3
+	slli	a4, a7, 4
+	add	a4, a4, t0
+.LBB1_4:                                # %casloop.start
+                                        #   Parent Loop BB1_1 Depth=1
+                                        #     Parent Loop BB1_2 Depth=2
+                                        #       Parent Loop BB1_3 Depth=3
+                                        # =>      This Loop Header: Depth=4
+                                        #           Child Loop BB1_9 Depth 5
+	flw	ft1, 0(t3)
+	vsetvli	zero, zero, e32, m1
+	vlwu.v	v9, (a4)
+	vfmul.vv	v9, v8, v9
+	vfmv.s.f	v10, ft1
+	vfredsum.vs	v9, v9, v10
+	vfmv.f.s	ft1, v9
+	fmv.x.w	a3, ft1
+	fmv.x.w	a1, ft0
+.LBB1_9:                                # %casloop.start
+                                        #   Parent Loop BB1_1 Depth=1
+                                        #     Parent Loop BB1_2 Depth=2
+                                        #       Parent Loop BB1_3 Depth=3
+                                        #         Parent Loop BB1_4 Depth=4
+                                        # =>        This Inner Loop Header: Depth=5
+	lr.w	s1, (t3)
+	bne	s1, a1, .LBB1_11
+# %bb.10:                               # %casloop.start
+                                        #   in Loop: Header=BB1_9 Depth=5
+	sc.w	a5, a3, (t3)
+	bnez	a5, .LBB1_9
+.LBB1_11:                               # %casloop.start
+                                        #   in Loop: Header=BB1_4 Depth=4
+	fmv.w.x	ft0, s1
+	bne	s1, a1, .LBB1_4
+# %bb.5:                                # %casloop.end
+                                        #   in Loop: Header=BB1_3 Depth=3
+	addi	a7, a7, 1
+	bne	a7, ra, .LBB1_3
+# %bb.6:                                # %"end for idw_halide_parallel_vec_.s1.r52$x.r52$x"
+                                        #   in Loop: Header=BB1_2 Depth=2
+	addi	s11, s11, 1
+	li	a1, 480
+	bne	s11, a1, .LBB1_2
+# %bb.7:                                # %"end for idw_halide_parallel_vec_.s1.x.x_inner"
+                                        #   in Loop: Header=BB1_1 Depth=1
+	ld	a3, 16(sp)                      # 8-byte Folded Reload
+	addiw	a3, a3, 1
+	li	a1, 270
+	bne	a3, a1, .LBB1_1
+# %bb.8:                                # %destructor_block
+	li	a0, 0
+	ld	ra, 200(sp)                     # 8-byte Folded Reload
+	ld	s0, 192(sp)                     # 8-byte Folded Reload
+	ld	s1, 184(sp)                     # 8-byte Folded Reload
+	ld	s2, 176(sp)                     # 8-byte Folded Reload
+	ld	s3, 168(sp)                     # 8-byte Folded Reload
+	ld	s4, 160(sp)                     # 8-byte Folded Reload
+	ld	s5, 152(sp)                     # 8-byte Folded Reload
+	ld	s6, 144(sp)                     # 8-byte Folded Reload
+	ld	s7, 136(sp)                     # 8-byte Folded Reload
+	ld	s8, 128(sp)                     # 8-byte Folded Reload
+	ld	s9, 120(sp)                     # 8-byte Folded Reload
+	ld	s10, 112(sp)                    # 8-byte Folded Reload
+	ld	s11, 104(sp)                    # 8-byte Folded Reload
+	addi	sp, sp, 208
+	ret
+.Lfunc_end1:
+	.size	idw_halide_parallel_vec__par_for_idw_halide_parallel_vec__s1_x_x_outer_tile_index__1, .Lfunc_end1-idw_halide_parallel_vec__par_for_idw_halide_parallel_vec__s1_x_x_outer_tile_index__1
+                                        # -- End function
+	.section	.text.idw_halide_parallel_vec_,"ax",@progbits
+	.globl	idw_halide_parallel_vec_        # -- Begin function idw_halide_parallel_vec_
+	.p2align	1
+	.type	idw_halide_parallel_vec_,@function
+idw_halide_parallel_vec_:               # @idw_halide_parallel_vec_
+# %bb.0:                                # %entry
+	addi	sp, sp, -176
+	sd	ra, 168(sp)                     # 8-byte Folded Spill
+	sd	s0, 160(sp)                     # 8-byte Folded Spill
+	sd	s1, 152(sp)                     # 8-byte Folded Spill
+	sd	s2, 144(sp)                     # 8-byte Folded Spill
+	sd	s3, 136(sp)                     # 8-byte Folded Spill
+	sd	s4, 128(sp)                     # 8-byte Folded Spill
+	sd	s5, 120(sp)                     # 8-byte Folded Spill
+	sd	s6, 112(sp)                     # 8-byte Folded Spill
+	sd	s7, 104(sp)                     # 8-byte Folded Spill
+	sd	s8, 96(sp)                      # 8-byte Folded Spill
+	sd	s9, 88(sp)                      # 8-byte Folded Spill
+	ld	t5, 40(a0)
+	lw	a4, 4(t5)
+	lw	t2, 20(t5)
+	li	a2, -480
+	subw	a6, a2, a4
+	addiw	a2, a4, -1
+	blt	a2, a6, .LBB2_2
+# %bb.1:                                # %entry
+	mv	a6, a2
+.LBB2_2:                                # %entry
+.Lpcrel_hi0:
+	auipc	a2, %pcrel_hi(.Lb34.buffer)
+	addi	a3, a2, %pcrel_lo(.Lpcrel_hi0)
+	sgtz	t4, a4
+	slti	a5, a4, -479
+	slti	s0, t2, -269
+	lui	s1, 559241
+	li	a7, 480
+	addiw	t1, s1, -1911
+	mv	t0, a4
+	blt	a4, a7, .LBB2_4
+# %bb.3:                                # %entry
+	li	t0, 480
+.LBB2_4:                                # %entry
+.Lpcrel_hi1:
+	auipc	t6, %pcrel_hi(.Lb35.buffer)
+	ld	s2, 16(a3)
+	add	a1, a4, a5
+	addiw	a1, a1, 479
+	mul	a2, a1, t1
+	srli	a2, a2, 32
+	add	a1, a1, a2
+	srliw	a2, a1, 31
+	sraiw	a1, a1, 8
+	add	a1, a1, a2
+	subw	s9, a1, a5
+	add	a1, t2, s0
+	addiw	a1, a1, 269
+	lui	a2, 994205
+	addiw	a2, a2, 1609
+	mul	a2, a1, a2
+	srli	a2, a2, 32
+	add	a1, a1, a2
+	srliw	a2, a1, 31
+	sraiw	a1, a1, 8
+	add	a1, a1, a2
+	subw	a1, a1, s0
+	mulw	s7, a1, s9
+	slti	a1, s7, 1
+	negw	a2, a1
+	seqz	s1, s9
+	add	a1, a1, s7
+	addiw	a1, a1, -1
+	add	s0, s9, s1
+	divw	a1, a1, s0
+	add	a5, a5, a5
+	addiw	a5, a5, -1
+	and	a2, a2, a5
+	add	a1, a1, a2
+	li	a2, 270
+	mulw	a1, a1, a2
+	addi	s1, s1, -1
+	and	t3, s1, a1
+	addi	t4, t4, -1
+	and	t4, t4, t3
+	addiw	a2, t2, -270
+	addi	a5, t6, %pcrel_lo(.Lpcrel_hi1)
+	blt	t4, a2, .LBB2_6
+# %bb.5:                                # %entry
+	mv	t4, a2
+.LBB2_6:                                # %entry
+	ld	s4, 16(a5)
+	lw	s3, 0(t5)
+	ld	a1, 0(a3)
+	lw	s6, 16(t5)
+	lw	s5, 24(t5)
+	or	a1, s2, a1
+	mv	s0, s4
+	beqz	a1, .LBB2_10
+# %bb.7:                                # %after_bb
+	ld	a1, 0(a5)
+	ld	s8, 16(a0)
+	or	a1, a1, s0
+	beqz	a1, .LBB2_11
+.LBB2_8:                                # %after_bb1
+	beqz	s8, .LBB2_12
+.LBB2_9:
+	li	a0, 0
+	j	.LBB2_18
+.LBB2_10:                               # %then_bb
+	lui	a1, 128
+	addiw	a1, a1, 9
+	slli	a1, a1, 13
+	sd	a1, 32(a3)
+	ld	a1, 40(a3)
+	sd	zero, 0(a3)
+	sd	zero, 8(a3)
+	sd	zero, 16(a3)
+	sw	zero, 0(a1)
+	li	a2, 300
+	sw	a2, 4(a1)
+	li	a2, 1
+	sw	a2, 8(a1)
+	sw	zero, 12(a1)
+	ld	s0, 16(a5)
+	sd	zero, 24(a3)
+	ld	a1, 0(a5)
+	ld	s8, 16(a0)
+	or	a1, a1, s0
+	bnez	a1, .LBB2_8
+.LBB2_11:                               # %then_bb2
+	sd	zero, 16(a5)
+	sd	zero, 8(a5)
+	sd	zero, 0(a5)
+	lui	a1, 128
+	addiw	a1, a1, 9
+	ld	a2, 40(a5)
+	slli	a1, a1, 13
+	addi	a1, a1, 2
+	sd	a1, 32(a5)
+	sw	zero, 0(a2)
+	li	a1, 100
+	sw	a1, 4(a2)
+	li	a1, 1
+	sw	a1, 8(a2)
+	sw	zero, 12(a2)
+	sd	zero, 24(a5)
+	bnez	s8, .LBB2_9
+.LBB2_12:                               # %_halide_buffer_is_bounds_query.exit31
+	ld	a1, 0(a0)
+	bnez	a1, .LBB2_17
+# %bb.13:                               # %then_bb5
+	srai	a1, a6, 31
+	subw	a2, a6, a1
+	mul	s1, a2, t1
+	srli	s1, s1, 32
+	add	a2, a2, s1
+	srliw	s1, a2, 31
+	sraiw	a2, a2, 8
+	addw	a2, a2, s1
+	add	a1, a1, a2
+	sgtz	a2, a1
+	negw	a2, a2
+	and	a1, a1, a2
+	mulw	a1, a1, a7
+	subw	a1, a1, t0
+	add	t0, t0, s3
+	addiw	t0, t0, -480
+	addiw	a6, a1, 960
+	add	a7, t4, s6
+	bgtz	a4, .LBB2_15
+# %bb.14:                               # %then_bb5
+	li	s0, 270
+	j	.LBB2_16
+.LBB2_15:
+	addi	s0, t3, 270
+.LBB2_16:                               # %then_bb5
+	subw	a1, s0, t4
+	sd	zero, 16(a0)
+	sd	zero, 8(a0)
+	sd	zero, 0(a0)
+	lui	a2, 256
+	addiw	a2, a2, 9
+	slli	a2, a2, 13
+	addi	a2, a2, 2
+	sd	a2, 32(a0)
+	sw	t0, 0(t5)
+	sw	a6, 4(t5)
+	li	a2, 1
+	sw	a2, 8(t5)
+	sw	zero, 12(t5)
+	sw	a7, 16(t5)
+	sw	a1, 20(t5)
+	sw	a6, 24(t5)
+	sw	zero, 28(t5)
+	sd	zero, 24(a0)
+.LBB2_17:                               # %land.rhs.i44
+	ld	a0, 0(a0)
+	seqz	a0, a0
+.LBB2_18:                               # %_halide_buffer_is_bounds_query.exit45
+	ld	a1, 16(a3)
+	ld	a2, 0(a3)
+	ld	a3, 16(a5)
+	ld	a5, 0(a5)
+	or	a1, a1, a2
+	seqz	a1, a1
+	or	a3, a3, a5
+	seqz	a2, a3
+	or	a1, a1, a2
+	or	a0, a0, a1
+	bnez	a0, .LBB2_20
+# %bb.19:                               # %then_bb8
+	mulw	a0, s6, s5
+	negw	s0, a0
+	sd	s8, 8(sp)
+	sw	a4, 16(sp)
+	sw	t2, 20(sp)
+	sw	s6, 24(sp)
+	sw	s5, 28(sp)
+	sw	s9, 32(sp)
+	sw	s0, 36(sp)
+.Lpcrel_hi2:
+	auipc	a0, %got_pcrel_hi(idw_halide_parallel_vec__par_for_idw_halide_parallel_vec__s0_x_x_outer_tile_index__1)
+	ld	a1, %pcrel_lo(.Lpcrel_hi2)(a0)
+	addi	a4, sp, 8
+	li	a0, 0
+	li	a2, 0
+	mv	a3, s7
+	call	halide_do_par_for@plt
+	sd	s2, 40(sp)
+	sd	s4, 48(sp)
+	sd	s8, 56(sp)
+	sw	s3, 64(sp)
+	sw	s6, 68(sp)
+	sw	s5, 72(sp)
+	sw	s9, 76(sp)
+	sw	s0, 80(sp)
+.Lpcrel_hi3:
+	auipc	a0, %got_pcrel_hi(idw_halide_parallel_vec__par_for_idw_halide_parallel_vec__s1_x_x_outer_tile_index__1)
+	ld	a1, %pcrel_lo(.Lpcrel_hi3)(a0)
+	addi	a4, sp, 40
+	li	a0, 0
+	li	a2, 0
+	mv	a3, s7
+	call	halide_do_par_for@plt
+.LBB2_20:                               # %destructor_block
+	li	a0, 0
+	ld	ra, 168(sp)                     # 8-byte Folded Reload
+	ld	s0, 160(sp)                     # 8-byte Folded Reload
+	ld	s1, 152(sp)                     # 8-byte Folded Reload
+	ld	s2, 144(sp)                     # 8-byte Folded Reload
+	ld	s3, 136(sp)                     # 8-byte Folded Reload
+	ld	s4, 128(sp)                     # 8-byte Folded Reload
+	ld	s5, 120(sp)                     # 8-byte Folded Reload
+	ld	s6, 112(sp)                     # 8-byte Folded Reload
+	ld	s7, 104(sp)                     # 8-byte Folded Reload
+	ld	s8, 96(sp)                      # 8-byte Folded Reload
+	ld	s9, 88(sp)                      # 8-byte Folded Reload
+	addi	sp, sp, 176
+	ret
+.Lfunc_end2:
+	.size	idw_halide_parallel_vec_, .Lfunc_end2-idw_halide_parallel_vec_
+                                        # -- End function
+	.section	.text.idw_halide_parallel_vec__argv,"ax",@progbits
+	.globl	idw_halide_parallel_vec__argv   # -- Begin function idw_halide_parallel_vec__argv
+	.p2align	1
+	.type	idw_halide_parallel_vec__argv,@function
+idw_halide_parallel_vec__argv:          # @idw_halide_parallel_vec__argv
+# %bb.0:                                # %entry
+	addi	sp, sp, -16
+	sd	ra, 8(sp)                       # 8-byte Folded Spill
+	ld	a0, 0(a0)
+	call	idw_halide_parallel_vec_@plt
+	li	a0, 0
+	ld	ra, 8(sp)                       # 8-byte Folded Reload
+	addi	sp, sp, 16
+	ret
+.Lfunc_end3:
+	.size	idw_halide_parallel_vec__argv, .Lfunc_end3-idw_halide_parallel_vec__argv
+                                        # -- End function
+	.section	.text.idw_halide_parallel_vec__metadata,"ax",@progbits
+	.globl	idw_halide_parallel_vec__metadata # -- Begin function idw_halide_parallel_vec__metadata
+	.p2align	1
+	.type	idw_halide_parallel_vec__metadata,@function
+idw_halide_parallel_vec__metadata:      # @idw_halide_parallel_vec__metadata
+# %bb.0:                                # %entry
+.Lpcrel_hi4:
+	auipc	a0, %pcrel_hi(.Lidw_halide_parallel_vec__metadata_storage)
+	addi	a0, a0, %pcrel_lo(.Lpcrel_hi4)
+	ret
+.Lfunc_end4:
+	.size	idw_halide_parallel_vec__metadata, .Lfunc_end4-idw_halide_parallel_vec__metadata
+                                        # -- End function
+	.type	.Lb34.shape,@object             # @b34.shape
+	.section	.rodata,"a",@progbits
+	.p2align	5, 0x0
+.Lb34.shape:
+	.asciz	"\000\000\000\000,\001\000\000\001\000\000\000\000\000\000"
+	.size	.Lb34.shape, 16
+
+	.type	.Lb34.data,@object              # @b34.data
+	.p2align	5, 0x0
+.Lb34.data:
+	.asciz	"\000\000\000\000\000\000\000\000H\000\000\000\000\000\000\000\325\000\000\000O\000\000\000\000\000\000\000\252\001\000\000<\000\000\000\000\000\000\000\200\002\000\000L\000\000\000\000\000\000\000U\003\000\000\200\000\000\000\000\000\000\000*\004\000\000C\000\000\000\000\000\000\000\000\005\000\000A\000\000\000\000\000\000\000\325\005\000\000@\000\000\000\000\000\000\000\252\006\000\000<\000\000\000\000\000\000\000\200\007\000\000=\000\000\000x\000\000\000\000\000\000\000Q\000\000\000x\000\000\000\325\000\000\000O\000\000\000x\000\000\000\252\001\000\000:\000\000\000x\000\000\000\200\002\000\000\204\000\000\000x\000\000\000U\003\000\000\225\000\000\000x\000\000\000*\004\000\000\216\000\000\000x\000\000\000\000\005\000\000@\000\000\000x\000\000\000\325\005\000\000E\000\000\000x\000\000\000\252\006\000\000A\000\000\000x\000\000\000\200\007\000\000@\000\000\000\360\000\000\000\000\000\000\000K\000\000\000\360\000\000\000\325\000\000\000D\000\000\000\360\000\000\000\252\001\000\000\214\000\000\000\360\000\000\000\200\002\000\000\231\000\000\000\360\000\000\000U\003\000\000\221\000\000\000\360\000\000\000*\004\000\000\204\000\000\000\360\000\000\000\000\005\000\000\230\000\000\000\360\000\000\000\325\005\000\000}\000\000\000\360\000\000\000\252\006\000\000B\000\000\000\360\000\000\000\200\007\000\000:\000\000\000h\001\000\000\000\000\000\000N\000\000\000h\001\000\000\325\000\000\000<\000\000\000h\001\000\000\252\001\000\000\213\000\000\000h\001\000\000\200\002\000\000\250\000\000\000h\001\000\000U\003\000\000\232\000\000\000h\001\000\000*\004\000\000\212\000\000\000h\001\000\000\000\005\000\000\221\000\000\000h\001\000\000\325\005\000\000\240\000\000\000h\001\000\000\252\006\000\000D\000\000\000h\001\000\000\200\007\000\000<\000\000\000\340\001\000\000\000\000\000\000M\000\000\000\340\001\000\000\325\000\000\000;\000\000\000\340\001\000\000\252\001\000\000\245\000\000\000\340\001\000\000\200\002\000\000\267\000\000\000\340\001\000\000U\003\000\000\246\000\000\000\340\001\000\000*\004\000\000\216\000\000\000\340\001\000\000\000\005\000\000{\000\000\000\340\001\000\000\325\005\000\000\233\000\000\000\340\001\000\000\252\006\000\000\220\000\000\000\340\001\000\000\200\007\000\000<\000\000\000X\002\000\000\000\000\000\000S\000\000\000X\002\000\000\325\000\000\000A\000\000\000X\002\000\000\252\001\000\000\262\000\000\000X\002\000\000\200\002\000\000\270\000\000\000X\002\000\000U\003\000\000\212\000\000\000X\002\000\000*\004\000\000|\000\000\000X\002\000\000\000\005\000\000\204\000\000\000X\002\000\000\325\005\000\000\257\000\000\000X\002\000\000\252\006\000\000\257\000\000\000X\002\000\000\200\007\000\000<\000\000\000\320\002\000\000\000\000\000\000V\000\000\000\320\002\000\000\325\000\000\000<\000\000\000\320\002\000\000\252\001\000\000\266\000\000\000\320\002\000\000\200\002\000\000\263\000\000\000\320\002\000\000U\003\000\000\230\000\000\000\320\002\000\000*\004\000\000\214\000\000\000\320\002\000\000\000\005\000\000s\000\000\000\320\002\000\000\325\005\000\000\234\000\000\000\320\002\000\000\252\006\000\000\254\000\000\000\320\002\000\000\200\007\000\000A\000\000\000H\003\000\000\000\000\000\000Z\000\000\000H\003\000\000\325\000\000\000E\000\000\000H\003\000\000\252\001\000\000\244\000\000\000H\003\000\000\200\002\000\000\272\000\000\000H\003\000\000U\003\000\000\222\000\000\000H\003\000\000*\004\000\000\223\000\000\000H\003\000\000\000\005\000\000z\000\000\000H\003\000\000\325\005\000\000\230\000\000\000H\003\000\000\252\006\000\000\246\000\000\000H\003\000\000\200\007\000\000E\000\000\000\300\003\000\000\000\000\000\000U\000\000\000\300\003\000\000\325\000\000\000C\000\000\000\300\003\000\000\252\001\000\000\272\000\000\000\300\003\000\000\200\002\000\000\251\000\000\000\300\003\000\000U\003\000\000\242\000\000\000\300\003\000\000*\004\000\000\223\000\000\000\300\003\000\000\000\005\000\000|\000\000\000\300\003\000\000\325\005\000\000\225\000\000\000\300\003\000\000\252\006\000\000\235\000\000\000\300\003\000\000\200\007\000\000E\000\000\0008\004\000\000\000\000\000\000U\000\000\0008\004\000\000\325\000\000\000F\000\000\0008\004\000\000\252\001\000\000\301\000\000\0008\004\000\000\200\002\000\000\300\000\000\0008\004\000\000U\003\000\000\233\000\000\0008\004\000\000*\004\000\000\207\000\000\0008\004\000\000\000\005\000\000x\000\000\0008\004\000\000\325\005\000\000\212\000\000\0008\004\000\000\252\006\000\000\200\000\000\0008\004\000\000\200\007\000\000F\000\000"
+	.size	.Lb34.data, 1200
+
+	.type	.Lb34.buffer,@object            # @b34.buffer
+	.data
+	.p2align	4, 0x0
+.Lb34.buffer:
+	.quad	0                               # 0x0
+	.quad	0
+	.quad	.Lb34.data
+	.quad	1                               # 0x1
+	.byte	0                               # 0x0
+	.byte	32                              # 0x20
+	.half	1                               # 0x1
+	.word	1                               # 0x1
+	.quad	.Lb34.shape
+	.quad	0
+	.size	.Lb34.buffer, 56
+
+	.type	.Lb35.shape,@object             # @b35.shape
+	.section	.rodata,"a",@progbits
+	.p2align	5, 0x0
+.Lb35.shape:
+	.asciz	"\000\000\000\000d\000\000\000\001\000\000\000\000\000\000"
+	.size	.Lb35.shape, 16
+
+	.type	.Lb35.data,@object              # @b35.data
+	.p2align	5, 0x0
+.Lb35.data:
+	.ascii	"\211\004&=\244\022\233\275\217U%;\272)J>#\026\024\276@_\234>\340\243|\275\334\034K\275\232\300\004:A\313\245\274\303\024r\275\345\313a\275x\231\300>\201\334\"\276\236}\321\275H\346\257\276\202V\310>\225\375(>},\026\274E\235W\275\241\236\236<\245\220\222=\004\361\263\276\2620\240\274\265\036\240=\\:\336=\240\016\263\276\260\224\371\275p\235(=\311\\x;\226y\\\275\360\227\377=\275\364\200=rw/<@U\017=c\327\003=nV\267\273\354<Y\276\257$\260>3\207\311\274<&{\274h\b\031>\203\222\003\276\213\3649\275d\177\017\276\330\306\246\275'\314:>\203\024\253=S\331M\276\257\263S=b\221q\275\336\t\306=#^\005\2763)\256\274\344\255G>\224!\374=\374\360o\275\215#\031\276\322`J\276y6\360=\273\303;\275N\"N>M\022K\276\016\035C=1Fb\275u\017\220\275\0040\033>,\037\223=t\335\304\275\363\344\271=\320\341\260\275\2265\255=\310\"b=&\031\330\275;\314\000>>\347o\275\255\222\020=C\rU<\370d\307\275\373\306\206=I\345\373\274Y|\034>WY;\276O\240?>B\347\226\275\203\271\203\275W\243\313<\214\364\343\274Ey\033\276J4\214=\261\304\241\274-\271p>X\313l\276*_\366\275\260^\\=\237l1=\371\272\200=\002\373?\274*\3578<]Gv="
+	.size	.Lb35.data, 400
+
+	.type	.Lb35.buffer,@object            # @b35.buffer
+	.data
+	.p2align	4, 0x0
+.Lb35.buffer:
+	.quad	0                               # 0x0
+	.quad	0
+	.quad	.Lb35.data
+	.quad	1                               # 0x1
+	.byte	2                               # 0x2
+	.byte	32                              # 0x20
+	.half	1                               # 0x1
+	.word	1                               # 0x1
+	.quad	.Lb35.shape
+	.quad	0
+	.size	.Lb35.buffer, 56
+
+	.type	.L__unnamed_1,@object           # @0
+	.section	.rodata,"a",@progbits
+	.p2align	4, 0x0
+.L__unnamed_1:
+	.zero	32
+	.size	.L__unnamed_1, 32
+
+	.type	.Lstr,@object                   # @str
+	.p2align	5, 0x0
+.Lstr:
+	.asciz	"idw_halide_parallel_vec_"
+	.size	.Lstr, 25
+
+	.type	.L__unnamed_2,@object           # @1
+	.section	.data.rel.ro,"aw",@progbits
+	.p2align	4, 0x0
+.L__unnamed_2:
+	.quad	.Lstr
+	.word	2                               # 0x2
+	.word	2                               # 0x2
+	.byte	2                               # 0x2
+	.byte	32                              # 0x20
+	.half	1                               # 0x1
+	.zero	4
+	.quad	0
+	.quad	0
+	.quad	0
+	.quad	0
+	.quad	.L__unnamed_1
+	.size	.L__unnamed_2, 64
+
+	.type	.Lstr.4,@object                 # @str.4
+	.section	.rodata,"a",@progbits
+	.p2align	5, 0x0
+.Lstr.4:
+	.asciz	"riscv-64-linux-no_asserts-no_runtime-rvv-vector_bits_128"
+	.size	.Lstr.4, 57
+
+	.type	.Lidw_halide_parallel_vec__metadata_storage,@object # @idw_halide_parallel_vec__metadata_storage
+	.section	.data.rel.ro,"aw",@progbits
+	.p2align	4, 0x0
+.Lidw_halide_parallel_vec__metadata_storage:
+	.word	1                               # 0x1
+	.word	1                               # 0x1
+	.quad	.L__unnamed_2
+	.quad	.Lstr.4
+	.quad	.Lstr
+	.size	.Lidw_halide_parallel_vec__metadata_storage, 32
+
+	.ident	"clang version 16.0.0 (https://github.com/dkurt/llvm-rvv-071 b027aa1b59c9f53240bdc836f39656723fdf9df0)"
+	.ident	"clang version 16.0.0 (https://github.com/dkurt/llvm-rvv-071 b027aa1b59c9f53240bdc836f39656723fdf9df0)"
+	.ident	"clang version 16.0.0 (https://github.com/dkurt/llvm-rvv-071 b027aa1b59c9f53240bdc836f39656723fdf9df0)"
+	.ident	"clang version 16.0.0 (https://github.com/dkurt/llvm-rvv-071 b027aa1b59c9f53240bdc836f39656723fdf9df0)"
+	.ident	"clang version 16.0.0 (https://github.com/dkurt/llvm-rvv-071 b027aa1b59c9f53240bdc836f39656723fdf9df0)"
+	.ident	"clang version 16.0.0 (https://github.com/dkurt/llvm-rvv-071 b027aa1b59c9f53240bdc836f39656723fdf9df0)"
+	.section	".note.GNU-stack","",@progbits
diff --git a/aot/idw_halide_vec.h b/aot/idw_halide_vec.h
new file mode 100644
index 0000000..6de0b40
--- /dev/null
+++ b/aot/idw_halide_vec.h
@@ -0,0 +1,56 @@
+#ifndef HALIDE__idw_halide_vec_h
+#define HALIDE__idw_halide_vec_h
+#include <stdint.h>
+
+// Forward declarations of the types used in the interface
+// to the Halide pipeline.
+//
+// For the definitions of these structs, include HalideRuntime.h
+
+// Halide's representation of a multi-dimensional array.
+// Halide::Runtime::Buffer is a more user-friendly wrapper
+// around this. Its declaration is in HalideBuffer.h
+struct halide_buffer_t;
+
+// Metadata describing the arguments to the generated function.
+// Used to construct calls to the _argv version of the function.
+struct halide_filter_metadata_t;
+
+#ifndef HALIDE_MUST_USE_RESULT
+#ifdef __has_attribute
+#if __has_attribute(nodiscard)
+#define HALIDE_MUST_USE_RESULT [[nodiscard]]
+#elif __has_attribute(warn_unused_result)
+#define HALIDE_MUST_USE_RESULT __attribute__((warn_unused_result))
+#else
+#define HALIDE_MUST_USE_RESULT
+#endif
+#else
+#define HALIDE_MUST_USE_RESULT
+#endif
+#endif
+
+#ifndef HALIDE_FUNCTION_ATTRS
+#define HALIDE_FUNCTION_ATTRS
+#endif
+
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+HALIDE_FUNCTION_ATTRS
+int idw_halide_vec_(struct halide_buffer_t *_idw_halide_vec__buffer);
+
+HALIDE_FUNCTION_ATTRS
+int idw_halide_vec__argv(void **args);
+
+HALIDE_FUNCTION_ATTRS
+const struct halide_filter_metadata_t *idw_halide_vec__metadata();
+
+#ifdef __cplusplus
+}  // extern "C"
+#endif
+
+#endif
diff --git a/aot/idw_halide_vec.s b/aot/idw_halide_vec.s
new file mode 100644
index 0000000..f595e22
--- /dev/null
+++ b/aot/idw_halide_vec.s
@@ -0,0 +1,450 @@
+	.text
+	.attribute	4, 16
+	.attribute	5, "rv64gcv0p7"
+	.file	"halide_buffer_t.cpp"
+	.section	.text.idw_halide_vec_,"ax",@progbits
+	.globl	idw_halide_vec_                 # -- Begin function idw_halide_vec_
+	.p2align	1
+	.type	idw_halide_vec_,@function
+idw_halide_vec_:                        # @idw_halide_vec_
+# %bb.0:                                # %entry
+	addi	sp, sp, -192
+	sd	ra, 184(sp)                     # 8-byte Folded Spill
+	sd	s0, 176(sp)                     # 8-byte Folded Spill
+	sd	s1, 168(sp)                     # 8-byte Folded Spill
+	sd	s2, 160(sp)                     # 8-byte Folded Spill
+	sd	s3, 152(sp)                     # 8-byte Folded Spill
+	sd	s4, 144(sp)                     # 8-byte Folded Spill
+	sd	s5, 136(sp)                     # 8-byte Folded Spill
+	sd	s6, 128(sp)                     # 8-byte Folded Spill
+	sd	s7, 120(sp)                     # 8-byte Folded Spill
+	sd	s8, 112(sp)                     # 8-byte Folded Spill
+	sd	s9, 104(sp)                     # 8-byte Folded Spill
+	sd	s10, 96(sp)                     # 8-byte Folded Spill
+	sd	s11, 88(sp)                     # 8-byte Folded Spill
+.Lpcrel_hi0:
+	auipc	a1, %pcrel_hi(.Lb44.buffer)
+	addi	a1, a1, %pcrel_lo(.Lpcrel_hi0)
+	ld	s6, 16(a1)
+	ld	a3, 40(a0)
+.Lpcrel_hi1:
+	auipc	a2, %pcrel_hi(.Lb45.buffer)
+	addi	a2, a2, %pcrel_lo(.Lpcrel_hi1)
+	ld	s9, 16(a2)
+	lw	s4, 0(a3)
+	lwu	s8, 4(a3)
+	lw	a4, 16(a3)
+	sd	a4, 8(sp)                       # 8-byte Folded Spill
+	ld	a4, 0(a1)
+	lwu	a5, 20(a3)
+	sd	a5, 32(sp)                      # 8-byte Folded Spill
+	lw	s1, 24(a3)
+	or	a5, s6, a4
+	mv	a4, s9
+	bnez	a5, .LBB0_2
+# %bb.1:                                # %then_bb
+	lui	a4, 128
+	addiw	a4, a4, 9
+	slli	a4, a4, 13
+	sd	a4, 32(a1)
+	ld	a4, 40(a1)
+	sd	zero, 0(a1)
+	sd	zero, 8(a1)
+	sd	zero, 16(a1)
+	sw	zero, 0(a4)
+	li	a5, 300
+	sw	a5, 4(a4)
+	li	a5, 1
+	sw	a5, 8(a4)
+	sw	zero, 12(a4)
+	ld	a4, 16(a2)
+	sd	zero, 24(a1)
+.LBB0_2:                                # %after_bb
+	ld	a5, 0(a2)
+	ld	s11, 16(a0)
+	or	a4, a4, a5
+	bnez	a4, .LBB0_4
+# %bb.3:                                # %then_bb2
+	sd	zero, 16(a2)
+	sd	zero, 8(a2)
+	sd	zero, 0(a2)
+	lui	a4, 128
+	addiw	a4, a4, 9
+	ld	a5, 40(a2)
+	slli	a4, a4, 13
+	addi	a4, a4, 2
+	sd	a4, 32(a2)
+	sw	zero, 0(a5)
+	li	a4, 100
+	sw	a4, 4(a5)
+	li	a4, 1
+	sw	a4, 8(a5)
+	sw	zero, 12(a5)
+	sd	zero, 24(a2)
+.LBB0_4:                                # %after_bb1
+	sext.w	a4, s8
+	sd	a4, 24(sp)                      # 8-byte Folded Spill
+	lw	a4, 32(sp)                      # 8-byte Folded Reload
+	sd	s1, 16(sp)                      # 8-byte Folded Spill
+	beqz	s11, .LBB0_6
+# %bb.5:
+	li	a0, 0
+	j	.LBB0_9
+.LBB0_6:                                # %_halide_buffer_is_bounds_query.exit36
+	ld	a5, 0(a0)
+	bnez	a5, .LBB0_8
+# %bb.7:                                # %then_bb5
+	sd	zero, 16(a0)
+	sd	zero, 8(a0)
+	sd	zero, 0(a0)
+	lui	a5, 256
+	addiw	a5, a5, 9
+	slli	a5, a5, 13
+	addi	a5, a5, 2
+	sd	a5, 32(a0)
+	sw	s4, 0(a3)
+	ld	s1, 24(sp)                      # 8-byte Folded Reload
+	sw	s1, 4(a3)
+	li	a5, 1
+	sw	a5, 8(a3)
+	sw	zero, 12(a3)
+	ld	a5, 8(sp)                       # 8-byte Folded Reload
+	sw	a5, 16(a3)
+	sw	a4, 20(a3)
+	sw	s1, 24(a3)
+	sw	zero, 28(a3)
+	sd	zero, 24(a0)
+.LBB0_8:                                # %land.rhs.i49
+	ld	a0, 0(a0)
+	seqz	a0, a0
+.LBB0_9:                                # %_halide_buffer_is_bounds_query.exit50
+	ld	a3, 16(a1)
+	ld	a1, 0(a1)
+	ld	a5, 16(a2)
+	ld	a2, 0(a2)
+	or	a1, a1, a3
+	seqz	a1, a1
+	or	a2, a2, a5
+	seqz	a2, a2
+	or	a1, a1, a2
+	or	a0, a0, a1
+	slti	a1, a4, 1
+	or	a0, a0, a1
+	bnez	a0, .LBB0_20
+# %bb.10:                               # %"for idw_halide_vec_.s0.y.rebased.preheader"
+	ld	s3, 16(sp)                      # 8-byte Folded Reload
+	ld	a0, 24(sp)                      # 8-byte Folded Reload
+	blez	a0, .LBB0_13
+# %bb.11:                               # %"for idw_halide_vec_.s0.y.rebased.us.preheader"
+	li	s0, 0
+	ld	s2, 24(sp)                      # 8-byte Folded Reload
+	slli	s2, s2, 2
+	ld	s1, 32(sp)                      # 8-byte Folded Reload
+.LBB0_12:                               # %"for idw_halide_vec_.s0.y.rebased.us"
+                                        # =>This Inner Loop Header: Depth=1
+	slli	a0, s0, 2
+	add	a0, a0, s11
+	li	a1, 0
+	mv	a2, s2
+	call	memset@plt
+	addi	s1, s1, -1
+	addw	s0, s0, s3
+	bnez	s1, .LBB0_12
+.LBB0_13:                               # %"for idw_halide_vec_.s1.y.rebased.preheader"
+	sd	zero, 40(sp)                    # 8-byte Folded Spill
+	li	a1, 6
+	li	a2, 1
+	addi	a3, sp, 64
+	addi	a7, sp, 76
+	addi	t0, sp, 72
+	addi	t1, sp, 68
+	li	s0, 4
+	addi	s1, sp, 48
+	addi	t2, sp, 52
+	addi	t3, sp, 60
+	addi	t6, sp, 56
+	li	ra, 2
+	li	a6, 3
+.LBB0_14:                               # %"for idw_halide_vec_.s1.y.rebased"
+                                        # =>This Loop Header: Depth=1
+                                        #     Child Loop BB0_16 Depth 2
+                                        #       Child Loop BB0_17 Depth 3
+	ld	a0, 24(sp)                      # 8-byte Folded Reload
+	blez	a0, .LBB0_19
+# %bb.15:                               # %"for idw_halide_vec_.s1.x.rebased.preheader"
+                                        #   in Loop: Header=BB0_14 Depth=1
+	li	t4, 0
+	ld	a0, 16(sp)                      # 8-byte Folded Reload
+	ld	a4, 40(sp)                      # 8-byte Folded Reload
+	mulw	t5, a0, a4
+	ld	a0, 8(sp)                       # 8-byte Folded Reload
+	addw	a0, a0, a4
+.LBB0_16:                               # %"for idw_halide_vec_.s1.x.rebased"
+                                        #   Parent Loop BB0_14 Depth=1
+                                        # =>  This Loop Header: Depth=2
+                                        #       Child Loop BB0_17 Depth 3
+	add	a4, t4, t5
+	slli	a4, a4, 2
+	add	s2, s11, a4
+	flw	ft0, 0(s2)
+	addw	a4, s4, t4
+	li	s5, 25
+	mv	s3, s6
+	mv	s10, s9
+.LBB0_17:                               # %"for idw_halide_vec_.s1.r24$x.r24$x"
+                                        #   Parent Loop BB0_14 Depth=1
+                                        #     Parent Loop BB0_16 Depth=2
+                                        # =>    This Inner Loop Header: Depth=3
+	addi	s7, s3, 4
+	vsetvli	zero, a1, e32, m2
+	vlwu.v	v8, (s7)
+	vsetvli	zero, a2, e32, m2
+	vslidedown.vi	v10, v8, 3
+	addi	a5, s3, 20
+	vsetvli	zero, a1, e32, m2
+	vlwu.v	v12, (a5)
+	vsetvli	zero, a2, e32, m2
+	vslidedown.vi	v14, v12, 2
+	vslidedown.vi	v12, v12, 5
+	vsw.v	v8, (a3)
+	vsw.v	v12, (a7)
+	vsw.v	v14, (t0)
+	vsw.v	v10, (t1)
+	vsetvli	zero, s0, e32, m1
+	vlwu.v	v8, (a3)
+	vsetvli	a5, zero, e32, m1
+	vrsub.vx	v8, v8, a4
+	vsetvli	zero, a1, e32, m2
+	vlwu.v	v10, (s3)
+	vsetvli	zero, a2, e32, m2
+	vslidedown.vi	v12, v10, 3
+	addi	a5, s3, 16
+	vsetvli	zero, a1, e32, m2
+	vlwu.v	v14, (a5)
+	vsetvli	zero, a2, e32, m2
+	vslidedown.vi	v16, v14, 2
+	vslidedown.vi	v14, v14, 5
+	vsw.v	v10, (s1)
+	vsw.v	v12, (t2)
+	vsw.v	v14, (t3)
+	vsw.v	v16, (t6)
+	vsetvli	zero, s0, e32, m1
+	vlwu.v	v9, (s1)
+	vsetvli	a5, zero, e32, m1
+	vrsub.vx	v9, v9, a0
+	vmul.vv	v9, v9, v9
+	vmacc.vv	v9, v8, v8
+	vfcvt.f.x.v	v8, v9
+	vfmv.f.s	ft1, v8
+	fsqrt.s	ft1, ft1
+	vfmv.s.f	v9, ft1
+	vsetvli	zero, a2, e32, m1
+	vslidedown.vi	v10, v8, 1
+	vfmv.f.s	ft1, v10
+	fsqrt.s	ft1, ft1
+	vsetvli	a5, zero, e32, m1
+	vfmv.s.f	v10, ft1
+	vsetvli	zero, ra, e32, m1
+	vslideup.vi	v9, v10, 1
+	vsetvli	zero, a2, e32, m1
+	vslidedown.vi	v10, v8, 2
+	vfmv.f.s	ft1, v10
+	fsqrt.s	ft1, ft1
+	vsetvli	a5, zero, e32, m1
+	vfmv.s.f	v10, ft1
+	vsetvli	zero, a6, e32, m1
+	vslideup.vi	v9, v10, 2
+	vsetvli	zero, a2, e32, m1
+	vslidedown.vi	v8, v8, 3
+	vfmv.f.s	ft1, v8
+	fsqrt.s	ft1, ft1
+	vsetvli	a5, zero, e32, m1
+	vfmv.s.f	v8, ft1
+	vsetvli	zero, s0, e32, m1
+	vslideup.vi	v9, v8, 3
+	vsetvli	zero, zero, e32, m1
+	vlwu.v	v8, (s10)
+	vfmul.vv	v8, v9, v8
+	vfmv.s.f	v9, ft0
+	vfredsum.vs	v8, v8, v9
+	vfmv.f.s	ft0, v8
+	addi	s5, s5, -1
+	addi	s10, s10, 16
+	addi	s3, s3, 48
+	bnez	s5, .LBB0_17
+# %bb.18:                               # %"end for idw_halide_vec_.s1.r24$x.r24$x"
+                                        #   in Loop: Header=BB0_16 Depth=2
+	addi	t4, t4, 1
+	fsw	ft0, 0(s2)
+	bne	t4, s8, .LBB0_16
+.LBB0_19:                               # %"end for idw_halide_vec_.s1.x.rebased"
+                                        #   in Loop: Header=BB0_14 Depth=1
+	ld	a4, 40(sp)                      # 8-byte Folded Reload
+	addi	a4, a4, 1
+	ld	a0, 32(sp)                      # 8-byte Folded Reload
+	sd	a4, 40(sp)                      # 8-byte Folded Spill
+	bne	a4, a0, .LBB0_14
+.LBB0_20:                               # %destructor_block
+	li	a0, 0
+	ld	ra, 184(sp)                     # 8-byte Folded Reload
+	ld	s0, 176(sp)                     # 8-byte Folded Reload
+	ld	s1, 168(sp)                     # 8-byte Folded Reload
+	ld	s2, 160(sp)                     # 8-byte Folded Reload
+	ld	s3, 152(sp)                     # 8-byte Folded Reload
+	ld	s4, 144(sp)                     # 8-byte Folded Reload
+	ld	s5, 136(sp)                     # 8-byte Folded Reload
+	ld	s6, 128(sp)                     # 8-byte Folded Reload
+	ld	s7, 120(sp)                     # 8-byte Folded Reload
+	ld	s8, 112(sp)                     # 8-byte Folded Reload
+	ld	s9, 104(sp)                     # 8-byte Folded Reload
+	ld	s10, 96(sp)                     # 8-byte Folded Reload
+	ld	s11, 88(sp)                     # 8-byte Folded Reload
+	addi	sp, sp, 192
+	ret
+.Lfunc_end0:
+	.size	idw_halide_vec_, .Lfunc_end0-idw_halide_vec_
+                                        # -- End function
+	.section	.text.idw_halide_vec__argv,"ax",@progbits
+	.globl	idw_halide_vec__argv            # -- Begin function idw_halide_vec__argv
+	.p2align	1
+	.type	idw_halide_vec__argv,@function
+idw_halide_vec__argv:                   # @idw_halide_vec__argv
+# %bb.0:                                # %entry
+	addi	sp, sp, -16
+	sd	ra, 8(sp)                       # 8-byte Folded Spill
+	ld	a0, 0(a0)
+	call	idw_halide_vec_@plt
+	li	a0, 0
+	ld	ra, 8(sp)                       # 8-byte Folded Reload
+	addi	sp, sp, 16
+	ret
+.Lfunc_end1:
+	.size	idw_halide_vec__argv, .Lfunc_end1-idw_halide_vec__argv
+                                        # -- End function
+	.section	.text.idw_halide_vec__metadata,"ax",@progbits
+	.globl	idw_halide_vec__metadata        # -- Begin function idw_halide_vec__metadata
+	.p2align	1
+	.type	idw_halide_vec__metadata,@function
+idw_halide_vec__metadata:               # @idw_halide_vec__metadata
+# %bb.0:                                # %entry
+.Lpcrel_hi2:
+	auipc	a0, %pcrel_hi(.Lidw_halide_vec__metadata_storage)
+	addi	a0, a0, %pcrel_lo(.Lpcrel_hi2)
+	ret
+.Lfunc_end2:
+	.size	idw_halide_vec__metadata, .Lfunc_end2-idw_halide_vec__metadata
+                                        # -- End function
+	.type	.Lb44.shape,@object             # @b44.shape
+	.section	.rodata,"a",@progbits
+	.p2align	5, 0x0
+.Lb44.shape:
+	.asciz	"\000\000\000\000,\001\000\000\001\000\000\000\000\000\000"
+	.size	.Lb44.shape, 16
+
+	.type	.Lb44.data,@object              # @b44.data
+	.p2align	5, 0x0
+.Lb44.data:
+	.asciz	"\000\000\000\000\000\000\000\000H\000\000\000\000\000\000\000\325\000\000\000O\000\000\000\000\000\000\000\252\001\000\000<\000\000\000\000\000\000\000\200\002\000\000L\000\000\000\000\000\000\000U\003\000\000\200\000\000\000\000\000\000\000*\004\000\000C\000\000\000\000\000\000\000\000\005\000\000A\000\000\000\000\000\000\000\325\005\000\000@\000\000\000\000\000\000\000\252\006\000\000<\000\000\000\000\000\000\000\200\007\000\000=\000\000\000x\000\000\000\000\000\000\000Q\000\000\000x\000\000\000\325\000\000\000O\000\000\000x\000\000\000\252\001\000\000:\000\000\000x\000\000\000\200\002\000\000\204\000\000\000x\000\000\000U\003\000\000\225\000\000\000x\000\000\000*\004\000\000\216\000\000\000x\000\000\000\000\005\000\000@\000\000\000x\000\000\000\325\005\000\000E\000\000\000x\000\000\000\252\006\000\000A\000\000\000x\000\000\000\200\007\000\000@\000\000\000\360\000\000\000\000\000\000\000K\000\000\000\360\000\000\000\325\000\000\000D\000\000\000\360\000\000\000\252\001\000\000\214\000\000\000\360\000\000\000\200\002\000\000\231\000\000\000\360\000\000\000U\003\000\000\221\000\000\000\360\000\000\000*\004\000\000\204\000\000\000\360\000\000\000\000\005\000\000\230\000\000\000\360\000\000\000\325\005\000\000}\000\000\000\360\000\000\000\252\006\000\000B\000\000\000\360\000\000\000\200\007\000\000:\000\000\000h\001\000\000\000\000\000\000N\000\000\000h\001\000\000\325\000\000\000<\000\000\000h\001\000\000\252\001\000\000\213\000\000\000h\001\000\000\200\002\000\000\250\000\000\000h\001\000\000U\003\000\000\232\000\000\000h\001\000\000*\004\000\000\212\000\000\000h\001\000\000\000\005\000\000\221\000\000\000h\001\000\000\325\005\000\000\240\000\000\000h\001\000\000\252\006\000\000D\000\000\000h\001\000\000\200\007\000\000<\000\000\000\340\001\000\000\000\000\000\000M\000\000\000\340\001\000\000\325\000\000\000;\000\000\000\340\001\000\000\252\001\000\000\245\000\000\000\340\001\000\000\200\002\000\000\267\000\000\000\340\001\000\000U\003\000\000\246\000\000\000\340\001\000\000*\004\000\000\216\000\000\000\340\001\000\000\000\005\000\000{\000\000\000\340\001\000\000\325\005\000\000\233\000\000\000\340\001\000\000\252\006\000\000\220\000\000\000\340\001\000\000\200\007\000\000<\000\000\000X\002\000\000\000\000\000\000S\000\000\000X\002\000\000\325\000\000\000A\000\000\000X\002\000\000\252\001\000\000\262\000\000\000X\002\000\000\200\002\000\000\270\000\000\000X\002\000\000U\003\000\000\212\000\000\000X\002\000\000*\004\000\000|\000\000\000X\002\000\000\000\005\000\000\204\000\000\000X\002\000\000\325\005\000\000\257\000\000\000X\002\000\000\252\006\000\000\257\000\000\000X\002\000\000\200\007\000\000<\000\000\000\320\002\000\000\000\000\000\000V\000\000\000\320\002\000\000\325\000\000\000<\000\000\000\320\002\000\000\252\001\000\000\266\000\000\000\320\002\000\000\200\002\000\000\263\000\000\000\320\002\000\000U\003\000\000\230\000\000\000\320\002\000\000*\004\000\000\214\000\000\000\320\002\000\000\000\005\000\000s\000\000\000\320\002\000\000\325\005\000\000\234\000\000\000\320\002\000\000\252\006\000\000\254\000\000\000\320\002\000\000\200\007\000\000A\000\000\000H\003\000\000\000\000\000\000Z\000\000\000H\003\000\000\325\000\000\000E\000\000\000H\003\000\000\252\001\000\000\244\000\000\000H\003\000\000\200\002\000\000\272\000\000\000H\003\000\000U\003\000\000\222\000\000\000H\003\000\000*\004\000\000\223\000\000\000H\003\000\000\000\005\000\000z\000\000\000H\003\000\000\325\005\000\000\230\000\000\000H\003\000\000\252\006\000\000\246\000\000\000H\003\000\000\200\007\000\000E\000\000\000\300\003\000\000\000\000\000\000U\000\000\000\300\003\000\000\325\000\000\000C\000\000\000\300\003\000\000\252\001\000\000\272\000\000\000\300\003\000\000\200\002\000\000\251\000\000\000\300\003\000\000U\003\000\000\242\000\000\000\300\003\000\000*\004\000\000\223\000\000\000\300\003\000\000\000\005\000\000|\000\000\000\300\003\000\000\325\005\000\000\225\000\000\000\300\003\000\000\252\006\000\000\235\000\000\000\300\003\000\000\200\007\000\000E\000\000\0008\004\000\000\000\000\000\000U\000\000\0008\004\000\000\325\000\000\000F\000\000\0008\004\000\000\252\001\000\000\301\000\000\0008\004\000\000\200\002\000\000\300\000\000\0008\004\000\000U\003\000\000\233\000\000\0008\004\000\000*\004\000\000\207\000\000\0008\004\000\000\000\005\000\000x\000\000\0008\004\000\000\325\005\000\000\212\000\000\0008\004\000\000\252\006\000\000\200\000\000\0008\004\000\000\200\007\000\000F\000\000"
+	.size	.Lb44.data, 1200
+
+	.type	.Lb44.buffer,@object            # @b44.buffer
+	.data
+	.p2align	4, 0x0
+.Lb44.buffer:
+	.quad	0                               # 0x0
+	.quad	0
+	.quad	.Lb44.data
+	.quad	1                               # 0x1
+	.byte	0                               # 0x0
+	.byte	32                              # 0x20
+	.half	1                               # 0x1
+	.word	1                               # 0x1
+	.quad	.Lb44.shape
+	.quad	0
+	.size	.Lb44.buffer, 56
+
+	.type	.Lb45.shape,@object             # @b45.shape
+	.section	.rodata,"a",@progbits
+	.p2align	5, 0x0
+.Lb45.shape:
+	.asciz	"\000\000\000\000d\000\000\000\001\000\000\000\000\000\000"
+	.size	.Lb45.shape, 16
+
+	.type	.Lb45.data,@object              # @b45.data
+	.p2align	5, 0x0
+.Lb45.data:
+	.ascii	"\211\004&=\244\022\233\275\217U%;\272)J>#\026\024\276@_\234>\340\243|\275\334\034K\275\232\300\004:A\313\245\274\303\024r\275\345\313a\275x\231\300>\201\334\"\276\236}\321\275H\346\257\276\202V\310>\225\375(>},\026\274E\235W\275\241\236\236<\245\220\222=\004\361\263\276\2620\240\274\265\036\240=\\:\336=\240\016\263\276\260\224\371\275p\235(=\311\\x;\226y\\\275\360\227\377=\275\364\200=rw/<@U\017=c\327\003=nV\267\273\354<Y\276\257$\260>3\207\311\274<&{\274h\b\031>\203\222\003\276\213\3649\275d\177\017\276\330\306\246\275'\314:>\203\024\253=S\331M\276\257\263S=b\221q\275\336\t\306=#^\005\2763)\256\274\344\255G>\224!\374=\374\360o\275\215#\031\276\322`J\276y6\360=\273\303;\275N\"N>M\022K\276\016\035C=1Fb\275u\017\220\275\0040\033>,\037\223=t\335\304\275\363\344\271=\320\341\260\275\2265\255=\310\"b=&\031\330\275;\314\000>>\347o\275\255\222\020=C\rU<\370d\307\275\373\306\206=I\345\373\274Y|\034>WY;\276O\240?>B\347\226\275\203\271\203\275W\243\313<\214\364\343\274Ey\033\276J4\214=\261\304\241\274-\271p>X\313l\276*_\366\275\260^\\=\237l1=\371\272\200=\002\373?\274*\3578<]Gv="
+	.size	.Lb45.data, 400
+
+	.type	.Lb45.buffer,@object            # @b45.buffer
+	.data
+	.p2align	4, 0x0
+.Lb45.buffer:
+	.quad	0                               # 0x0
+	.quad	0
+	.quad	.Lb45.data
+	.quad	1                               # 0x1
+	.byte	2                               # 0x2
+	.byte	32                              # 0x20
+	.half	1                               # 0x1
+	.word	1                               # 0x1
+	.quad	.Lb45.shape
+	.quad	0
+	.size	.Lb45.buffer, 56
+
+	.type	.L__unnamed_1,@object           # @0
+	.section	.rodata,"a",@progbits
+	.p2align	4, 0x0
+.L__unnamed_1:
+	.zero	32
+	.size	.L__unnamed_1, 32
+
+	.type	.Lstr,@object                   # @str
+	.p2align	5, 0x0
+.Lstr:
+	.asciz	"idw_halide_vec_"
+	.size	.Lstr, 16
+
+	.type	.L__unnamed_2,@object           # @1
+	.section	.data.rel.ro,"aw",@progbits
+	.p2align	4, 0x0
+.L__unnamed_2:
+	.quad	.Lstr
+	.word	2                               # 0x2
+	.word	2                               # 0x2
+	.byte	2                               # 0x2
+	.byte	32                              # 0x20
+	.half	1                               # 0x1
+	.zero	4
+	.quad	0
+	.quad	0
+	.quad	0
+	.quad	0
+	.quad	.L__unnamed_1
+	.size	.L__unnamed_2, 64
+
+	.type	.Lstr.4,@object                 # @str.4
+	.section	.rodata,"a",@progbits
+	.p2align	5, 0x0
+.Lstr.4:
+	.asciz	"riscv-64-linux-no_asserts-no_runtime-rvv-vector_bits_128"
+	.size	.Lstr.4, 57
+
+	.type	.Lidw_halide_vec__metadata_storage,@object # @idw_halide_vec__metadata_storage
+	.section	.data.rel.ro,"aw",@progbits
+	.p2align	4, 0x0
+.Lidw_halide_vec__metadata_storage:
+	.word	1                               # 0x1
+	.word	1                               # 0x1
+	.quad	.L__unnamed_2
+	.quad	.Lstr.4
+	.quad	.Lstr
+	.size	.Lidw_halide_vec__metadata_storage, 32
+
+	.ident	"clang version 16.0.0 (https://github.com/dkurt/llvm-rvv-071 b027aa1b59c9f53240bdc836f39656723fdf9df0)"
+	.ident	"clang version 16.0.0 (https://github.com/dkurt/llvm-rvv-071 b027aa1b59c9f53240bdc836f39656723fdf9df0)"
+	.ident	"clang version 16.0.0 (https://github.com/dkurt/llvm-rvv-071 b027aa1b59c9f53240bdc836f39656723fdf9df0)"
+	.ident	"clang version 16.0.0 (https://github.com/dkurt/llvm-rvv-071 b027aa1b59c9f53240bdc836f39656723fdf9df0)"
+	.ident	"clang version 16.0.0 (https://github.com/dkurt/llvm-rvv-071 b027aa1b59c9f53240bdc836f39656723fdf9df0)"
+	.ident	"clang version 16.0.0 (https://github.com/dkurt/llvm-rvv-071 b027aa1b59c9f53240bdc836f39656723fdf9df0)"
+	.section	".note.GNU-stack","",@progbits
diff --git a/include/algos.hpp b/include/algos.hpp
index d6c9f0c..e1f005f 100644
--- a/include/algos.hpp
+++ b/include/algos.hpp
@@ -42,6 +42,12 @@ void idw_ref(const uint8_t* src, uint8_t* dst, int height, int width, int* point
 
 void idw_halide(const uint8_t* src, uint8_t* dst, int height, int width, int* pointsBuf, float* weightsBuf);
 
+void idw_halide_parallel(const uint8_t* src, uint8_t* dst, int height, int width, int* pointsBuf, float* weightsBuf);
+
+void idw_halide_vec(const uint8_t* src, uint8_t* dst, int height, int width, int* pointsBuf, float* weightsBuf);
+
+void idw_halide_parallel_vec(const uint8_t* src, uint8_t* dst, int height, int width, int* pointsBuf, float* weightsBuf);
+
 void voxel_up(float* src, float* kernel, float* dst,
                     int inpChannels, int height, int width, int depth);
 void upscale(const std::vector<std::string> img_path, int width, int height);
diff --git a/perf/perf_main.cpp b/perf/perf_main.cpp
index 905450a..a20069f 100644
--- a/perf/perf_main.cpp
+++ b/perf/perf_main.cpp
@@ -14,7 +14,14 @@
 
 #include "algos.hpp"
 
-#include <Halide.h>
+// #ifdef __riscv
+//   #include <HalideBuffer.h>
+//   #include "idw.h"
+//   using namespace Halide::Runtime;
+// #else
+//   #include <Halide.h>
+//   using namespace Halide;
+// #endif
 
 // #include <iostream>
 
@@ -215,21 +222,16 @@ PERF_TEST(convolution_nhwc, halide) {
 // 300 elems (100 points)
 static int idwPoints[] = {0, 0, 72, 0, 213, 79, 0, 426, 60, 0, 640, 76, 0, 853, 128, 0, 1066, 67, 0, 1280, 65, 0, 1493, 64, 0, 1706, 60, 0, 1920, 61, 120, 0, 81, 120, 213, 79, 120, 426, 58, 120, 640, 132, 120, 853, 149, 120, 1066, 142, 120, 1280, 64, 120, 1493, 69, 120, 1706, 65, 120, 1920, 64, 240, 0, 75, 240, 213, 68, 240, 426, 140, 240, 640, 153, 240, 853, 145, 240, 1066, 132, 240, 1280, 152, 240, 1493, 125, 240, 1706, 66, 240, 1920, 58, 360, 0, 78, 360, 213, 60, 360, 426, 139, 360, 640, 168, 360, 853, 154, 360, 1066, 138, 360, 1280, 145, 360, 1493, 160, 360, 1706, 68, 360, 1920, 60, 480, 0, 77, 480, 213, 59, 480, 426, 165, 480, 640, 183, 480, 853, 166, 480, 1066, 142, 480, 1280, 123, 480, 1493, 155, 480, 1706, 144, 480, 1920, 60, 600, 0, 83, 600, 213, 65, 600, 426, 178, 600, 640, 184, 600, 853, 138, 600, 1066, 124, 600, 1280, 132, 600, 1493, 175, 600, 1706, 175, 600, 1920, 60, 720, 0, 86, 720, 213, 60, 720, 426, 182, 720, 640, 179, 720, 853, 152, 720, 1066, 140, 720, 1280, 115, 720, 1493, 156, 720, 1706, 172, 720, 1920, 65, 840, 0, 90, 840, 213, 69, 840, 426, 164, 840, 640, 186, 840, 853, 146, 840, 1066, 147, 840, 1280, 122, 840, 1493, 152, 840, 1706, 166, 840, 1920, 69, 960, 0, 85, 960, 213, 67, 960, 426, 186, 960, 640, 169, 960, 853, 162, 960, 1066, 147, 960, 1280, 124, 960, 1493, 149, 960, 1706, 157, 960, 1920, 69, 1080, 0, 85, 1080, 213, 70, 1080, 426, 193, 1080, 640, 192, 1080, 853, 155, 1080, 1066, 135, 1080, 1280, 120, 1080, 1493, 138, 1080, 1706, 128, 1080, 1920, 70};
 // 100 elems
-float idwWeights[] = {0.04053167, -0.07571915, 0.0025228, 0.1974248, -0.14461569, 0.30541419, -0.06167972, -0.04958807, 0.00050641, -0.02023852, -0.05910183, -0.05512609, 0.37617088, -0.15904428, -0.10229038, -0.34355378, 0.39128499, 0.16502984, -0.00916588, -0.05264022, 0.01936275, 0.07156495, -0.35144818, -0.01955447, 0.07818357, 0.10850975, -0.34972095, -0.12186563, 0.04116577, 0.00378971, -0.05382689, 0.12480152, 0.0629668, 0.01070963, 0.03499341, 0.03218783, -0.00559502, -0.21214646, 0.34402987, -0.0246006, -0.01532894, 0.14944613, -0.12848858, -0.04539923, -0.1401344, -0.08143395, 0.18241941, 0.08353522, -0.20102434, 0.05168503, -0.05897654, 0.09669851, -0.13024191, -0.02125988, 0.19499928, 0.12311092, -0.05857943, -0.14954968, -0.19763496, 0.1172914, -0.04584096, 0.20130274, -0.198312, 0.04763513, -0.05524272, -0.07034198, 0.15155036, 0.0718368, -0.09612551, 0.09076872, -0.0863682, 0.08457486, 0.05520895, -0.10551672, 0.12577908, -0.05857014, 0.03529613, 0.01300365, -0.09736055, 0.06580921, -0.03074898, 0.1528181, -0.18295799, 0.18713497, -0.07368328, -0.06431868, 0.02485816, -0.02782657, -0.15182979, 0.06845911, -0.01974711, 0.23508139, -0.23124445, -0.1202987, 0.05380124, 0.04331648, 0.06285662, -0.01171756, 0.01128749, 0.06012665};
+static float idwWeights[] = {0.04053167, -0.07571915, 0.0025228, 0.1974248, -0.14461569, 0.30541419, -0.06167972, -0.04958807, 0.00050641, -0.02023852, -0.05910183, -0.05512609, 0.37617088, -0.15904428, -0.10229038, -0.34355378, 0.39128499, 0.16502984, -0.00916588, -0.05264022, 0.01936275, 0.07156495, -0.35144818, -0.01955447, 0.07818357, 0.10850975, -0.34972095, -0.12186563, 0.04116577, 0.00378971, -0.05382689, 0.12480152, 0.0629668, 0.01070963, 0.03499341, 0.03218783, -0.00559502, -0.21214646, 0.34402987, -0.0246006, -0.01532894, 0.14944613, -0.12848858, -0.04539923, -0.1401344, -0.08143395, 0.18241941, 0.08353522, -0.20102434, 0.05168503, -0.05897654, 0.09669851, -0.13024191, -0.02125988, 0.19499928, 0.12311092, -0.05857943, -0.14954968, -0.19763496, 0.1172914, -0.04584096, 0.20130274, -0.198312, 0.04763513, -0.05524272, -0.07034198, 0.15155036, 0.0718368, -0.09612551, 0.09076872, -0.0863682, 0.08457486, 0.05520895, -0.10551672, 0.12577908, -0.05857014, 0.03529613, 0.01300365, -0.09736055, 0.06580921, -0.03074898, 0.1528181, -0.18295799, 0.18713497, -0.07368328, -0.06431868, 0.02485816, -0.02782657, -0.15182979, 0.06845911, -0.01974711, 0.23508139, -0.23124445, -0.1202987, 0.05380124, 0.04331648, 0.06285662, -0.01171756, 0.01128749, 0.06012665};
 
 PERF_TEST(idw, reference) {
     const int width = 1920;
     const int height = 1080;
-
-    // 300 elems (100 points)
-    int points[] = {0, 0, 72, 0, 213, 79, 0, 426, 60, 0, 640, 76, 0, 853, 128, 0, 1066, 67, 0, 1280, 65, 0, 1493, 64, 0, 1706, 60, 0, 1920, 61, 120, 0, 81, 120, 213, 79, 120, 426, 58, 120, 640, 132, 120, 853, 149, 120, 1066, 142, 120, 1280, 64, 120, 1493, 69, 120, 1706, 65, 120, 1920, 64, 240, 0, 75, 240, 213, 68, 240, 426, 140, 240, 640, 153, 240, 853, 145, 240, 1066, 132, 240, 1280, 152, 240, 1493, 125, 240, 1706, 66, 240, 1920, 58, 360, 0, 78, 360, 213, 60, 360, 426, 139, 360, 640, 168, 360, 853, 154, 360, 1066, 138, 360, 1280, 145, 360, 1493, 160, 360, 1706, 68, 360, 1920, 60, 480, 0, 77, 480, 213, 59, 480, 426, 165, 480, 640, 183, 480, 853, 166, 480, 1066, 142, 480, 1280, 123, 480, 1493, 155, 480, 1706, 144, 480, 1920, 60, 600, 0, 83, 600, 213, 65, 600, 426, 178, 600, 640, 184, 600, 853, 138, 600, 1066, 124, 600, 1280, 132, 600, 1493, 175, 600, 1706, 175, 600, 1920, 60, 720, 0, 86, 720, 213, 60, 720, 426, 182, 720, 640, 179, 720, 853, 152, 720, 1066, 140, 720, 1280, 115, 720, 1493, 156, 720, 1706, 172, 720, 1920, 65, 840, 0, 90, 840, 213, 69, 840, 426, 164, 840, 640, 186, 840, 853, 146, 840, 1066, 147, 840, 1280, 122, 840, 1493, 152, 840, 1706, 166, 840, 1920, 69, 960, 0, 85, 960, 213, 67, 960, 426, 186, 960, 640, 169, 960, 853, 162, 960, 1066, 147, 960, 1280, 124, 960, 1493, 149, 960, 1706, 157, 960, 1920, 69, 1080, 0, 85, 1080, 213, 70, 1080, 426, 193, 1080, 640, 192, 1080, 853, 155, 1080, 1066, 135, 1080, 1280, 120, 1080, 1493, 138, 1080, 1706, 128, 1080, 1920, 70};
-    // 100 elems
-    float weights[] = {0.04053167, -0.07571915, 0.0025228, 0.1974248, -0.14461569, 0.30541419, -0.06167972, -0.04958807, 0.00050641, -0.02023852, -0.05910183, -0.05512609, 0.37617088, -0.15904428, -0.10229038, -0.34355378, 0.39128499, 0.16502984, -0.00916588, -0.05264022, 0.01936275, 0.07156495, -0.35144818, -0.01955447, 0.07818357, 0.10850975, -0.34972095, -0.12186563, 0.04116577, 0.00378971, -0.05382689, 0.12480152, 0.0629668, 0.01070963, 0.03499341, 0.03218783, -0.00559502, -0.21214646, 0.34402987, -0.0246006, -0.01532894, 0.14944613, -0.12848858, -0.04539923, -0.1401344, -0.08143395, 0.18241941, 0.08353522, -0.20102434, 0.05168503, -0.05897654, 0.09669851, -0.13024191, -0.02125988, 0.19499928, 0.12311092, -0.05857943, -0.14954968, -0.19763496, 0.1172914, -0.04584096, 0.20130274, -0.198312, 0.04763513, -0.05524272, -0.07034198, 0.15155036, 0.0718368, -0.09612551, 0.09076872, -0.0863682, 0.08457486, 0.05520895, -0.10551672, 0.12577908, -0.05857014, 0.03529613, 0.01300365, -0.09736055, 0.06580921, -0.03074898, 0.1528181, -0.18295799, 0.18713497, -0.07368328, -0.06431868, 0.02485816, -0.02782657, -0.15182979, 0.06845911, -0.01974711, 0.23508139, -0.23124445, -0.1202987, 0.05380124, 0.04331648, 0.06285662, -0.01171756, 0.01128749, 0.06012665};
-
+    
     Mat dst(height, width, CV_8U);
 
     PERF_SAMPLE_BEGIN()
-        idw_ref(NULL, dst.ptr<uint8_t>(), height, width, points, weights);
+        idw_ref(NULL, dst.ptr<uint8_t>(), height, width, idwPoints, idwWeights);
     PERF_SAMPLE_END()
 
     SANITY_CHECK_NOTHING();
@@ -248,6 +250,45 @@ PERF_TEST(idw, halide) {
     SANITY_CHECK_NOTHING();
 }
 
+PERF_TEST(idw, halide_parallel) {
+    const int width = 1920;
+    const int height = 1080;
+
+    Mat dst(height, width, CV_8U);
+
+    PERF_SAMPLE_BEGIN()
+        idw_halide_parallel(NULL, dst.ptr<uint8_t>(), height, width, idwPoints, idwWeights);
+    PERF_SAMPLE_END()
+
+    SANITY_CHECK_NOTHING();
+}
+
+PERF_TEST(idw, halide_vec) {
+    const int width = 1920;
+    const int height = 1080;
+
+    Mat dst(height, width, CV_8U);
+
+    PERF_SAMPLE_BEGIN()
+        idw_halide_vec(NULL, dst.ptr<uint8_t>(), height, width, idwPoints, idwWeights);
+    PERF_SAMPLE_END()
+
+    SANITY_CHECK_NOTHING();
+}
+
+PERF_TEST(idw, halide_parallel_vec) {
+    const int width = 1920;
+    const int height = 1080;
+
+    Mat dst(height, width, CV_8U);
+
+    PERF_SAMPLE_BEGIN()
+        idw_halide_parallel_vec(NULL, dst.ptr<uint8_t>(), height, width, idwPoints, idwWeights);
+    PERF_SAMPLE_END()
+
+    SANITY_CHECK_NOTHING();
+}
+
 PERF_TEST(voxel_up, halide) {
     static const int ic = 4;
     static const int height = 100;
diff --git a/src/idw.cpp b/src/idw.cpp
index 622867e..2933703 100644
--- a/src/idw.cpp
+++ b/src/idw.cpp
@@ -9,7 +9,10 @@ using namespace std;
 
 #ifdef __riscv
   #include <HalideBuffer.h>
-  #include "idw.h"
+  #include "idw_halide.h"
+  #include "idw_halide_vec.h"
+  #include "idw_halide_parallel.h"
+  #include "idw_halide_parallel_vec.h"
   using namespace Halide::Runtime;
 #else
   #include <Halide.h>
@@ -27,9 +30,154 @@ void idw_halide(const uint8_t* src, uint8_t* dst, int height, int width, int* po
     Buffer<int> points(pointsBuf, {pointCount*3});
     Buffer<float> weights(weightsBuf, {pointCount});
 #ifdef __riscv
-    idw(mask);
+    idw_halide_(mask);
 #else
-    static Func f("idw");
+    static Func f("idw_halide_");
+
+    // try {
+    if (!f.defined()) {
+        Var x("x"), y("y");
+        RDom r(0, pointCount);
+        f(x, y) = 0.F;
+        Expr x0 = points(3*r+1);
+        Expr y0 = points(3*r);
+        Expr dx = x - x0;
+        Expr dy = y - y0;
+        Expr weight = weights(r);
+        f(x, y) += hypot(dx, dy) * weight;
+
+        // f.vectorize(r, 8);
+        const int factor = 4;
+        // f.update().atomic().vectorize(r, factor);
+        // f.update().parallel(x);
+
+        // Compile
+        Target target;
+        target.os = Target::OS::Linux;
+        target.arch = Target::Arch::RISCV;
+        target.bits = 64;
+        target.vector_bits = factor * sizeof(float) * 8;
+
+        // Tested XuanTie C906 has 128-bit vector unit
+        CV_Assert(target.vector_bits <= 128);
+
+        std::vector<Target::Feature> features;
+        features.push_back(Target::RVV);
+        features.push_back(Target::NoAsserts);
+        features.push_back(Target::NoRuntime);
+        target.set_features(features);
+
+        std::cout << target << std::endl;
+        // f.print_loop_nest();
+
+        // Dump AOT code
+        f.compile_to_header("idw_halide.h", {}, "idw_halide_", target);
+        f.compile_to_assembly("idw_halide.s", {}, "idw_halide_", target);
+    }
+    // }
+    // catch (Halide::Error &e) {
+    //     cout << e.what() << '\n';
+    // }
+
+    f.realize(mask);
+#endif
+
+    float maxVal = 193.0;
+    float minVal = 58.0;
+    float diff = maxVal - minVal;
+    for (int y = 0; y < height; y++) {
+        for (int x = 0; x < width; x++) {
+            dst[y*width + x] = (uint8_t) (255 * (maskBuf[y*width + x] - minVal) / diff);
+        }
+    }
+    delete[] maskBuf;
+}
+
+void idw_halide_parallel(const uint8_t* src, uint8_t* dst, int height, int width, int* pointsBuf, float* weightsBuf) {
+    float* maskBuf = new float[height * width]();
+
+    Buffer<float> mask(maskBuf, {width, height});
+    Buffer<uint8_t> output(dst, {width, height});
+
+    Buffer<int> points(pointsBuf, {pointCount*3});
+    Buffer<float> weights(weightsBuf, {pointCount});
+#ifdef __riscv
+    idw_halide_parallel_(mask);
+#else
+    static Func f("idw_halide_parallel_");
+
+    try {
+    if (!f.defined()) {
+        Var x("x"), y("y"), x_outer("x_outer"), y_outer("y_outer"), x_inner("x_inner"), y_inner("y_inner"), tile_index("tile_index");
+        RDom r(0, pointCount);
+        f(x, y) = 0.F;
+        Expr x0 = points(3*r+1);
+        Expr y0 = points(3*r);
+        Expr dx = x - x0;
+        Expr dy = y - y0;
+        Expr weight = weights(r);
+        f(x, y) += hypot(dx, dy) * weight;
+
+        f.tile(x, y, x_outer, y_outer, x_inner, y_inner, width / 4, height / 4).fuse(x_outer, y_outer, tile_index).parallel(tile_index);
+        f.update().tile(x, y, x_outer, y_outer, x_inner, y_inner, width / 4, height / 4).fuse(x_outer, y_outer, tile_index).parallel(tile_index);
+        
+        const int factor = 4;
+        // f.update().atomic().vectorize(r, factor);
+
+        // Compile
+        Target target;
+        target.os = Target::OS::Linux;
+        target.arch = Target::Arch::RISCV;
+        target.bits = 64;
+        target.vector_bits = factor * sizeof(float) * 8;
+
+        // Tested XuanTie C906 has 128-bit vector unit
+        CV_Assert(target.vector_bits <= 128);
+
+        std::vector<Target::Feature> features;
+        features.push_back(Target::RVV);
+        features.push_back(Target::NoAsserts);
+        features.push_back(Target::NoRuntime);
+        target.set_features(features);
+
+        std::cout << target << std::endl;
+        // f.print_loop_nest();
+
+        // Dump AOT code
+        f.compile_to_header("idw_halide_parallel.h", {}, "idw_halide_parallel_", target);
+        f.compile_to_assembly("idw_halide_parallel.s", {}, "idw_halide_parallel_", target);
+    }
+
+    f.realize(mask);
+    }
+    catch (Halide::Error &e) {
+        cout << e.what() << '\n';
+    }
+#endif
+
+    float maxVal = 193.0;
+    float minVal = 58.0;
+    float diff = maxVal - minVal;
+    for (int y = 0; y < height; y++) {
+        for (int x = 0; x < width; x++) {
+            dst[y*width + x] = (uint8_t) (255 * (maskBuf[y*width + x] - minVal) / diff);
+        }
+    }
+    delete[] maskBuf;
+}
+
+void idw_halide_vec(const uint8_t* src, uint8_t* dst, int height, int width, int* pointsBuf, float* weightsBuf) {
+    float* maskBuf = new float[height * width]();
+
+    Buffer<float> mask(maskBuf, {width, height});
+    Buffer<uint8_t> output(dst, {width, height});
+
+    Buffer<int> points(pointsBuf, {pointCount*3});
+    Buffer<float> weights(weightsBuf, {pointCount});
+#ifdef __riscv
+    idw_halide_vec_(mask);
+#else
+    static Func f("idw_halide_vec_");
 
     // try {
     if (!f.defined()) {
@@ -46,6 +194,7 @@ void idw_halide(const uint8_t* src, uint8_t* dst, int height, int width, int* po
         // f.vectorize(r, 8);
         const int factor = 4;
         f.update().atomic().vectorize(r, factor);
+        // f.update().parallel(x);
 
         // Compile
         Target target;
@@ -64,11 +213,11 @@ void idw_halide(const uint8_t* src, uint8_t* dst, int height, int width, int* po
         target.set_features(features);
 
         std::cout << target << std::endl;
-        f.print_loop_nest();
+        // f.print_loop_nest();
 
         // Dump AOT code
-        f.compile_to_header("idw.h", {}, "idw", target);
-        f.compile_to_assembly("idw.s", {}, "idw", target);
+        f.compile_to_header("idw_halide_vec.h", {}, "idw_halide_vec_", target);
+        f.compile_to_assembly("idw_halide_vec.s", {}, "idw_halide_vec_", target);
     }
     // }
     // catch (Halide::Error &e) {
@@ -89,6 +238,79 @@ void idw_halide(const uint8_t* src, uint8_t* dst, int height, int width, int* po
     delete[] maskBuf;
 }
 
+void idw_halide_parallel_vec(const uint8_t* src, uint8_t* dst, int height, int width, int* pointsBuf, float* weightsBuf) {
+    float* maskBuf = new float[height * width]();
+
+    Buffer<float> mask(maskBuf, {width, height});
+    Buffer<uint8_t> output(dst, {width, height});
+
+    Buffer<int> points(pointsBuf, {pointCount*3});
+    Buffer<float> weights(weightsBuf, {pointCount});
+#ifdef __riscv
+    idw_halide_parallel_vec_(mask);
+#else
+    static Func f("idw_halide_parallel_vec_");
+
+    try {
+    if (!f.defined()) {
+        Var x("x"), y("y"), x_outer("x_outer"), y_outer("y_outer"), x_inner("x_inner"), y_inner("y_inner"), tile_index("tile_index");
+        RDom r(0, pointCount);
+        f(x, y) = 0.F;
+        Expr x0 = points(3*r+1);
+        Expr y0 = points(3*r);
+        Expr dx = x - x0;
+        Expr dy = y - y0;
+        Expr weight = weights(r);
+        f(x, y) += hypot(dx, dy) * weight;
+
+        f.tile(x, y, x_outer, y_outer, x_inner, y_inner, width / 4, height / 4).fuse(x_outer, y_outer, tile_index).parallel(tile_index);
+        f.update().tile(x, y, x_outer, y_outer, x_inner, y_inner, width / 4, height / 4).fuse(x_outer, y_outer, tile_index).parallel(tile_index);
+        
+        const int factor = 4;
+        f.update().atomic().vectorize(r, factor);
+
+        // Compile
+        Target target;
+        target.os = Target::OS::Linux;
+        target.arch = Target::Arch::RISCV;
+        target.bits = 64;
+        target.vector_bits = factor * sizeof(float) * 8;
+
+        // Tested XuanTie C906 has 128-bit vector unit
+        CV_Assert(target.vector_bits <= 128);
+
+        std::vector<Target::Feature> features;
+        features.push_back(Target::RVV);
+        features.push_back(Target::NoAsserts);
+        features.push_back(Target::NoRuntime);
+        target.set_features(features);
+
+        std::cout << target << std::endl;
+        // f.print_loop_nest();
+
+        // Dump AOT code
+        f.compile_to_header("idw_halide_parallel_vec.h", {}, "idw_halide_parallel_vec_", target);
+        f.compile_to_assembly("idw_halide_parallel_vec.s", {}, "idw_halide_parallel_vec_", target);
+    }
+
+    f.realize(mask);
+    }
+    catch (Halide::Error &e) {
+        cout << e.what() << '\n';
+    }
+#endif
+
+    float maxVal = 193.0;
+    float minVal = 58.0;
+    float diff = maxVal - minVal;
+    for (int y = 0; y < height; y++) {
+        for (int x = 0; x < width; x++) {
+            dst[y*width + x] = (uint8_t) (255 * (maskBuf[y*width + x] - minVal) / diff);
+        }
+    }
+    delete[] maskBuf;
+}
+
 void idw_ref(const uint8_t* src, uint8_t* dst, int height, int width, int* points, float* weights) {
     float* mask = new float[height * width]();
 
diff --git a/test/test_main.cpp b/test/test_main.cpp
index 9fadfbd..0e89acf 100644
--- a/test/test_main.cpp
+++ b/test/test_main.cpp
@@ -167,6 +167,8 @@ TEST(convolution_nhwc, halide) {
 TEST(idw, halide) {
     Mat src(height, width, CV_8U); 
     Mat dst(height, width, CV_8U), cl_dst(height, width, CV_8UC3), dst_h(height, width, CV_8U), cl_dst_h(height, width, CV_8UC3);
+    Mat dst_hp(height, width, CV_8U), cl_dst_hp(height, width, CV_8UC3), dst_hv(height, width, CV_8U), cl_dst_hv(height, width, CV_8UC3), 
+        dst_hpv(height, width, CV_8U), cl_dst_hpv(height, width, CV_8UC3);
     // randu(src, 0, 256);
 
     // 300 elems (100 points)
@@ -176,13 +178,25 @@ TEST(idw, halide) {
 
     idw_ref(NULL, dst.ptr<uint8_t>(), height, width, points, weights);
     idw_halide(NULL, dst_h.ptr<uint8_t>(), height, width, points, weights);
+    idw_halide_parallel(NULL, dst_hp.ptr<uint8_t>(), height, width, points, weights);
+    idw_halide_vec(NULL, dst_hv.ptr<uint8_t>(), height, width, points, weights);
+    idw_halide_parallel_vec(NULL, dst_hpv.ptr<uint8_t>(), height, width, points, weights);
 
     applyColorMap(dst, cl_dst, COLORMAP_JET);
     applyColorMap(dst_h, cl_dst_h, COLORMAP_JET);
+    applyColorMap(dst_hp, cl_dst_hp, COLORMAP_JET);
+    applyColorMap(dst_hv, cl_dst_hv, COLORMAP_JET);
+    applyColorMap(dst_hpv, cl_dst_hpv, COLORMAP_JET);
 
     // imwrite("src.png", src);
     imwrite("res.png", dst);
     imwrite("cres.png", cl_dst);
     imwrite("res_halide.png", dst_h);
     imwrite("cres_halide.png", cl_dst_h);
+    imwrite("res_halide_par.png", dst_hp);
+    imwrite("cres_halide_par.png", cl_dst_hp);
+    imwrite("res_halide_vec.png", dst_hv);
+    imwrite("cres_halide_vec.png", cl_dst_hv);
+    imwrite("res_halide_parvec.png", dst_hpv);
+    imwrite("cres_halide_parvec.png", cl_dst_hpv);
 }
\ No newline at end of file