Experiment: opt-in SDR-to-EDR presentation

.github/workflows/tests.yml

@@ -58,7 +58,7 @@ jobs:
           make GAME="$GAME" -j3 t_perf_order
           ./t_perf_order --cpu 2>&1 | tee ../../../logs/t_perf_order.log
           make GAME="$GAME" -j3 perf-order-tsan 2>&1 | tee ../../../logs/t_perf_order-tsan.log
-          tests="t_gfx_suspend t_capture t_dirty_slots t_dirty_race t_tex_layout t_pace t_fstep t_link_ticks t_cpu_tex t_hmap t_cover t_pso_skey t_refl t_rt_range_ro t_present_rescue t_gpuprof t_sparse_layout t_timestamp t_rt_depth_ro t_rt_range_shape t_ub t_pso_pending t_vk_desc"
+          tests="t_gfx_suspend t_capture t_dirty_slots t_dirty_race t_tex_layout t_pace t_fstep t_link_ticks t_edr_policy t_cpu_tex t_hmap t_cover t_pso_skey t_refl t_rt_range_ro t_present_rescue t_gpuprof t_sparse_layout t_timestamp t_rt_depth_ro t_rt_range_shape t_ub t_pso_pending t_vk_desc"
           make GAME="$GAME" -j3 $tests 2>&1 | tee ../../../logs/gfx-build.log
           failed=0
           for t in $tests; do

dm/gfx/nvn/nvn_edr.h

@@ -0,0 +1,22 @@
+// Experimental SDR-to-EDR presentation policy. This expands an SDR picture's
+// highlights; it does not recover a game's original scene-referred HDR signal.
+#pragma once
+#include <math.h>
+#include <stdlib.h>
+#include <string.h>
+
+static inline int nvn_edr_requested(void) {
+    const char *value = getenv("DM_PRESENT_EDR");
+    return value && !strcmp(value, "1");
+}
+static inline float nvn_edr_headroom(int requested, int active, float current, float potential) {
+    if (!requested || !active || !isfinite(current) || !isfinite(potential)) return 1;
+    return fmaxf(1, fminf(2, fminf(current, potential)));
+}
+
+// The input is the same stored SDR bytes the stock present pass samples.
+// Decode once to extended-linear sRGB, keep shadows/midtones, and expand only
+// luminance above 0.5. A common RGB gain preserves the highlight's chromaticity.
+#define NVN_EDR_MSL \
+    "float3 edr_decode(float3 x) { x=clamp(x,0.0f,1.0f); return select(pow((x+0.055f)/1.055f,float3(2.4f)),x/12.92f,x<=0.04045f); }\n" \
+    "float3 edr_map(float3 rgb,float headroom,bool alreadyLinear) { float3 l=alreadyLinear?clamp(rgb,0.0f,1.0f):edr_decode(rgb); float y=dot(l,float3(0.2126f,0.7152f,0.0722f)); float w=smoothstep(0.5f,1.0f,y); return l*(1.0f+(headroom-1.0f)*w); }\n"

dm/gfx/nvn/nvn_mtl.m

@@ -32,6 +32,7 @@
 #include "dmsh_chain.h"   // DMSH_CHAIN: the translation chain version of the B shader library expected by this source tree, one for every game (beside this file, tools/dmsh_chain.py --header)
 #include "game_profile.h"   // game profile constants (profile id, the profile's switches)
 #include "../../host/front.h"
+#include "nvn_edr.h"
 // Texture slot table per stage: 0-127 the combined handles (nvnCommandBufferBindTexture), 128-255 the separate texture binds (NVN_SEPTEX_SLOT0, nvn_obj.h)
 #define NTEX 256
 // The NVN driver's constant segments c0 and c2 (byte offsets, slot counts) and the binding kinds of the shader reflection have one definition, which the shader build reads too (dmsh_build.py through dmsh_abi.py)
@@ -454,7 +455,7 @@ void mtl_init(void) {
     g_present_args = [g_dev newArgumentTableWithDescriptor:pad error:&ate];
     if (!g_present_args) MDIE("failed to create the present argument table: %s", ate.localizedDescription.UTF8String);
     for (int i = 0; i < 3; i++) {
-        g_present_uv[i] = [g_dev newBufferWithLength:sizeof(float) * 8 options:MTLResourceStorageModeShared];
+        g_present_uv[i] = [g_dev newBufferWithLength:sizeof(float) * 12 options:MTLResourceStorageModeShared];
         if (!g_present_uv[i]) MDIE("failed to create present UV buffer[%d]", i);
         res_add(g_present_uv[i]);
     }
@@ -10060,7 +10061,7 @@ void mtl_exec(const uint8_t *p) {
 }
 
 // ---------- Present: draw the swapchain texture to the CAMetalLayer; optional screenshot ----------
-static id<MTLRenderPipelineState> g_blit_pso; static MTLPixelFormat g_blit_fmt;
+static id<MTLRenderPipelineState> g_blit_pso; static MTLPixelFormat g_blit_fmt; static int g_blit_edr;
 static id<MTLSamplerState> g_blit_samp;
 static uint64_t g_present_n;
 uint64_t mtl_present_count(void) { return g_present_n; }   // Actual present count (heat attribution DM_HEAT computes the present frame rate)
@@ -10397,23 +10398,34 @@ int mtl_display_hz(void) {
 static _Atomic uint64_t g_present_stall_ns;   // For unit tests: an extra pause before each present on the present thread (simulating a drawable that stays unavailable)
 static CAMetalLayer *g_present_layer_resident;
 void mtl_present_test_stall(uint64_t ns) { g_present_stall_ns = ns; }
+__attribute__((weak)) float front_present_edr_headroom(void) { return 1; }   // iOS supplies scene headroom; Mac output remains SDR
+static int g_edr_mode = -1;
+static int present_edr(MTLPixelFormat format) {
+    if (g_edr_mode < 0) g_edr_mode = nvn_edr_requested();
+    return g_edr_mode && format == MTLPixelFormatRGBA16Float;
+}
 static void present_pipeline(MTLPixelFormat format) {
-    if (!g_blit_pso || g_blit_fmt != format) {
+    int edr = present_edr(format);
+    if (!g_blit_pso || g_blit_fmt != format || g_blit_edr != edr) {
         static const char *src_msl =
             "#include <metal_stdlib>\nusing namespace metal;\n"
             "struct V { float4 p [[position]]; float2 uv; };\n"
             "vertex V vs(uint id [[vertex_id]]) { float2 uv = float2((id << 1) & 2, id & 2); V o; o.p = float4(uv * float2(2, -2) + float2(-1, 1), 0, 1); o.uv = uv; return o; }\n"
-            "struct Crop { float4 rect; float4 edge; };\n"
-            "fragment float4 fs(V in [[stage_in]], texture2d<float> t [[texture(0)]], sampler s [[sampler(0)]], constant Crop& c [[buffer(0)]]) { return t.sample(s, clamp(in.uv * c.rect.zw + c.rect.xy, c.edge.xy, c.edge.zw)); }\n";
+            "struct Crop { float4 rect; float4 edge; float4 edr; };\n"
+            NVN_EDR_MSL
+            "fragment float4 fs(V in [[stage_in]], texture2d<float> t [[texture(0)]], sampler s [[sampler(0)]], constant Crop& c [[buffer(0)]]) { float4 color=t.sample(s, clamp(in.uv * c.rect.zw + c.rect.xy, c.edge.xy, c.edge.zw));\n"
+            "#if EDR_PRESENT\n color.rgb=edr_map(color.rgb,c.edr.x,c.edr.y>0.0f);\n#endif\n return color; }\n";
         NSError *e = nil;
-        id<MTLLibrary> lib = [g_dev newLibraryWithSource:@(src_msl) options:nil error:&e];
+        MTLCompileOptions *options = [MTLCompileOptions new]; options.preprocessorMacros = @{@"EDR_PRESENT":@(edr)};
+        id<MTLLibrary> lib = [g_dev newLibraryWithSource:@(src_msl) options:options error:&e];
         if (!lib) MDIE("present shader compile failed %s", e.localizedDescription.UTF8String);
         MTLRenderPipelineDescriptor *pd = [MTLRenderPipelineDescriptor new];
         pd.vertexFunction = [lib newFunctionWithName:@"vs"]; pd.fragmentFunction = [lib newFunctionWithName:@"fs"];
         pd.colorAttachments[0].pixelFormat = format;
         g_blit_pso = [g_dev newRenderPipelineStateWithDescriptor:pd error:&e];
         if (!g_blit_pso) MDIE("present pipeline failed %s", e.localizedDescription.UTF8String);
         g_blit_fmt = format;
+        g_blit_edr = edr;
         MTLSamplerDescriptor *sd = [MTLSamplerDescriptor new]; sd.minFilter = sd.magFilter = MTLSamplerMinMagFilterLinear; sd.supportArgumentBuffers = YES;
         g_blit_samp = [g_dev newSamplerStateWithDescriptor:sd];
     }
@@ -10473,9 +10485,12 @@ static int present_encode(id<MTLCommandBuffer> cb, id<MTLTexture> src, id<MTLTex
     rp.colorAttachments[0].storeAction = MTLStoreActionStore;
     id<MTLRenderCommandEncoder> en = [cb renderCommandEncoderWithDescriptor:rp];
     [en setViewport:vp];
-    [en setRenderPipelineState:g_blit_pso]; [en setFragmentTexture:present_raw_src(src) atIndex:0]; [en setFragmentSamplerState:g_blit_samp atIndex:0];
-    float uv[8] = {(float)cx / src.width, (float)cy / src.height, (float)cw / src.width, (float)ch / src.height,
-                   (cx + 0.5f) / src.width, (cy + 0.5f) / src.height, (cx + cw - 0.5f) / src.width, (cy + ch - 0.5f) / src.height};
+    id<MTLTexture> raw = present_raw_src(src);
+    [en setRenderPipelineState:g_blit_pso]; [en setFragmentTexture:raw atIndex:0]; [en setFragmentSamplerState:g_blit_samp atIndex:0];
+    float uv[12] = {(float)cx / src.width, (float)cy / src.height, (float)cw / src.width, (float)ch / src.height,
+                   (cx + 0.5f) / src.width, (cy + 0.5f) / src.height, (cx + cw - 0.5f) / src.width, (cy + ch - 0.5f) / src.height,
+                   present_edr(dst.pixelFormat) ? front_present_edr_headroom() : 1,
+                   raw.pixelFormat == MTLPixelFormatRGBA8Unorm_sRGB || raw.pixelFormat == MTLPixelFormatBGRA8Unorm_sRGB, 0, 0};
     if (atomic_load_explicit(&g_origin_ll, memory_order_relaxed)) { uv[1] += uv[3]; uv[3] = -uv[3]; }   // lower-left origin: the window image is stored bottom-up
     [en setFragmentBytes:uv length:sizeof uv atIndex:0];
     [en drawPrimitives:MTLPrimitiveTypeTriangle vertexStart:0 vertexCount:3];
@@ -10557,7 +10572,11 @@ static int present_encode_m4(id<MTL4CommandBuffer> cb, id<MTLTexture> src, id<MT
     double dw = dst.width, dh = dst.height;
     MTLViewport vp = present_viewport(r[2], r[3], dw, dh);
     int full = vp.originX == 0 && vp.originY == 0 && vp.width == dw && vp.height == dh;
-    id<MTLTexture> tex = present_raw_src(src), up = fx ? fx_encode(fx, cb, tex, r, vp, dst.pixelFormat, fx == &g_fx_present) : nil;
+    // MetalFX continues to process SDR perceptual bytes. The final present pass
+    // decodes/expands once; screenshots target BGRA8 and stay SDR even when EDR
+    // is requested. Do not feed linear EDR values into the perceptual scaler.
+    MTLPixelFormat upscale_format = present_edr(dst.pixelFormat) ? MTLPixelFormatBGRA8Unorm : dst.pixelFormat;
+    id<MTLTexture> tex = present_raw_src(src), up = fx ? fx_encode(fx, cb, tex, r, vp, upscale_format, fx == &g_fx_present) : nil;
     if (up) { tex = up; r[0] = r[1] = 0; r[2] = up.width; r[3] = up.height; }   // the scaled picture, the viewport's size: drawn 1:1
     MTL4RenderPassDescriptor *rp = [MTL4RenderPassDescriptor new];
     rp.colorAttachments[0].texture = dst;
@@ -10573,6 +10592,8 @@ static int present_encode_m4(id<MTL4CommandBuffer> cb, id<MTLTexture> src, id<MT
     float *uv = uvb.contents; NSUInteger tw = tex.width, th = tex.height, cx = r[0], cy = r[1], cw = r[2], ch = r[3];
     uv[0] = (float)cx / tw; uv[1] = (float)cy / th; uv[2] = (float)cw / tw; uv[3] = (float)ch / th;
     uv[4] = (cx + 0.5f) / tw; uv[5] = (cy + 0.5f) / th; uv[6] = (cx + cw - 0.5f) / tw; uv[7] = (cy + ch - 0.5f) / th;
+    uv[8] = present_edr(dst.pixelFormat) ? front_present_edr_headroom() : 1;
+    uv[9] = tex.pixelFormat == MTLPixelFormatRGBA8Unorm_sRGB || tex.pixelFormat == MTLPixelFormatBGRA8Unorm_sRGB; uv[10] = uv[11] = 0;
     if (atomic_load_explicit(&g_origin_ll, memory_order_relaxed)) { uv[1] += uv[3]; uv[3] = -uv[3]; }   // lower-left origin: the window image is stored bottom-up
     [g_present_args setAddress:uvb.gpuAddress atIndex:0];
     [en setViewport:vp]; [en setRenderPipelineState:g_blit_pso];
@@ -10584,7 +10605,7 @@ static int present_encode_m4(id<MTL4CommandBuffer> cb, id<MTLTexture> src, id<MT
 // returns whether it was submitted. Completion reports the encoding kind, or PRESENT_FAILED on a GPU error. Resources remain resident and retained until GPU completion.
 static int present_offscreen_m4(id<MTLTexture> src, const int crop[4], id<MTLTexture> out, void (^done)(int)) {
     id<MTL4CommandAllocator> al = [g_dev newCommandAllocator]; id<MTL4CommandBuffer> cb = [g_dev newCommandBuffer];
-    id<MTLBuffer> uvb = [g_dev newBufferWithLength:sizeof(float) * 8 options:MTLResourceStorageModeShared];
+    id<MTLBuffer> uvb = [g_dev newBufferWithLength:sizeof(float) * 12 options:MTLResourceStorageModeShared];
     if (!al || !cb || !uvb) { MLOG("[RED] offscreen present: no command allocator or buffer"); return 0; }
     id<MTLTexture> raw = present_raw_src(src);
     fx_state fx = {0};

dm/gfx/tests/Makefile

@@ -107,6 +107,10 @@ check_t_perf_order: t_perf_order
 	DM_NVN_FAST_WAKE=0 DM_NVN_SLICE_MERGE=0 DM_NVN_CS_HOIST=0 MTL_DEBUG_LAYER=1 ./t_perf_order
 t_link_ticks: t_link_ticks.m $(MTL_T_DEP)
 	clang $(CFLAGS) -fobjc-arc -o $@ t_link_ticks.m $(filter-out ../nvn/hle_nvn.c,$(MTL_T_SRC)) $(FW)
+t_edr_policy: t_edr_policy.c ../nvn/nvn_edr.h
+	clang $(CFLAGS) -o $@ t_edr_policy.c
+t_edr_present: t_edr_present.m edr_probe_host.c ../nvn/nvn_edr.h $(MTL_T_DEP)
+	clang $(CFLAGS) -fobjc-arc -o $@ t_edr_present.m edr_probe_host.c $(MTL_T_SRC) $(FW)
 t_cpu_tex: t_cpu_tex.m $(MTL_T_DEP)
 	clang $(CFLAGS) -fobjc-arc -o $@ t_cpu_tex.m $(filter-out ../nvn/hle_nvn.c,$(MTL_T_SRC)) $(FW)
 .PHONY: check_t_cpu_tex

dm/gfx/tests/edr_probe_host.c

@@ -0,0 +1,4 @@
+// Synthetic headroom for the GPU color-transform test, never a display claim.
+#include <stdatomic.h>
+_Atomic float edr_probe_headroom = 2;
+float front_present_edr_headroom(void) { return atomic_load(&edr_probe_headroom); }

dm/gfx/tests/t_edr_policy.c

@@ -0,0 +1,20 @@
+#include "../nvn/nvn_edr.h"
+#include <stdio.h>
+static int checks, failed;
+#define CHECK(name, ok) do { checks++; int pass = !!(ok); failed += !pass; printf("%s: %s\n", name, pass ? "PASS" : "FAIL"); } while (0)
+int main(void) {
+    unsetenv("DM_PRESENT_EDR"); CHECK("EDR is off by default", !nvn_edr_requested());
+    setenv("DM_PRESENT_EDR", "0", 1); CHECK("zero keeps SDR", !nvn_edr_requested());
+    setenv("DM_PRESENT_EDR", "1", 1); CHECK("one explicitly opts in", nvn_edr_requested());
+    setenv("DM_PRESENT_EDR", "true", 1); CHECK("other values do not accidentally opt in", !nvn_edr_requested());
+    CHECK("off keeps SDR headroom", nvn_edr_headroom(0, 1, 8, 8) == 1);
+    CHECK("an inactive scene keeps SDR", nvn_edr_headroom(1, 0, 8, 8) == 1);
+    CHECK("SDR-only display keeps SDR", nvn_edr_headroom(1, 1, 1, 1) == 1);
+    CHECK("current headroom limits output", fabsf(nvn_edr_headroom(1, 1, 1.2f, 8) - 1.2f) < 0.00001f);
+    CHECK("potential headroom also limits output", nvn_edr_headroom(1, 1, 8, 1.5f) == 1.5f);
+    CHECK("experimental boost is capped at two", nvn_edr_headroom(1, 1, 8, 8) == 2);
+    CHECK("invalid headroom falls back to SDR", nvn_edr_headroom(1, 1, NAN, 8) == 1);
+    CHECK("invalid potential falls back to SDR", nvn_edr_headroom(1, 1, 2, INFINITY) == 1);
+    printf("edr policy: %d checks, %d failed\n", checks, failed);
+    return failed != 0;
+}