From aeb58e09e000a030db62ac8931a8ac758bd3b478 Mon Sep 17 00:00:00 2001
From: seedds <seedds@gmail.com>
Date: Wed, 11 Nov 2026 23:54:30 +0100
Subject: [PATCH] shader: report a one-pixel fragment size

Report Ampere special register 0x84 as a 1 x 1 fragment and make other unknown special registers visible in translator notes. Cover S2R, S2UR, byte extraction, and the diagnostic with Linux SPIR-V tests.\n\nAdapted from seedds/nxnative#2373.
---
 dm/gfx/tests/test_dmsh_fragment_size.py       | 102 ++++++++++++++++++
 .../Instructions/InstEmitAmpereAlu.cs         |  19 +++-
 2 files changed, 120 insertions(+), 1 deletion(-)
 create mode 100644 dm/gfx/tests/test_dmsh_fragment_size.py

diff --git a/dm/gfx/tests/test_dmsh_fragment_size.py b/dm/gfx/tests/test_dmsh_fragment_size.py
new file mode 100644
index 00000000..2c85cf8c
--- /dev/null
+++ b/dm/gfx/tests/test_dmsh_fragment_size.py
@@ -0,0 +1,102 @@
+#!/usr/bin/env python3
+"""Ampere S2UR special register 0x84 through the offline translator.
+
+Bytes 1 and 2 are the variable-rate-shading fragment width and height.  The
+host shades one pixel per fragment, so the register must be 0x00010100 and its
+width-times-height product must fold to one.  Needs DMSH_TEST_XLATE and the
+SPIR-V tools on PATH.
+"""
+import json
+import os
+import re
+import shutil
+import struct
+import subprocess
+import tempfile
+import unittest
+from pathlib import Path
+
+
+EXIT = (0x000fea0003800000, 0x000000000000794d)
+S2UR = (0x000ea20000008400, 0x00000000000479c3)       # S2UR UR4, SR 0x84
+S2UR_UNKNOWN = (0x000ea20000008500, S2UR[1])          # S2UR UR4, SR 0x85
+S2R = (0x000e260000008400, 0x0000000000067919)        # S2R R6, SR 0x84
+BYTE2 = (0x004fe2000800003f, 0x0000777204057896)      # UPRMT UR5, UR4, 0x7772
+BYTE1 = (0x000fc4000800003f, 0x0000777104047896)      # UPRMT UR4, UR4, 0x7771
+PRODUCT = (0x000fe2000f8e023f, 0x00000004050472a4)    # UIMAD UR4, UR5, UR4, URZ
+MOV_UR4 = (0x000fe20000000f00, 0x0000000400067c02)    # MOV R6, UR4
+STORE = (0x000fe8000000c800, 0x0000080609008388)      # STS.U32 [R9 * 16 + 8], R6
+
+
+def inst(hi, lo):
+    return struct.pack('<QQ', lo, hi)
+
+
+def waiting(i):
+    return (i[0] | 0x3f << 52, i[1])
+
+
+def s2r_tid(rd):
+    return (0x000e260000002100, 0x0000000000007919 | rd << 16)
+
+
+@unittest.skipUnless(
+    os.environ.get('DMSH_TEST_XLATE') and shutil.which('spirv-val') and shutil.which('spirv-dis'),
+    'set DMSH_TEST_XLATE; needs spirv-val and spirv-dis',
+)
+class FragmentSize(unittest.TestCase):
+    def translate(self, *body, move_uniform=True):
+        with tempfile.TemporaryDirectory(prefix='ampere_fragment_size_') as d:
+            work = Path(d)
+            code = inst(*s2r_tid(9))
+            tail = (MOV_UR4, STORE) if move_uniform else (STORE,)
+            code += b''.join(inst(*waiting(i)) for i in (*body, *tail))
+            code += inst(*EXIT)
+            (work / 'k.bin').write_bytes(code)
+            job = [dict(
+                key='k', stage='cs', arch='sm86', code=str(work / 'k.bin'),
+                tex={}, img={}, attr={}, fout={}, local=[32, 1, 1], shared=0x1000,
+                debug=True, nvn2_cs=[0, 0, 3, 0x40, 0],
+            )]
+            (work / 'jobs.json').write_text(json.dumps(job))
+            r = subprocess.run(
+                [os.environ['DMSH_TEST_XLATE'], str(work / 'jobs.json'), d],
+                capture_output=True, text=True,
+            )
+            self.log = r.stdout + r.stderr
+            self.assertEqual(r.returncode, 0, r.stdout + r.stderr)
+            spv = work / 'k.spv'
+            self.assertTrue(spv.exists(), r.stdout + r.stderr)
+            self.result = json.loads((work / 'k.res.json').read_text())
+            v = subprocess.run(
+                ['spirv-val', '--target-env', 'vulkan1.3', str(spv)],
+                capture_output=True, text=True,
+            )
+            self.assertEqual(v.returncode, 0, v.stdout + v.stderr)
+            return subprocess.run(
+                ['spirv-dis', '--raw-id', str(spv)],
+                capture_output=True, text=True, check=True,
+            ).stdout
+
+    def test_register_value(self):
+        dis = self.translate(S2UR)
+        self.assertRegex(dis, r'OpConstant %\d+ 65792\b')
+
+    def test_nonuniform_register_value(self):
+        dis = self.translate(S2R, move_uniform=False)
+        self.assertRegex(dis, r'OpConstant %\d+ 65792\b')
+
+    def test_one_pixel_fragment(self):
+        dis = self.translate(S2UR, BYTE2, BYTE1, PRODUCT)
+        one = re.search(r'^\s*(%\d+) = OpConstant %\d+ 1$', dis, re.MULTILINE)
+        self.assertIsNotNone(one, dis)
+        self.assertRegex(dis, rf'OpStore %\d+ {re.escape(one.group(1))}\b')
+
+    def test_unknown_register_is_diagnostic(self):
+        self.translate(S2UR_UNKNOWN)
+        message = 'S2R of special register 0x85, which the translator does not know: read as 0'
+        self.assertIn(message, json.dumps(self.result), self.result)
+
+
+if __name__ == '__main__':
+    unittest.main()
diff --git a/dm/third_party/ryujinx/src/Ryujinx.Graphics.Shader/Instructions/InstEmitAmpereAlu.cs b/dm/third_party/ryujinx/src/Ryujinx.Graphics.Shader/Instructions/InstEmitAmpereAlu.cs
index b181718d..4cbfcadf 100644
--- a/dm/third_party/ryujinx/src/Ryujinx.Graphics.Shader/Instructions/InstEmitAmpereAlu.cs
+++ b/dm/third_party/ryujinx/src/Ryujinx.Graphics.Shader/Instructions/InstEmitAmpereAlu.cs
@@ -627,12 +627,29 @@ public static void ACs2r(EmitterContext context)
             }
         }
 
+        // Special register 0x84 holds the variable-rate-shading fragment size in
+        // bytes 1 and 2. The host path shades every pixel, hence a 1 x 1 fragment.
+        private const int SRegFragmentSize = 0x84;
+        private const int FragmentSizeOneByOne = 0x00010100;
+
         // S2R Rd, SR / S2UR URd, SR (72..80)
         public static void AS2r(EmitterContext context)
         {
             AmpereOp x = new(context.CurrOp);
+            int sr = x.Bits(72, 80);
+
+            if (sr == SRegFragmentSize)
+            {
+                ASet(context, x, Const(FragmentSizeOneByOne), AUniform(context));
+                return;
+            }
+
+            if (!Enum.IsDefined(typeof(SReg), sr))
+            {
+                context.TranslatorContext.GpuAccessor.Log($"S2R of special register 0x{sr:x2}, which the translator does not know: read as 0");
+            }
 
-            ASet(context, x, EmitSReg(context, (SReg)x.Bits(72, 80)), AUniform(context));
+            ASet(context, x, EmitSReg(context, (SReg)sr), AUniform(context));
         }
 
         // ULDC URd, c[bank][offset] (0xab9) or c[bank][URa + offset] (0xabb, URa at 24..30): a byte, short, word or word pair (size 73..76)