shader: support S_QUADMASK_B32

tests/ShaderRecompilerComputeTests.cpp

@@ -23748,6 +23748,46 @@ TestCase ScalarNotB64UpdatesScc() {
            O::S_ENDPGM}};
 }
 
+TestCase ScalarQuadmaskB32(u32 wave_size) {
+  using O = ShaderOpcode;
+  TestCase test;
+  test.name = wave_size == 32 ? "ScalarQuadmaskB32Wave32" : "ScalarQuadmaskB32Wave64";
+  test.initial = {0, 0xffffffffu, 0x80000011u, 0x01010101u};
+  for (u32 bit = 0; bit < 32; ++bit) test.initial.push_back(1u << bit);
+  test.expected = test.initial;
+  auto &code = test.code;
+  for (u32 i = 0; i < test.initial.size(); ++i) {
+    u32 quads = 0;
+    for (u32 q = 0; q < 8; ++q) {
+      if ((test.initial[i] >> (4 * q)) & 15u) quads |= 1u << q;
+    }
+    AppendVMovU32(&code, 30, i * sizeof(u32));
+    AppendBufferLoadDword(&code, 2, 30);
+    for (u32 dst : {12u, 106u}) {
+      code.push_back(EncodeVop1(0x02, dst, Vgpr(2)));
+      AppendSMovLiteral(&code, dst + 1, 0xdeadbeefu);
+      code.push_back(EncodeSopc(0x06, InlineU32(quads == 0), InlineU32(1)));
+      code.push_back(EncodeSop1(0x2c, dst, dst));
+      code.push_back(EncodeSop2(0x0a, 14, InlineU32(1), InlineU32(0)));
+      AppendStoreSgpr(&code, dst, static_cast<u32>(test.expected.size()));
+      AppendStoreSgpr(&code, dst + 1, static_cast<u32>(test.expected.size()) + 1);
+      AppendStoreSgpr(&code, 14, static_cast<u32>(test.expected.size()) + 2);
+      test.expected.insert(test.expected.end(), {quads, 0xdeadbeefu, quads != 0});
+    }
+  }
+  code.push_back(EncodeSop1(0x2c, 12, 255));
+  code.push_back(0x000f000fu);
+  AppendStoreSgpr(&code, 12, static_cast<u32>(test.expected.size()));
+  test.expected.push_back(0x11u);
+  AppendEnd(&code);
+  test.opcodes = {O::S_QUADMASK_B32, O::S_CMP_EQ_U32, O::S_CSELECT_B32, O::S_MOV_B32,
+                  O::V_READFIRSTLANE_B32, O::V_MOV_B32, O::BUFFER_LOAD_DWORD,
+                  O::BUFFER_STORE_DWORD, O::S_ENDPGM};
+  test.compute_info.wave_size = wave_size;
+  test.has_compute_info = true;
+  return test;
+}
+
 TestCase ScalarQuadmaskB64() {
   using O = ShaderOpcode;
 
@@ -46047,6 +46087,8 @@ std::vector<TestCase> MakeCases() {
   AddCase(ScalarCompareOps);
   AddCase(ScalarShiftAddAndMaskOps);
   AddCase(ScalarNotB64UpdatesScc);
+  cases.push_back(ScalarQuadmaskB32(32));
+  cases.push_back(ScalarQuadmaskB32(64));
   AddCase(ScalarQuadmaskB64);
   AddCase(ScalarFlbitI32B64Gpu);
   AddCase(ScalarFlbitI32B32Gpu);
@@ -52562,6 +52604,13 @@ int main(int argc, char **argv) {
   std::setvbuf(stdout, nullptr, _IONBF, 0);
   EnsureConfigInitialized();
   CheckLeastRecentlyUsedCacheOrdering();
+  if (argc == 2 && std::strcmp(argv[1], "--quadmask-only") == 0) {
+    VulkanHarness vulkan;
+    RunCase(&vulkan, ScalarQuadmaskB32(32));
+    RunCase(&vulkan, ScalarQuadmaskB32(64));
+    RunCase(&vulkan, ScalarQuadmaskB64());
+    return 0;
+  }
   if (argc == 2 && std::strcmp(argv[1], "--vop3-conversions-only") == 0) {
     VulkanHarness vulkan;
     RunCase(&vulkan, Vop3FloatConversions(32));