Merge main -> staging-sm6.9

.github/workflows/clang-format-checker.yml

@@ -12,7 +12,7 @@ jobs:
     permissions:
       pull-requests: write
     steps:
-      - name: Fetch LLVM sources
+      - name: Fetch DirectXShaderCompiler sources
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
         with:
           ref: ${{ github.event.pull_request.head.sha }}
@@ -31,6 +31,20 @@ jobs:
           separator: ","
           skip_initial_fetch: true
 
+      # We need to pull the script from the main branch, so that we ensure
+      # we get the latest version of this script.
+      - name: Fetch code formatting utils
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
+        with:
+          repository: microsoft/DirectXShaderCompiler
+          ref: ${{ github.base_ref }}
+          sparse-checkout: |
+            utils/git/requirements_formatting.txt
+            utils/git/code-format-helper.py
+            utils/git/code-format-save-diff.py
+          sparse-checkout-cone-mode: false
+          path: code-format-tools
+
       - name: "Listed files"
         env:
           LISTED_FILES: ${{ steps.changed-files.outputs.all_changed_files }}
@@ -48,10 +62,10 @@ jobs:
         with:
           python-version: '3.11'
           cache: 'pip'
-          cache-dependency-path: 'utils/git/requirements_formatting.txt'
+          cache-dependency-path: 'code-format-tools/utils/git/requirements_formatting.txt'
 
       - name: Install python dependencies
-        run: pip install -r utils/git/requirements_formatting.txt
+        run: pip install -r code-format-tools/utils/git/requirements_formatting.txt
 
       - name: Run code formatter
         id: formatter
@@ -61,7 +75,7 @@ jobs:
           END_REV: ${{ github.event.pull_request.head.sha }}
           CHANGED_FILES: ${{ steps.changed-files.outputs.all_changed_files }}
         run: |
-           python utils/git/code-format-helper.py \
+           python code-format-tools/utils/git/code-format-helper.py \
             --token ${{ secrets.GITHUB_TOKEN }} \
             --issue-number $GITHUB_PR_NUMBER \
             --start-rev $START_REV \
@@ -92,28 +106,37 @@ jobs:
             } catch (err) {
               core.setFailed(`Request failed with error ${err}`)
             } 
-      - name: Fetch LLVM sources
-        uses: actions/checkout@v4
+
+      # We need to pull the script from the main branch, so that we ensure
+      # we get the latest version of this script.
+      - name: Fetch code formatting utils
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
         with:
-          fetch-depth: 2
-          path: build/main_src
+          repository: microsoft/DirectXShaderCompiler
+          ref: ${{ github.base_ref }}
+          sparse-checkout: |
+            utils/git/requirements_formatting.txt
+            utils/git/code-format-helper.py
+            utils/git/code-format-save-diff.py
+          sparse-checkout-cone-mode: false
+          path: code-format-tools
 
       - name: Setup Python env
         uses: actions/setup-python@v4
         with:
           python-version: '3.11'
           cache: 'pip'
-          cache-dependency-path: 'build/main_src/utils/git/requirements_formatting.txt'
+          cache-dependency-path: 'code-format-tools/utils/git/requirements_formatting.txt'
 
       - name: Install python dependencies
-        run: pip install -r build/main_src/utils/git/requirements_formatting.txt
+        run: pip install -r code-format-tools/utils/git/requirements_formatting.txt
 
       - name: Apply code diff
         env:
           GITHUB_PR_NUMBER: ${{ github.event.issue.number }}
           COMMENT_ID: ${{ github.event.comment.id }}
         run: |
-          python build/main_src/utils/git/code-format-save-diff.py \
+          python code-format-tools/utils/git/code-format-save-diff.py \
             --token ${{ secrets.GITHUB_TOKEN }} \
             --issue-number $GITHUB_PR_NUMBER \
             --tmp-diff-file $TMP_DIFF_FILE \

include/dxc/Support/HLSLOptions.td

@@ -317,8 +317,6 @@ def print_before_all : Flag<["-", "/"], "print-before-all">, Group<hlslcomp_Grou
   HelpText<"Print LLVM IR before each pass.">;
 def print_before : Separate<["-", "/"], "print-before">, Group<hlslcomp_Group>, Flags<[CoreOption, HelpHidden]>,
   HelpText<"Print LLVM IR before a specific pass. May be specificied multiple times.">;
-def select_validator : Separate<["-", "/"], "select-validator">, Group<hlslcomp_Group>, Flags<[CoreOption, HelpHidden]>,
-  HelpText<"Select validator: auto: (default) use DXIL.dll if found, otherwise use internal;  internal: internal non-signing validator;  external: use DXIL.dll if found, otherwise fail compilation.">;
 def print_after_all : Flag<["-", "/"], "print-after-all">, Group<hlslcomp_Group>, Flags<[CoreOption, HelpHidden]>,
   HelpText<"Print LLVM IR after each pass.">;
 def print_after : Separate<["-", "/"], "print-after">, Group<hlslcomp_Group>, Flags<[CoreOption, HelpHidden]>,

lib/DXIL/DxilOperations.cpp

@@ -765,32 +765,32 @@ const OP::OpCodeProperty OP::m_OpCodeProps[(unsigned)OP::OpCode::NumOpCodes] = {
      "unary",
      Attribute::ReadNone,
      1,
-     {{0x403}},
-     {{0x3}}}, // Overloads: hf<hf
+     {{0x3}},
+     {{0x0}}}, // Overloads: hf
     {OC::DerivCoarseY,
      "DerivCoarseY",
      OCC::Unary,
      "unary",
      Attribute::ReadNone,
      1,
-     {{0x403}},
-     {{0x3}}}, // Overloads: hf<hf
+     {{0x3}},
+     {{0x0}}}, // Overloads: hf
     {OC::DerivFineX,
      "DerivFineX",
      OCC::Unary,
      "unary",
      Attribute::ReadNone,
      1,
-     {{0x403}},
-     {{0x3}}}, // Overloads: hf<hf
+     {{0x3}},
+     {{0x0}}}, // Overloads: hf
     {OC::DerivFineY,
      "DerivFineY",
      OCC::Unary,
      "unary",
      Attribute::ReadNone,
      1,
-     {{0x403}},
-     {{0x3}}}, // Overloads: hf<hf
+     {{0x3}},
+     {{0x0}}}, // Overloads: hf
 
     // Pixel shader
     {OC::EvalSnapped,

lib/DxcSupport/HLSLOptions.cpp

@@ -1033,20 +1033,6 @@ int ReadDxcOpts(const OptTable *optionTable, unsigned flagsToInclude,
     opts.ValVerMinor = (unsigned long)minor64;
   }
 
-  llvm::StringRef valSelectStr = Args.getLastArgValue(OPT_select_validator);
-  if (!valSelectStr.empty()) {
-    opts.SelectValidator = llvm::StringSwitch<ValidatorSelection>(valSelectStr)
-                               .Case("auto", ValidatorSelection::Auto)
-                               .Case("internal", ValidatorSelection::Internal)
-                               .Case("external", ValidatorSelection::External)
-                               .Default(ValidatorSelection::Invalid);
-    if (opts.SelectValidator == ValidatorSelection::Invalid) {
-      errors << "Unsupported value '" << valSelectStr
-             << "for -select-validator option.";
-      return 1;
-    }
-  }
-
   if (opts.IsLibraryProfile() && Minor == 0xF) {
     if (opts.ValVerMajor != UINT_MAX && opts.ValVerMajor != 0) {
       errors << "Offline library profile cannot be used with non-zero "

lib/DxilValidation/DxilValidation.cpp

@@ -2287,6 +2287,32 @@ static void ValidateDxilOperationCallInProfile(CallInst *CI,
     break;
   }
 
+  // Shader Execution Reordering - from ray query
+  case DXIL::OpCode::HitObject_FromRayQuery:
+  case DXIL::OpCode::HitObject_FromRayQueryWithAttrs: {
+    for (unsigned i = 1; i < CI->getNumOperands(); ++i) {
+      Value *Arg = CI->getArgOperand(i);
+      if (isa<UndefValue>(Arg))
+        ValCtx.EmitInstrError(CI, ValidationRule::InstrNoReadingUninitialized);
+    }
+    break;
+  }
+
+  case DXIL::OpCode::HitObject_Invoke: {
+    if (isa<UndefValue>(CI->getArgOperand(1)))
+      ValCtx.EmitInstrError(CI, ValidationRule::InstrUndefHitObject);
+    if (isa<UndefValue>(CI->getArgOperand(2)))
+      ValCtx.EmitInstrError(CI, ValidationRule::InstrNoReadingUninitialized);
+  } break;
+  case DXIL::OpCode::HitObject_TraceRay: {
+    Value *Hdl = CI->getArgOperand(
+        DxilInst_HitObject_TraceRay::arg_accelerationStructure);
+    ValidateASHandle(CI, Hdl, ValCtx);
+    for (unsigned ArgIdx = 2; ArgIdx < CI->getNumArgOperands(); ++ArgIdx)
+      if (isa<UndefValue>(CI->getArgOperand(ArgIdx)))
+        ValCtx.EmitInstrError(CI, ValidationRule::InstrNoReadingUninitialized);
+    DxilInst_HitObject_TraceRay HOTraceRay(CI);
+  } break;
   case DXIL::OpCode::AtomicBinOp:
   case DXIL::OpCode::AtomicCompareExchange: {
     Type *pOverloadType = OP::GetOverloadType(Opcode, CI->getCalledFunction());

lib/Transforms/Scalar/ScalarReplAggregatesHLSL.cpp

@@ -2819,7 +2819,6 @@ void SROA_Helper::RewriteCall(CallInst *CI) {
         DeadInsts.push_back(CI);
         break;
       }
-        LLVM_FALLTHROUGH;
       default:
         // RayQuery this pointer replacement.
         if (OldVal->getType()->isPointerTy() &&

tools/clang/lib/SPIRV/EmitVisitor.cpp

@@ -2000,7 +2000,13 @@ bool EmitVisitor::visit(SpirvIntrinsicInstruction *inst) {
     }
   }
 
-  finalizeInstruction(&mainBinary);
+  auto opcode = static_cast<spv::Op>(inst->getInstruction());
+  if ((opcode == spv::Op::OpSpecConstant || opcode == spv::Op::OpConstant) &&
+      !inst->getInstructionSet()) {
+    finalizeInstruction(&typeConstantBinary);
+  } else {
+    finalizeInstruction(&mainBinary);
+  }
   return true;
 }
 

tools/clang/test/CodeGenDXIL/hlsl/types/longvec-intrinsics.hlsl

@@ -2,7 +2,6 @@
 // RUN: %dxc -T cs_6_9 -enable-16bit-types -DNUM=7   %s | FileCheck %s
 // RUN: %dxc -T cs_6_9 -enable-16bit-types -DNUM=125 %s | FileCheck %s
 // RUN: %dxc -T cs_6_9 -enable-16bit-types -DNUM=256 %s | FileCheck %s
-// RUN: %dxc -T cs_6_9 -enable-16bit-types -DNUM=1024 %s | FileCheck %s
 
 // Test vector-enabled non-trivial intrinsics that take parameters of various types.
 
@@ -203,6 +202,36 @@ void main() {
   // CHECK: fmul fast <[[NUM]] x float> [[tmp]], <float 0x3FE62E4300000000
   fRes += log(fVec1);
 
+  // CHECK-NOT: extractelement
+  // CHECK-NOT: insertelement
+  // CHECK: [[tmp:%.*]] = call <[[NUM]] x half> @dx.op.unary.[[HTY]](i32 23, <[[NUM]] x half> [[hvec2]])  ; Log(value)
+  // CHECK: [[tmp2:%.*]] = fmul fast <[[NUM]] x half> [[tmp]], [[hvec1]]
+  // CHECK: call <[[NUM]] x half> @dx.op.unary.[[HTY]](i32 21, <[[NUM]] x half> [[tmp2]])  ; Exp(value)
+  hRes += pow(hVec2, hVec1);
+
+  // CHECK-NOT: extractelement
+  // CHECK-NOT: insertelement
+  // CHECK: [[tmp:%.*]] = call <[[NUM]] x float> @dx.op.unary.[[FTY]](i32 23, <[[NUM]] x float> [[fvec2]])  ; Log(value)
+  // CHECK: [[tmp2:%.*]] = fmul fast <[[NUM]] x float> [[tmp]], [[fvec1]]
+  // CHECK: call <[[NUM]] x float> @dx.op.unary.[[FTY]](i32 21, <[[NUM]] x float> [[tmp2]])  ; Exp(value)
+  fRes += pow(fVec2, fVec1);
+
+  vector<half, NUM> hVal;
+  // CHECK-NOT: extractelement
+  // CHECK-NOT: insertelement
+  // CHECK: [[tmp:%.*]] = call <[[NUM]] x half> @dx.op.unary.[[HTY]](i32 29, <[[NUM]] x half> [[hvec1]])  ; Round_z(value)
+  // CHECK: fsub fast <[[NUM]] x half> [[hvec1]], [[tmp]]
+  hRes *= modf(hVec1, hVal);
+  hRes += hVal;
+
+  vector<float, NUM> fVal;
+  // CHECK-NOT: extractelement
+  // CHECK-NOT: insertelement
+  // CHECK: [[tmp:%.*]] = call <[[NUM]] x float> @dx.op.unary.[[FTY]](i32 29, <[[NUM]] x float> [[fvec1]])  ; Round_z(value)
+  // CHECK: fsub fast <[[NUM]] x float> [[fvec1]], [[tmp]]
+  fRes *= modf(fVec1, fVal);
+  fRes += fVal;
+
   // CHECK-NOT: extractelement
   // CHECK-NOT: insertelement
   // CHECK: [[sub:%.*]] = fsub fast <[[NUM]] x half> [[hvec2]], [[hvec1]]
@@ -227,6 +256,25 @@ void main() {
   // CHECK: fmul fast <[[NUM]] x float> [[mul]], [[sub]]
   fRes += smoothstep(fVec1, fVec2, fVec3);
 
+  // Note that Fabs is tested in longvec-trivial-unary-float-intrinsics.
+  // CHECK-NOT: extractelement
+  // CHECK-NOT: insertelement
+  // CHECK: [[tmp:%.*]] = sub <[[NUM]] x i16> zeroinitializer, [[svec1]]
+  // CHECK: call <[[NUM]] x i16> @dx.op.binary.[[STY]](i32 37, <[[NUM]] x i16> [[svec1]], <[[NUM]] x i16> [[tmp]])  ; IMax(a,b)
+  sRes += abs(sVec1);
+
+  // CHECK-NOT: extractelement
+  // CHECK-NOT: insertelement
+  // CHECK: [[tmp:%.*]] = sub <[[NUM]] x i32> zeroinitializer, [[ivec1]]
+  // CHECK: call <[[NUM]] x i32> @dx.op.binary.[[ITY]](i32 37, <[[NUM]] x i32> [[ivec1]], <[[NUM]] x i32> [[tmp]])  ; IMax(a,b)
+  iRes += abs(iVec1);
+
+  // CHECK-NOT: extractelement
+  // CHECK-NOT: insertelement
+  // CHECK: [[tmp:%.*]] = sub <[[NUM]] x i64> zeroinitializer, [[lvec1]]
+  // CHECK: call <[[NUM]] x i64> @dx.op.binary.[[LTY]](i32 37, <[[NUM]] x i64> [[lvec1]], <[[NUM]] x i64> [[tmp]])  ; IMax(a,b)
+  lRes += abs(lVec1);
+
   // Intrinsics that expand into llvm ops.
 
   // CHECK-NOT: extractelement

tools/clang/test/CodeGenDXIL/hlsl/types/longvec-trivial-scalarized-intrinsics.hlsl

@@ -9,6 +9,13 @@
 // RUN: %dxc -DFUNC=countbits   -DARITY=1 -DTYPE=uint -T ps_6_9 %s | FileCheck %s --check-prefixes=CHECK,UNARY
 // RUN: %dxc -DFUNC=firstbithigh -DARITY=1 -DTYPE=uint -T ps_6_9 %s | FileCheck %s --check-prefixes=CHECK,UNARY
 // RUN: %dxc -DFUNC=firstbitlow  -DARITY=1 -DTYPE=uint -T ps_6_9 %s | FileCheck %s --check-prefixes=CHECK,UNARY
+// RUN: %dxc -DFUNC=ddx         -DARITY=1 -T ps_6_9 %s | FileCheck %s --check-prefixes=CHECK,UNARY
+// RUN: %dxc -DFUNC=ddx_coarse  -DARITY=1 -T ps_6_9 %s | FileCheck %s --check-prefixes=CHECK,UNARY
+// RUN: %dxc -DFUNC=ddx_fine    -DARITY=1 -T ps_6_9 %s | FileCheck %s --check-prefixes=CHECK,UNARY
+// RUN: %dxc -DFUNC=ddy         -DARITY=1 -T ps_6_9 %s | FileCheck %s --check-prefixes=CHECK,UNARY
+// RUN: %dxc -DFUNC=ddy_coarse  -DARITY=1 -T ps_6_9 %s | FileCheck %s --check-prefixes=CHECK,UNARY
+// RUN: %dxc -DFUNC=ddy_fine    -DARITY=1 -T ps_6_9 %s | FileCheck %s --check-prefixes=CHECK,UNARY
+// RUN: %dxc -DFUNC=fwidth      -DARITY=1 -T ps_6_9 %s | FileCheck %s --check-prefixes=CHECK,UNARY
 // RUN: %dxc -DFUNC=QuadReadLaneAt         -DARITY=4 -T ps_6_9 %s | FileCheck %s --check-prefixes=CHECK,QUAD
 // RUN: %dxc -DFUNC=QuadReadAcrossX        -DARITY=1 -T ps_6_9 %s | FileCheck %s --check-prefixes=CHECK,QUAD
 // RUN: %dxc -DFUNC=QuadReadAcrossY        -DARITY=1 -T ps_6_9 %s | FileCheck %s --check-prefixes=CHECK,QUAD

tools/clang/test/CodeGenDXIL/hlsl/types/longvec-trivial-unary-float-intrinsics.hlsl

@@ -1,3 +1,5 @@
+// RUN: %dxc -T cs_6_9 -enable-16bit-types -DFUNC=abs  -DOP=6 -DNUM=7    %s | FileCheck %s
+// RUN: %dxc -T cs_6_9 -enable-16bit-types -DFUNC=abs  -DOP=6 -DNUM=1022 %s | FileCheck %s
 // RUN: %dxc -T cs_6_9 -enable-16bit-types -DFUNC=saturate  -DOP=7 -DNUM=7    %s | FileCheck %s
 // RUN: %dxc -T cs_6_9 -enable-16bit-types -DFUNC=saturate  -DOP=7 -DNUM=1022 %s | FileCheck %s
 // RUN: %dxc -T cs_6_9 -enable-16bit-types -DFUNC=cos  -DOP=12 -DNUM=7    %s | FileCheck %s

tools/clang/test/CodeGenSPIRV/spv.intrinsicConstantValue.hlsl

@@ -0,0 +1,13 @@
+// RUN: %dxc -Od -T cs_6_8 -spirv -fcgl %s | FileCheck %s
+
+// CHECK: %spirvIntrinsicType = OpTypeInt 8 0
+using uint8_t [[vk::ext_capability(/* Int8 */ 39)]] =
+    vk::SpirvType</* OpTypeInt */ 21, 8, 8, vk::Literal<vk::integral_constant<uint, 8> >,
+                  vk::Literal<vk::integral_constant<bool, false> > >;
+
+[[vk::ext_instruction(/* OpConstant */ 43)]] uint8_t mkconsant([[vk::ext_literal]] int v);
+
+// CHECK: OpConstant %spirvIntrinsicType 42
+static const uint8_t K = mkconsant(42);
+
+[numthreads(1, 1, 1)] void main() {}

tools/clang/test/DXC/deprecated-select-validator.hlsl

@@ -0,0 +1,14 @@
+// Test that the deprecated option, select-validator, doesn't work.
+// RUN: not %dxc -E main -T vs_6_7 -select-validator internal %s 2>&1 | FileCheck %s
+
+// CHECK: dxc failed : Unknown argument: '-select-validator'
+
+float4 main(int loc : SV_StartVertexLocation
+           , uint loc2 : SV_StartInstanceLocation
+           ) : SV_Position
+{
+    float4 r = 0;
+    r += loc;
+    r += loc2;
+    return r;
+}

tools/clang/test/HLSLFileCheck/hlsl/types/struct/struct-annotations.hlsl

@@ -1,5 +1,5 @@
-// RUN: %dxc -T ps_6_8 -E main -Qkeep_reflect_in_dxil -select-validator internal %s | FileCheck -check-prefix=CHECK68 %s
-// RUN: %dxc -T ps_6_7 -E main -Qkeep_reflect_in_dxil -select-validator internal %s | FileCheck -check-prefix=CHECK67 %s
+// RUN: %dxc -T ps_6_8 -E main -Qkeep_reflect_in_dxil %s | FileCheck -check-prefix=CHECK68 %s
+// RUN: %dxc -T ps_6_7 -E main -Qkeep_reflect_in_dxil %s | FileCheck -check-prefix=CHECK67 %s
 
 // Make sure the vector is annotated with vector size (DXIL 1.8 and higher),
 // matrix is annotated with matrix size and orientation, and scalar does not
@@ -47,4 +47,4 @@ StructuredBuffer<MyStruct> g_myStruct;
 float main() : SV_Target 
 { 
     return g_myStruct[0].vec.x + g_myStruct[0].vec.y; 
-}
+}

tools/clang/test/HLSLFileCheck/infra/auto-dxilver.hlsl

@@ -5,14 +5,17 @@
 // This should implicitly require dxilver 1.8.
 
 // RUN: %dxc -T vs_6_8 -Vd %s | FileCheck %s
-// Even though this is using -Vd, the validator version is set by the available
-// validator.  If that isn't version 1.8 or above, we'll see an error.
+// Even though this is using -Vd, the validator version being checked is the internal
+// validator's version. If a pre-DXIL-1.8 DXC was used to run this test, we expect failure,
+// since the internal validator will be the same version as the older DXC.
 // The implicit dxilver logic should not skip the check when -Vd is used.
 // CHECK-NOT: error: validator version {{.*}} does not support target profile.
 
 // RUN: %dxc -T vs_6_0 -validator-version 1.8 %s | FileCheck %s
 // Even though target is 6.0, the explicit -validator-version should add an
-// implicit dxilver 1.8 requirement.
+// implicit dxilver 1.8 requirement. The requirement should pass for DXCs that
+// are newer than DXIL Version 1.8, since then, the internal validator's version will
+// be sufficiently new for this check.
 // CHECK-NOT: error: The module cannot be validated by the version of the validator currently attached.
 
 // This error would occur if run against wrong compiler.
@@ -21,8 +24,6 @@
 // Catch any other unexpected error cases.
 // CHECK-NOT: error
 
-// RUN: %dxc -T vs_6_8 -select-validator internal %s | FileCheck %s
-// This should always be run, and always succeed.
 // CHECK: define void @main()
 
 void main() {}

tools/clang/test/LitDXILValidation/load-store-validation.ll

@@ -1,3 +1,4 @@
+; REQUIRES: dxil-1-9
 ; RUN: not %dxv %s 2>&1 | FileCheck %s
 
 ; Ensure proper validation errors are produced for invalid parameters to load and store operations.