[clang] [CIR] Implement HLSL Matrix Truncation to Matrix dest (PR #242703)
via cfe-commits
cfe-commits at lists.llvm.org
Sat Nov 14 13:12:41 PST 2026
llvmorg-github-actions[bot] wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-clang
Author: Amr Hesham (AmrDeveloper)
<details>
<summary>Changes</summary>
Implement support for HLSL Matrix Truncation to Matrix dest
---
Full diff: https://github.com/llvm/llvm-project/pull/242703.diff
7 Files Affected:
- (modified) clang/include/clang/CIR/Dialect/IR/CIROps.td (+4-5)
- (modified) clang/include/clang/CIR/Dialect/IR/CIRTypeConstraints.td (+6)
- (modified) clang/lib/CIR/CodeGen/CIRGenBuilder.h (+13)
- (modified) clang/lib/CIR/CodeGen/CIRGenExpr.cpp (-6)
- (modified) clang/lib/CIR/CodeGen/CIRGenExprScalar.cpp (+31)
- (modified) clang/lib/CIR/Dialect/IR/CIRDialect.cpp (+15-10)
- (added) clang/test/CIR/CodeGenHLSL/matrixx-truncation.hlsl (+24)
``````````diff
diff --git a/clang/include/clang/CIR/Dialect/IR/CIROps.td b/clang/include/clang/CIR/Dialect/IR/CIROps.td
index 82e3095e527b13..c11b444e85057c 100644
--- a/clang/include/clang/CIR/Dialect/IR/CIROps.td
+++ b/clang/include/clang/CIR/Dialect/IR/CIROps.td
@@ -6841,12 +6841,12 @@ def CIR_VecShuffleOp : CIR_Op<"vec.shuffle", [
}];
let arguments = (ins
- CIR_VectorType:$vec1,
- CIR_VectorType:$vec2,
+ CIR_AnyVectorOrMatrixType:$vec1,
+ CIR_AnyVectorOrMatrixType:$vec2,
CIR_IntArrayAttr:$indices
);
- let results = (outs CIR_VectorType:$result);
+ let results = (outs CIR_AnyVectorOrMatrixType:$result);
let assemblyFormat = [{
`(` $vec1 `,` $vec2 `:` qualified(type($vec1)) `)` $indices `:`
qualified(type($result)) attr-dict
@@ -6962,8 +6962,7 @@ def CIR_VecSplatOp : CIR_Op<"vec.splat", [
}];
let arguments = (ins CIR_VectorElementType:$value);
- let results = (outs AnyTypeOf<[CIR_VectorType, CIR_MatrixType],
- "vector or matrix type">:$result);
+ let results = (outs CIR_AnyVectorOrMatrixType:$result);
let assemblyFormat = [{
$value `:` type($value) `,` qualified(type($result)) attr-dict
diff --git a/clang/include/clang/CIR/Dialect/IR/CIRTypeConstraints.td b/clang/include/clang/CIR/Dialect/IR/CIRTypeConstraints.td
index cfa498f87df9c4..c2b0f0d864ff88 100644
--- a/clang/include/clang/CIR/Dialect/IR/CIRTypeConstraints.td
+++ b/clang/include/clang/CIR/Dialect/IR/CIRTypeConstraints.td
@@ -441,6 +441,12 @@ def CIR_AnyIntOrVecOrMatrixOfIntType
let cppFunctionName = "isIntOrVectorOrMatrixOfIntType";
}
+def CIR_AnyVectorOrMatrixType
+ : AnyTypeOf<[CIR_AnyVectorType, CIR_AnyMatrixType],
+ "vector or matrix type"> {
+ let cppFunctionName = "isVectorOrMatrixType";
+}
+
def CIR_AnyFloatOrVecOrMatrixOfFloatType
: AnyTypeOf<[CIR_AnyFloatType, CIR_VectorOfFloatType,
CIR_MatrixOfFloatType],
diff --git a/clang/lib/CIR/CodeGen/CIRGenBuilder.h b/clang/lib/CIR/CodeGen/CIRGenBuilder.h
index 3f96bc927b6245..114c2956d022e0 100644
--- a/clang/lib/CIR/CodeGen/CIRGenBuilder.h
+++ b/clang/lib/CIR/CodeGen/CIRGenBuilder.h
@@ -812,6 +812,19 @@ class CIRGenBuilderTy : public cir::CIRBaseBuilderTy {
getArrayAttr(maskAttrs));
}
+ cir::VecShuffleOp createVecShuffle(mlir::Location loc, mlir::Value vec1,
+ mlir::Type resultTy,
+ llvm::ArrayRef<int64_t> mask) {
+ cir::ConstantOp poison =
+ getConstant(loc, cir::PoisonAttr::get(vec1.getType()));
+ auto maskAttrs = llvm::to_vector_of<mlir::Attribute>(
+ llvm::map_range(mask, [&](int32_t idx) {
+ return cir::IntAttr::get(getSInt32Ty(), idx);
+ }));
+ return cir::VecShuffleOp::create(*this, loc, resultTy, vec1, poison,
+ getArrayAttr(maskAttrs));
+ }
+
cir::VecShuffleOp createVecShuffle(mlir::Location loc, mlir::Value vec1,
mlir::Value vec2,
llvm::ArrayRef<int64_t> mask) {
diff --git a/clang/lib/CIR/CodeGen/CIRGenExpr.cpp b/clang/lib/CIR/CodeGen/CIRGenExpr.cpp
index 76273e1f95e1f1..5750233d796a03 100644
--- a/clang/lib/CIR/CodeGen/CIRGenExpr.cpp
+++ b/clang/lib/CIR/CodeGen/CIRGenExpr.cpp
@@ -3095,12 +3095,6 @@ Address CIRGenFunction::createMemTemp(QualType ty, CharUnits align,
Address result =
createTempAlloca(convertTypeForMem(ty), /*destAddrSpace=*/{}, align, loc,
name, /*arraySize=*/nullptr, alloca, ip);
- if (ty->isConstantMatrixType()) {
- assert(!cir::MissingFeatures::matrixType());
- cgm.errorNYI(loc, "createMemTemp constant matrix type");
- return Address::invalid();
- }
-
return result;
}
diff --git a/clang/lib/CIR/CodeGen/CIRGenExprScalar.cpp b/clang/lib/CIR/CodeGen/CIRGenExprScalar.cpp
index ee885fe1574f14..890afbbaa8cadc 100644
--- a/clang/lib/CIR/CodeGen/CIRGenExprScalar.cpp
+++ b/clang/lib/CIR/CodeGen/CIRGenExprScalar.cpp
@@ -2975,7 +2975,38 @@ mlir::Value ScalarExprEmitter::VisitCastExpr(CastExpr *ce) {
}
return builder.createExtractElement(loc, vec, /*idx=*/0);
}
+ case CK_HLSLMatrixTruncation: {
+ assert((destTy->isMatrixType() || destTy->isBuiltinType()) &&
+ "Destination type must be a matrix or builtin type.");
+ mlir::Value matrix = Visit(subExpr);
+ mlir::Location loc = cgf.getLoc(subExpr->getSourceRange());
+ if (const auto *matTy = destTy->getAs<ConstantMatrixType>()) {
+ llvm::SmallVector<int64_t> mask(matTy->getNumElementsFlattened());
+ unsigned numCols = matTy->getNumColumns();
+ unsigned numRows = matTy->getNumRows();
+
+ auto *srcMatTy = subExpr->getType()->getAs<ConstantMatrixType>();
+ assert(srcMatTy && "Source type must be a matrix type.");
+ assert(numRows <= srcMatTy->getNumRows());
+ assert(numCols <= srcMatTy->getNumColumns());
+
+ for (unsigned row = 0; row < numRows; row++)
+ for (unsigned column = 0; column < numCols; column++)
+ mask[matTy->getColumnMajorFlattenedIndex(row, column)] =
+ srcMatTy->getColumnMajorFlattenedIndex(row, column);
+
+ auto elementTy =
+ mlir::cast<cir::MatrixType>(matrix.getType()).getElementType();
+ auto resultTy = cir::MatrixType::get(builder.getContext(), elementTy,
+ numRows, numCols);
+ return builder.createVecShuffle(loc, matrix, resultTy, mask);
+ }
+ cgf.getCIRGenModule().errorNYI(subExpr->getSourceRange(),
+ "ScalarExprEmitter::VisitCastExpr: HLSL "
+ "Matrix truncation with non-matrix dest");
+ return {};
+ }
default:
cgf.getCIRGenModule().errorNYI(subExpr->getSourceRange(),
"CastExpr: ", ce->getCastKindName());
diff --git a/clang/lib/CIR/Dialect/IR/CIRDialect.cpp b/clang/lib/CIR/Dialect/IR/CIRDialect.cpp
index 10a61434a4a30d..3d89f0b2700102 100644
--- a/clang/lib/CIR/Dialect/IR/CIRDialect.cpp
+++ b/clang/lib/CIR/Dialect/IR/CIRDialect.cpp
@@ -4666,31 +4666,36 @@ OpFoldResult cir::VecShuffleOp::fold(FoldAdaptor adaptor) {
}
return cir::ConstVectorAttr::get(
- getType(), mlir::ArrayAttr::get(getContext(), elements));
+ mlir::cast<cir::VectorType>(getType()),
+ mlir::ArrayAttr::get(getContext(), elements));
}
LogicalResult cir::VecShuffleOp::verify() {
- if (getVec1().getType().isScalable() || getVec2().getType().isScalable() ||
- getResult().getType().isScalable())
+ if (!mlir::isa<cir::VectorType>(getType()))
+ return success();
+
+ auto vec1Ty = mlir::cast<cir::VectorType>(getVec1().getType());
+ auto vec2Ty = mlir::cast<cir::VectorType>(getVec2().getType());
+ auto resultTy = mlir::cast<cir::VectorType>(getResult().getType());
+ if (vec1Ty.isScalable() || vec2Ty.isScalable() || resultTy.isScalable())
return emitOpError() << "scalable vectors are not supported";
// The number of elements in the indices array must match the number of
// elements in the result type.
- if (getIndices().size() != getResult().getType().getFixedSize()) {
+ if (getIndices().size() != resultTy.getFixedSize()) {
return emitOpError() << ": the number of elements in " << getIndices()
- << " and " << getResult().getType() << " don't match";
+ << " and " << resultTy << " don't match";
}
// The element types of the two input vectors and of the result type must
// match.
- if (getVec1().getType().getElementType() !=
- getResult().getType().getElementType()) {
+ if (vec1Ty.getElementType() != resultTy.getElementType()) {
return emitOpError() << ": element types of " << getVec1().getType()
- << " and " << getResult().getType() << " don't match";
+ << " and " << resultTy << " don't match";
}
- const uint64_t maxValidIndex = getVec1().getType().getFixedSize() +
- getVec2().getType().getFixedSize() - 1;
+ const uint64_t maxValidIndex =
+ vec1Ty.getFixedSize() + vec2Ty.getFixedSize() - 1;
if (llvm::any_of(
getIndices().getAsRange<cir::IntAttr>(), [&](cir::IntAttr idxAttr) {
return idxAttr.getSInt() != -1 && idxAttr.getUInt() > maxValidIndex;
diff --git a/clang/test/CIR/CodeGenHLSL/matrixx-truncation.hlsl b/clang/test/CIR/CodeGenHLSL/matrixx-truncation.hlsl
new file mode 100644
index 00000000000000..10cef611a3f968
--- /dev/null
+++ b/clang/test/CIR/CodeGenHLSL/matrixx-truncation.hlsl
@@ -0,0 +1,24 @@
+// RUN: %clang_cc1 -x hlsl -std=hlsl2021 -triple spirv-unknown-vulkan-library -finclude-default-header -fclangir -emit-cir %s -o %t.cir
+// RUN: FileCheck --input-file=%t.cir %s -check-prefix=CIR
+// RUN: %clang_cc1 -x hlsl -std=hlsl2021 -triple spirv-unknown-vulkan-library -finclude-default-header -fclangir -emit-llvm %s -o %t-cir.ll
+// RUN: FileCheck --input-file=%t-cir.ll %s -check-prefix=LLVM
+// RUN: %clang_cc1 -x hlsl -std=hlsl2021 -triple spirv-unknown-vulkan-library -finclude-default-header -emit-llvm -disable-llvm-passes %s -o %t.ll
+// RUN: FileCheck --input-file=%t.ll %s -check-prefix=LLVM
+
+void trunc_cast() {
+ int4x4 a;
+ int3x4 b = (int3x4) a;
+}
+
+// CIR: %[[A_ADDR:.*]] = cir.alloca "a" {{.*}} : !cir.ptr<!cir.matrix<4 x 4 x !s32i>>
+// CIR: %[[B_ADDR:.*]] = cir.alloca "b" {{.*}} init : !cir.ptr<!cir.matrix<3 x 4 x !s32i>>
+// CIR: %[[TMP_A:.*]] = cir.load {{.*}} %[[A_ADDR]] : !cir.ptr<!cir.matrix<4 x 4 x !s32i>>, !cir.matrix<4 x 4 x !s32i>
+// CIR: %[[POISON:.*]] = cir.const #cir.poison : !cir.matrix<4 x 4 x !s32i>
+// CIR: %[[RESULT:.*]] = cir.vec.shuffle(%[[TMP_A]], %[[POISON]] : !cir.matrix<4 x 4 x !s32i>) [#cir.int<0> : !s32i, #cir.int<1> : !s32i, #cir.int<2> : !s32i, #cir.int<4> : !s32i, #cir.int<5> : !s32i, #cir.int<6> : !s32i, #cir.int<8> : !s32i, #cir.int<9> : !s32i, #cir.int<10> : !s32i, #cir.int<12> : !s32i, #cir.int<13> : !s32i, #cir.int<14> : !s32i] : !cir.matrix<3 x 4 x !s32i> loc(#loc6)
+// CIR: cir.store {{.*}} %[[RESULT]], %[[B_ADDR]] : !cir.matrix<3 x 4 x !s32i>, !cir.ptr<!cir.matrix<3 x 4 x !s32i>>
+
+// LLVM: %[[A_ADDR:.*]] = alloca [4 x <4 x i32>], align 4
+// LLVM: %[[B_ADDR:.*]] = alloca [4 x <3 x i32>], align 4
+// LLVM: %[[TMP_A:.*]] = load <16 x i32>, ptr %[[A_ADDR]], align 4
+// LLVM: %[[RESULT:.*]] = shufflevector <16 x i32> %[[TMP_A]], <16 x i32> poison, <12 x i32> <i32 0, i32 1, i32 2, i32 4, i32 5, i32 6, i32 8, i32 9, i32 10, i32 12, i32 13, i32 14>
+// LLVM: store <12 x i32> %[[RESULT]], ptr %[[B_ADDR]], align 4
``````````
</details>
https://github.com/llvm/llvm-project/pull/242703
More information about the cfe-commits
mailing list