Skip to content

Commit

Permalink
[CIR][Lowering] VecCreateOp and VecSplatOp lowering choose LLVM:Poiso…
Browse files Browse the repository at this point in the history 
…nOp (#959)

They should use PoisonOp (which becomes PoisonValue in LLVMIR) as it is
the OG's choice.
Proof:
We generate VecCreateOp [here
](https://github.com/llvm/clangir/blob/2ca12fe5ec3a1e7279256f069010be2d68200585/clang/lib/CIR/CodeGen/CIRGenExprScalar.cpp#L1975)
And it's OG counterpart is
[here](https://github.com/llvm/clangir/blob/2ca12fe5ec3a1e7279256f069010be2d68200585/clang/lib/CodeGen/CGExprScalar.cpp#L2096)
OG uses PoisonValue. 
As to VecSplatOp, OG unconditionally [chooses PoisonValue
](https://github.com/llvm/clangir/blob/2ca12fe5ec3a1e7279256f069010be2d68200585/llvm/lib/IR/IRBuilder.cpp#L1204)

A even more solid proof for this case is that 
when we use OG to generate code for our test case I changed in this PR ,
its always using poison instead of undef as far as VecSplat and
VecCreate is concerned. The [OG generated code for vectype-ext.cpp
](https://godbolt.org/z/eqx1rns86) here.
The [OG generated code for vectype.cpp
](https://godbolt.org/z/frMjbKGeT) here.

For reference, generated CIR for the test case vectype-ext.cpp is
[here](https://godbolt.org/z/frMjbKGeT)

This is to unblock #936 to help it
set on the right path.

Note: There might be other CIR vec ops that need to choose Poison to be
consistent with OG, but I'd limit the scope of this PR, and wait to see
issue pop up in the future.
  • Loading branch information
@ghehg @lanza
ghehg authored and lanza committed Nov 3, 2024
1 parent 5931d7e commit 32ce46a
Show file tree
Hide file tree
Showing 3 changed files with 12 additions and 11 deletions.
4 changes: 2 additions & 2 deletions clang/lib/CIR/Lowering/DirectToLLVM/LowerToLLVM.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -1610,7 +1610,7 @@ class CIRVectorCreateLowering
assert(vecTy && "result type of cir.vec.create op is not VectorType");
auto llvmTy = typeConverter->convertType(vecTy);
auto loc = op.getLoc();
mlir::Value result = rewriter.create<mlir::LLVM::UndefOp>(loc, llvmTy);
mlir::Value result = rewriter.create<mlir::LLVM::PoisonOp>(loc, llvmTy);
assert(vecTy.getSize() == op.getElements().size() &&
"cir.vec.create op count doesn't match vector type elements count");
for (uint64_t i = 0; i < vecTy.getSize(); ++i) {
Expand Down Expand Up @@ -1675,7 +1675,7 @@ class CIRVectorSplatLowering
assert(vecTy && "result type of cir.vec.splat op is not VectorType");
auto llvmTy = typeConverter->convertType(vecTy);
auto loc = op.getLoc();
mlir::Value undef = rewriter.create<mlir::LLVM::UndefOp>(loc, llvmTy);
mlir::Value undef = rewriter.create<mlir::LLVM::PoisonOp>(loc, llvmTy);
mlir::Value indexValue =
rewriter.create<mlir::LLVM::ConstantOp>(loc, rewriter.getI64Type(), 0);
mlir::Value elementValue = adaptor.getValue();
Expand Down
8 changes: 4 additions & 4 deletions clang/test/CIR/CodeGen/vectype-ext.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -25,7 +25,7 @@ void vector_int_test(int x) {
// LLVM: %[[#X1:]] = load i32, ptr %{{[0-9]+}}, align 4
// LLVM-NEXT: %[[#X2:]] = load i32, ptr %{{[0-9]+}}, align 4
// LLVM-NEXT: %[[#SUM:]] = add nsw i32 %[[#X2]], 1
// LLVM-NEXT: %[[#VEC1:]] = insertelement <4 x i32> undef, i32 %[[#X1]], i64 0
// LLVM-NEXT: %[[#VEC1:]] = insertelement <4 x i32> poison, i32 %[[#X1]], i64 0
// LLVM-NEXT: %[[#VEC2:]] = insertelement <4 x i32> %[[#VEC1]], i32 5, i64 1
// LLVM-NEXT: %[[#VEC3:]] = insertelement <4 x i32> %[[#VEC2]], i32 6, i64 2
// LLVM-NEXT: %[[#VEC4:]] = insertelement <4 x i32> %[[#VEC3]], i32 %[[#SUM]], i64 3
Expand All @@ -38,7 +38,7 @@ void vector_int_test(int x) {
// LLVM: %[[#X1:]] = load i32, ptr %{{[0-9]+}}, align 4
// LLVM-NEXT: %[[#X2:]] = load i32, ptr %{{[0-9]+}}, align 4
// LLVM-NEXT: %[[#SUM:]] = add nsw i32 %[[#X2]], 1
// LLVM-NEXT: %[[#VEC1:]] = insertelement <4 x i32> undef, i32 %[[#X1]], i64 0
// LLVM-NEXT: %[[#VEC1:]] = insertelement <4 x i32> poison, i32 %[[#X1]], i64 0
// LLVM-NEXT: %[[#VEC2:]] = insertelement <4 x i32> %[[#VEC1]], i32 %[[#SUM]], i64 1
// LLVM-NEXT: %[[#VEC3:]] = insertelement <4 x i32> %[[#VEC2]], i32 0, i64 2
// LLVM-NEXT: %[[#VEC4:]] = insertelement <4 x i32> %[[#VEC3]], i32 0, i64 3
Expand Down Expand Up @@ -211,7 +211,7 @@ void vector_double_test(int x, double y) {
// LLVM: %[[#Y1:]] = load double, ptr %{{[0-9]+}}, align 8
// LLVM-NEXT: %[[#Y2:]] = load double, ptr %{{[0-9]+}}, align 8
// LLVM-NEXT: %[[#SUM:]] = fadd double %[[#Y2]], 1.000000e+00
// LLVM-NEXT: %[[#VEC1:]] = insertelement <2 x double> undef, double %[[#Y1]], i64 0
// LLVM-NEXT: %[[#VEC1:]] = insertelement <2 x double> poison, double %[[#Y1]], i64 0
// LLVM-NEXT: %[[#VEC2:]] = insertelement <2 x double> %[[#VEC1]], double %[[#SUM]], i64 1
// LLVM-NEXT: store <2 x double> %[[#VEC2]], ptr %{{[0-9]+}}, align 16

Expand All @@ -221,7 +221,7 @@ void vector_double_test(int x, double y) {
// CIR: %{{[0-9]+}} = cir.vec.create(%{{[0-9]+}}, %[[#dzero]] : !cir.double, !cir.double) : !cir.vector<!cir.double x 2>

// LLVM: %[[#Y1:]] = load double, ptr %{{[0-9]+}}, align 8
// LLVM-NEXT: %[[#VEC1:]] = insertelement <2 x double> undef, double %[[#Y1]], i64 0
// LLVM-NEXT: %[[#VEC1:]] = insertelement <2 x double> poison, double %[[#Y1]], i64 0
// LLVM-NEXT: %[[#VEC2:]] = insertelement <2 x double> %[[#VEC1]], double 0.000000e+00, i64 1
// LLVM-NEXT: store <2 x double> %[[#VEC2]], ptr %{{[0-9]+}}, align 16

Expand Down
11 changes: 6 additions & 5 deletions clang/test/CIR/Lowering/vectype.cpp
View file
Original file line number Diff line number Diff line change
@@ -1,5 +1,6 @@
// RUN: %clang_cc1 -std=c++17 -triple x86_64-unknown-linux-gnu -fclangir -emit-cir %s -o %t.cir
// RUN: cir-opt %t.cir -cir-to-llvm -o %t.mlir
// RUN: %clang_cc1 -std=c++17 -triple x86_64-unknown-linux-gnu -fclangir -emit-llvm %s -o %t.ii
// RUN: FileCheck --input-file=%t.mlir %s

typedef int vi4 __attribute__((vector_size(16)));
Expand All @@ -22,7 +23,7 @@ void vector_int_test(int x) {
// CHECK: %[[#T46:]] = llvm.load %[[#T1]] {alignment = 4 : i64} : !llvm.ptr -> i32
// CHECK: %[[#T47:]] = llvm.mlir.constant(1 : i32) : i32
// CHECK: %[[#T48:]] = llvm.add %[[#T46]], %[[#T47]] overflow<nsw> : i32
// CHECK: %[[#T49:]] = llvm.mlir.undef : vector<4xi32>
// CHECK: %[[#T49:]] = llvm.mlir.poison : vector<4xi32>
// CHECK: %[[#T50:]] = llvm.mlir.constant(0 : i64) : i64
// CHECK: %[[#T51:]] = llvm.insertelement %[[#T43]], %[[#T49]][%[[#T50]] : i64] : vector<4xi32>
// CHECK: %[[#T52:]] = llvm.mlir.constant(1 : i64) : i64
Expand All @@ -41,10 +42,10 @@ void vector_int_test(int x) {

// Scalar to vector conversion, a.k.a. vector splat.
b = a + 7;
// CHECK: %[[#undef:]] = llvm.mlir.undef : vector<4xi32>
// CHECK: %[[#poison:]] = llvm.mlir.poison : vector<4xi32>
// CHECK: %[[#zeroInt:]] = llvm.mlir.constant(0 : i64) : i64
// CHECK: %[[#inserted:]] = llvm.insertelement %[[#seven:]], %[[#undef]][%[[#zeroInt]] : i64] : vector<4xi32>
// CHECK: %[[#shuffled:]] = llvm.shufflevector %[[#inserted]], %[[#undef]] [0, 0, 0, 0] : vector<4xi32>
// CHECK: %[[#inserted:]] = llvm.insertelement %[[#seven:]], %[[#poison]][%[[#zeroInt]] : i64] : vector<4xi32>
// CHECK: %[[#shuffled:]] = llvm.shufflevector %[[#inserted]], %[[#poison]] [0, 0, 0, 0] : vector<4xi32>

// Extract element.
int c = a[x];
Expand Down Expand Up @@ -233,7 +234,7 @@ void vector_double_test(int x, double y) {
// CHECK: %[[#T30:]] = llvm.load %[[#T3]] {alignment = 8 : i64} : !llvm.ptr -> f64
// CHECK: %[[#T31:]] = llvm.mlir.constant(1.000000e+00 : f64) : f64
// CHECK: %[[#T32:]] = llvm.fadd %[[#T30]], %[[#T31]] : f64
// CHECK: %[[#T33:]] = llvm.mlir.undef : vector<2xf64>
// CHECK: %[[#T33:]] = llvm.mlir.poison : vector<2xf64>
// CHECK: %[[#T34:]] = llvm.mlir.constant(0 : i64) : i64
// CHECK: %[[#T35:]] = llvm.insertelement %[[#T29]], %[[#T33]][%[[#T34]] : i64] : vector<2xf64>
// CHECK: %[[#T36:]] = llvm.mlir.constant(1 : i64) : i64
Expand Down

0 comments on commit 32ce46a

Please sign in to comment.