[llvm] [DXIL] Complete abs lowering (PR #86158)
via llvm-commits
llvm-commits at lists.llvm.org
Thu Mar 21 10:16:52 PDT 2024
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-backend-directx
Author: Farzon Lotfi (farzonl)
<details>
<summary>Changes</summary>
This change completes #<!-- -->86155
- `DXIL.td` - lowering `fabs` intrinsic to the float dxil op.
- `DXILIntrinsicExpansion.cpp` - Add intrinsic expansion for the abs case.
---
Full diff: https://github.com/llvm/llvm-project/pull/86158.diff
5 Files Affected:
- (modified) llvm/lib/Target/DirectX/DXIL.td (+2)
- (modified) llvm/lib/Target/DirectX/DXILIntrinsicExpansion.cpp (+23)
- (added) llvm/test/CodeGen/DirectX/abs-vec.ll (+34)
- (added) llvm/test/CodeGen/DirectX/abs.ll (+38)
- (added) llvm/test/CodeGen/DirectX/fabs.ll (+32)
``````````diff
diff --git a/llvm/lib/Target/DirectX/DXIL.td b/llvm/lib/Target/DirectX/DXIL.td
index 36eb29d53766f0..ba7cd6ae91833c 100644
--- a/llvm/lib/Target/DirectX/DXIL.td
+++ b/llvm/lib/Target/DirectX/DXIL.td
@@ -255,6 +255,8 @@ class DXILOpMapping<int opCode, DXILOpClass opClass,
}
// Concrete definition of DXIL Operation mapping to corresponding LLVM intrinsic
+def Abs : DXILOpMapping<6, unary, int_fabs,
+ "Returns the absolute value of the input.">;
def IsInf : DXILOpMapping<9, isSpecialFloat, int_dx_isinf,
"Determines if the specified value is infinite.",
[llvm_i1_ty, llvm_halforfloat_ty]>;
diff --git a/llvm/lib/Target/DirectX/DXILIntrinsicExpansion.cpp b/llvm/lib/Target/DirectX/DXILIntrinsicExpansion.cpp
index 0db42bc0a0fb64..b46564702c7aad 100644
--- a/llvm/lib/Target/DirectX/DXILIntrinsicExpansion.cpp
+++ b/llvm/lib/Target/DirectX/DXILIntrinsicExpansion.cpp
@@ -33,6 +33,7 @@ using namespace llvm;
static bool isIntrinsicExpansion(Function &F) {
switch (F.getIntrinsicID()) {
+ case Intrinsic::abs:
case Intrinsic::exp:
case Intrinsic::dx_any:
case Intrinsic::dx_clamp:
@@ -46,6 +47,26 @@ static bool isIntrinsicExpansion(Function &F) {
return false;
}
+static bool expandAbs(CallInst *Orig) {
+ Value *X = Orig->getOperand(0);
+ IRBuilder<> Builder(Orig->getParent());
+ Builder.SetInsertPoint(Orig);
+ Type *Ty = X->getType();
+ Type *EltTy = Ty->getScalarType();
+ Constant *Zero = Ty->isVectorTy()
+ ? ConstantVector::getSplat(
+ ElementCount::getFixed(
+ cast<FixedVectorType>(Ty)->getNumElements()),
+ ConstantInt::get(EltTy, 0))
+ : ConstantInt::get(EltTy, 0);
+ auto *V = Builder.CreateSub(Zero, X);
+ auto *MaxCall =
+ Builder.CreateIntrinsic(Ty, Intrinsic::smax, {X, V}, nullptr, "dx.max");
+ Orig->replaceAllUsesWith(MaxCall);
+ Orig->eraseFromParent();
+ return true;
+}
+
static bool expandIntegerDot(CallInst *Orig, Intrinsic::ID DotIntrinsic) {
assert(DotIntrinsic == Intrinsic::dx_sdot ||
DotIntrinsic == Intrinsic::dx_udot);
@@ -213,6 +234,8 @@ static bool expandClampIntrinsic(CallInst *Orig, Intrinsic::ID ClampIntrinsic) {
static bool expandIntrinsic(Function &F, CallInst *Orig) {
switch (F.getIntrinsicID()) {
+ case Intrinsic::abs:
+ return expandAbs(Orig);
case Intrinsic::exp:
return expandExpIntrinsic(Orig);
case Intrinsic::dx_any:
diff --git a/llvm/test/CodeGen/DirectX/abs-vec.ll b/llvm/test/CodeGen/DirectX/abs-vec.ll
new file mode 100644
index 00000000000000..1c40555eb390cd
--- /dev/null
+++ b/llvm/test/CodeGen/DirectX/abs-vec.ll
@@ -0,0 +1,34 @@
+; RUN: opt -S -dxil-intrinsic-expansion < %s | FileCheck %s
+
+; Make sure dxil operation function calls for abs are generated for int vectors.
+
+; CHECK-LABEL: abs_i16Vec2
+define noundef <2 x i16> @abs_i16Vec2(<2 x i16> noundef %a) #0 {
+entry:
+; CHECK: sub <2 x i16> zeroinitializer, %a
+; CHECK: call <2 x i16> @llvm.smax.v2i16(<2 x i16> %a, <2 x i16> %{{.*}})
+ %elt.abs = call <2 x i16> @llvm.abs.v2i16(<2 x i16> %a, i1 false)
+ ret <2 x i16> %elt.abs
+}
+
+; CHECK-LABEL: abs_i32Vec3
+define noundef <3 x i32> @abs_i32Vec3(<3 x i32> noundef %a) #0 {
+entry:
+; CHECK: sub <3 x i32> zeroinitializer, %a
+; CHECK: call <3 x i32> @llvm.smax.v3i32(<3 x i32> %a, <3 x i32> %{{.*}})
+ %elt.abs = call <3 x i32> @llvm.abs.v3i32(<3 x i32> %a, i1 false)
+ ret <3 x i32> %elt.abs
+}
+
+; CHECK-LABEL: abs_i64Vec4
+define noundef <4 x i64> @abs_i64Vec4(<4 x i64> noundef %a) #0 {
+entry:
+; CHECK: sub <4 x i64> zeroinitializer, %a
+; CHECK: call <4 x i64> @llvm.smax.v4i64(<4 x i64> %a, <4 x i64> %{{.*}})
+ %elt.abs = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a, i1 false)
+ ret <4 x i64> %elt.abs
+}
+
+declare <2 x i16> @llvm.abs.v2i16(<2 x i16>, i1 immarg)
+declare <3 x i32> @llvm.abs.v3i32(<3 x i32>, i1 immarg)
+declare <4 x i64> @llvm.abs.v4i64(<4 x i64>, i1 immarg)
diff --git a/llvm/test/CodeGen/DirectX/abs.ll b/llvm/test/CodeGen/DirectX/abs.ll
new file mode 100644
index 00000000000000..822580e8c089af
--- /dev/null
+++ b/llvm/test/CodeGen/DirectX/abs.ll
@@ -0,0 +1,38 @@
+; RUN: opt -S -dxil-intrinsic-expansion < %s | FileCheck %s --check-prefixes=CHECK,EXPCHECK
+; RUN: opt -S -dxil-op-lower < %s | FileCheck %s --check-prefixes=CHECK,DOPCHECK
+
+; Make sure dxil operation function calls for abs are generated for int16_t/int/int64_t.
+
+; CHECK-LABEL: abs_i16
+define noundef i16 @abs_i16(i16 noundef %a) {
+entry:
+; CHECK: sub i16 0, %a
+; EXPCHECK: call i16 @llvm.smax.i16(i16 %a, i16 %{{.*}})
+; DOPCHECK: call i16 @dx.op.binary.i16(i32 37, i16 %a, i16 %{{.*}})
+ %elt.abs = call i16 @llvm.abs.i16(i16 %a, i1 false)
+ ret i16 %elt.abs
+}
+
+; CHECK-LABEL: abs_i32
+define noundef i32 @abs_i32(i32 noundef %a) {
+entry:
+; CHECK: sub i32 0, %a
+; EXPCHECK: call i32 @llvm.smax.i32(i32 %a, i32 %{{.*}})
+; DOPCHECK: call i32 @dx.op.binary.i32(i32 37, i32 %a, i32 %{{.*}})
+ %elt.abs = call i32 @llvm.abs.i32(i32 %a, i1 false)
+ ret i32 %elt.abs
+}
+
+; CHECK-LABEL: abs_i64
+define noundef i64 @abs_i64(i64 noundef %a) {
+entry:
+; CHECK: sub i64 0, %a
+; EXPCHECK: call i64 @llvm.smax.i64(i64 %a, i64 %{{.*}})
+; DOPCHECK: call i64 @dx.op.binary.i64(i32 37, i64 %a, i64 %{{.*}})
+ %elt.abs = call i64 @llvm.abs.i64(i64 %a, i1 false)
+ ret i64 %elt.abs
+}
+
+declare i16 @llvm.abs.i16(i16, i1 immarg)
+declare i32 @llvm.abs.i32(i32, i1 immarg)
+declare i64 @llvm.abs.i64(i64, i1 immarg)
diff --git a/llvm/test/CodeGen/DirectX/fabs.ll b/llvm/test/CodeGen/DirectX/fabs.ll
new file mode 100644
index 00000000000000..3b3f8aa9a4a928
--- /dev/null
+++ b/llvm/test/CodeGen/DirectX/fabs.ll
@@ -0,0 +1,32 @@
+; RUN: opt -S -dxil-op-lower < %s | FileCheck %s
+
+; Make sure dxil operation function calls for abs are generated for float, half, and double.
+
+
+; CHECK-LABEL: fabs_half
+define noundef half @fabs_half(half noundef %a) {
+entry:
+ ; CHECK: call half @dx.op.unary.f16(i32 6, half %{{.*}})
+ %elt.abs = call half @llvm.fabs.f16(half %a)
+ ret half %elt.abs
+}
+
+; CHECK-LABEL: fabs_float
+define noundef float @fabs_float(float noundef %a) {
+entry:
+; CHECK: call float @dx.op.unary.f32(i32 6, float %{{.*}})
+ %elt.abs = call float @llvm.fabs.f32(float %a)
+ ret float %elt.abs
+}
+
+; CHECK-LABEL: fabs_double
+define noundef double @fabs_double(double noundef %a) {
+entry:
+; CHECK: call double @dx.op.unary.f64(i32 6, double %{{.*}})
+ %elt.abs = call double @llvm.fabs.f64(double %a)
+ ret double %elt.abs
+}
+
+declare half @llvm.fabs.f16(half)
+declare float @llvm.fabs.f32(float)
+declare double @llvm.fabs.f64(double)
``````````
</details>
https://github.com/llvm/llvm-project/pull/86158
More information about the llvm-commits
mailing list