751a546fa9
Issue: https://github.com/llvm/llvm-project/issues/99172 - [x] Implement `WavePrefixSum` clang builtin - [x] Link `WavePrefixSum` clang builtin with `hlsl_intrinsics.h` - [x] Add sema checks for `WavePrefixSum` to `CheckHLSLBuiltinFunctionCall` in `SemaChecking.cpp` - [x] Add codegen for `WavePrefixSum` to `EmitHLSLBuiltinExpr` in `CGBuiltin.cpp` - [x] Add codegen tests to `clang/test/CodeGenHLSL/builtins/WavePrefixSum.hlsl` - [x] Add sema tests to `clang/test/SemaHLSL/BuiltIns/WavePrefixSum-errors.hlsl` - [x] Create the `int_dx_WavePrefixSum` intrinsic in `IntrinsicsDirectX.td` - [x] Create the `DXILOpMapping` of `int_dx_WavePrefixSum` to `121` in `DXIL.td` - [x] Create the `WavePrefixSum.ll` and `WavePrefixSum_errors.ll` tests in `llvm/test/CodeGen/DirectX/` - [x] Create the `int_spv_WavePrefixSum` intrinsic in `IntrinsicsSPIRV.td` - [x] In SPIRVInstructionSelector.cpp create the `WavePrefixSum` lowering and map it to `int_spv_WavePrefixSum` in `SPIRVInstructionSelector::selectIntrinsic`. - [x] Create SPIR-V backend test case in `llvm/test/CodeGen/SPIRV/hlsl-intrinsics/WavePrefixSum.ll` I also added a new macro `GENERATE_HLSL_INTRINSIC_FUNCTION_SELECT_UNSIGNED` in conjunction with the new function `getUnsignedIntrinsicVariant` to make selecting unsigned variants of the intrinsic easier. As a result, I was able to replace `getWaveActiveSumIntrinsic`, `getWaveActiveMaxIntrinsic`, and `getWaveActiveMinIntrinsic` using the new macro.
144 lines
5.5 KiB
LLVM
144 lines
5.5 KiB
LLVM
; RUN: opt -S -scalarizer -dxil-op-lower -mtriple=dxil-pc-shadermodel6.3-library < %s | FileCheck %s
|
|
|
|
; Test that for scalar values, WavePrefixSum maps down to the DirectX op
|
|
|
|
define noundef half @wave_prefix_sum_half(half noundef %expr) {
|
|
entry:
|
|
; CHECK: call half @dx.op.wavePrefixOp.f16(i32 121, half %expr, i8 0, i8 0)
|
|
%ret = call half @llvm.dx.wave.prefix.sum.f16(half %expr)
|
|
ret half %ret
|
|
}
|
|
|
|
define noundef float @wave_prefix_sum_float(float noundef %expr) {
|
|
entry:
|
|
; CHECK: call float @dx.op.wavePrefixOp.f32(i32 121, float %expr, i8 0, i8 0)
|
|
%ret = call float @llvm.dx.wave.prefix.sum.f32(float %expr)
|
|
ret float %ret
|
|
}
|
|
|
|
define noundef double @wave_prefix_sum_double(double noundef %expr) {
|
|
entry:
|
|
; CHECK: call double @dx.op.wavePrefixOp.f64(i32 121, double %expr, i8 0, i8 0)
|
|
%ret = call double @llvm.dx.wave.prefix.sum.f64(double %expr)
|
|
ret double %ret
|
|
}
|
|
|
|
define noundef i16 @wave_prefix_sum_i16(i16 noundef %expr) {
|
|
entry:
|
|
; CHECK: call i16 @dx.op.wavePrefixOp.i16(i32 121, i16 %expr, i8 0, i8 0)
|
|
%ret = call i16 @llvm.dx.wave.prefix.sum.i16(i16 %expr)
|
|
ret i16 %ret
|
|
}
|
|
|
|
define noundef i32 @wave_prefix_sum_i32(i32 noundef %expr) {
|
|
entry:
|
|
; CHECK: call i32 @dx.op.wavePrefixOp.i32(i32 121, i32 %expr, i8 0, i8 0)
|
|
%ret = call i32 @llvm.dx.wave.prefix.sum.i32(i32 %expr)
|
|
ret i32 %ret
|
|
}
|
|
|
|
define noundef i64 @wave_prefix_sum_i64(i64 noundef %expr) {
|
|
entry:
|
|
; CHECK: call i64 @dx.op.wavePrefixOp.i64(i32 121, i64 %expr, i8 0, i8 0)
|
|
%ret = call i64 @llvm.dx.wave.prefix.sum.i64(i64 %expr)
|
|
ret i64 %ret
|
|
}
|
|
|
|
define noundef i16 @wave_prefix_usum_i16(i16 noundef %expr) {
|
|
entry:
|
|
; CHECK: call i16 @dx.op.wavePrefixOp.i16(i32 121, i16 %expr, i8 0, i8 1)
|
|
%ret = call i16 @llvm.dx.wave.prefix.usum.i16(i16 %expr)
|
|
ret i16 %ret
|
|
}
|
|
|
|
define noundef i32 @wave_prefix_usum_i32(i32 noundef %expr) {
|
|
entry:
|
|
; CHECK: call i32 @dx.op.wavePrefixOp.i32(i32 121, i32 %expr, i8 0, i8 1)
|
|
%ret = call i32 @llvm.dx.wave.prefix.usum.i32(i32 %expr)
|
|
ret i32 %ret
|
|
}
|
|
|
|
define noundef i64 @wave_prefix_usum_i64(i64 noundef %expr) {
|
|
entry:
|
|
; CHECK: call i64 @dx.op.wavePrefixOp.i64(i32 121, i64 %expr, i8 0, i8 1)
|
|
%ret = call i64 @llvm.dx.wave.prefix.usum.i64(i64 %expr)
|
|
ret i64 %ret
|
|
}
|
|
|
|
declare half @llvm.dx.wave.prefix.sum.f16(half)
|
|
declare float @llvm.dx.wave.prefix.sum.f32(float)
|
|
declare double @llvm.dx.wave.prefix.sum.f64(double)
|
|
|
|
declare i16 @llvm.dx.wave.prefix.sum.i16(i16)
|
|
declare i32 @llvm.dx.wave.prefix.sum.i32(i32)
|
|
declare i64 @llvm.dx.wave.prefix.sum.i64(i64)
|
|
|
|
declare i16 @llvm.dx.wave.prefix.usum.i16(i16)
|
|
declare i32 @llvm.dx.wave.prefix.usum.i32(i32)
|
|
declare i64 @llvm.dx.wave.prefix.usum.i64(i64)
|
|
|
|
; Test that for vector values, WavePrefixSum scalarizes and maps down to the
|
|
; DirectX op
|
|
|
|
define noundef <2 x half> @wave_prefix_sum_v2half(<2 x half> noundef %expr) {
|
|
entry:
|
|
; CHECK: call half @dx.op.wavePrefixOp.f16(i32 121, half %expr.i0, i8 0, i8 0)
|
|
; CHECK: call half @dx.op.wavePrefixOp.f16(i32 121, half %expr.i1, i8 0, i8 0)
|
|
%ret = call <2 x half> @llvm.dx.wave.prefix.sum.v2f16(<2 x half> %expr)
|
|
ret <2 x half> %ret
|
|
}
|
|
|
|
define noundef <3 x i32> @wave_prefix_sum_v3i32(<3 x i32> noundef %expr) {
|
|
entry:
|
|
; CHECK: call i32 @dx.op.wavePrefixOp.i32(i32 121, i32 %expr.i0, i8 0, i8 0)
|
|
; CHECK: call i32 @dx.op.wavePrefixOp.i32(i32 121, i32 %expr.i1, i8 0, i8 0)
|
|
; CHECK: call i32 @dx.op.wavePrefixOp.i32(i32 121, i32 %expr.i2, i8 0, i8 0)
|
|
%ret = call <3 x i32> @llvm.dx.wave.prefix.sum.v3i32(<3 x i32> %expr)
|
|
ret <3 x i32> %ret
|
|
}
|
|
|
|
define noundef <4 x double> @wave_prefix_sum_v4f64(<4 x double> noundef %expr) {
|
|
entry:
|
|
; CHECK: call double @dx.op.wavePrefixOp.f64(i32 121, double %expr.i0, i8 0, i8 0)
|
|
; CHECK: call double @dx.op.wavePrefixOp.f64(i32 121, double %expr.i1, i8 0, i8 0)
|
|
; CHECK: call double @dx.op.wavePrefixOp.f64(i32 121, double %expr.i2, i8 0, i8 0)
|
|
; CHECK: call double @dx.op.wavePrefixOp.f64(i32 121, double %expr.i3, i8 0, i8 0)
|
|
%ret = call <4 x double> @llvm.dx.wave.prefix.sum.v464(<4 x double> %expr)
|
|
ret <4 x double> %ret
|
|
}
|
|
|
|
declare <2 x half> @llvm.dx.wave.prefix.sum.v2f16(<2 x half>)
|
|
declare <3 x i32> @llvm.dx.wave.prefix.sum.v3i32(<3 x i32>)
|
|
declare <4 x double> @llvm.dx.wave.prefix.sum.v4f64(<4 x double>)
|
|
|
|
define noundef <2 x i16> @wave_prefix_usum_v2i16(<2 x i16> noundef %expr) {
|
|
entry:
|
|
; CHECK: call i16 @dx.op.wavePrefixOp.i16(i32 121, i16 %expr.i0, i8 0, i8 1)
|
|
; CHECK: call i16 @dx.op.wavePrefixOp.i16(i32 121, i16 %expr.i1, i8 0, i8 1)
|
|
%ret = call <2 x i16> @llvm.dx.wave.prefix.usum.v2f16(<2 x i16> %expr)
|
|
ret <2 x i16> %ret
|
|
}
|
|
|
|
define noundef <3 x i32> @wave_prefix_usum_v3i32(<3 x i32> noundef %expr) {
|
|
entry:
|
|
; CHECK: call i32 @dx.op.wavePrefixOp.i32(i32 121, i32 %expr.i0, i8 0, i8 1)
|
|
; CHECK: call i32 @dx.op.wavePrefixOp.i32(i32 121, i32 %expr.i1, i8 0, i8 1)
|
|
; CHECK: call i32 @dx.op.wavePrefixOp.i32(i32 121, i32 %expr.i2, i8 0, i8 1)
|
|
%ret = call <3 x i32> @llvm.dx.wave.prefix.usum.v3i32(<3 x i32> %expr)
|
|
ret <3 x i32> %ret
|
|
}
|
|
|
|
define noundef <4 x i64> @wave_prefix_usum_v4f64(<4 x i64> noundef %expr) {
|
|
entry:
|
|
; CHECK: call i64 @dx.op.wavePrefixOp.i64(i32 121, i64 %expr.i0, i8 0, i8 1)
|
|
; CHECK: call i64 @dx.op.wavePrefixOp.i64(i32 121, i64 %expr.i1, i8 0, i8 1)
|
|
; CHECK: call i64 @dx.op.wavePrefixOp.i64(i32 121, i64 %expr.i2, i8 0, i8 1)
|
|
; CHECK: call i64 @dx.op.wavePrefixOp.i64(i32 121, i64 %expr.i3, i8 0, i8 1)
|
|
%ret = call <4 x i64> @llvm.dx.wave.prefix.usum.v464(<4 x i64> %expr)
|
|
ret <4 x i64> %ret
|
|
}
|
|
|
|
declare <2 x i16> @llvm.dx.wave.prefix.usum.v2f16(<2 x i16>)
|
|
declare <3 x i32> @llvm.dx.wave.prefix.usum.v3i32(<3 x i32>)
|
|
declare <4 x i64> @llvm.dx.wave.prefix.usum.v4f64(<4 x i64>)
|