61 lines
2.5 KiB
Plaintext
61 lines
2.5 KiB
Plaintext
// This test ensures that the macro to convert scale to constant for gather has been inlined and
|
|
// optimised as expected and no extra gather calls or switch statement is remaining after the optimisation.
|
|
|
|
// RUN: %{ispc} %s --target=avx2-i64x4 -h %t.h --emit-llvm -o %t.bc
|
|
// RUN: llvm-dis %t.bc -o - | FileCheck %s -check-prefix=CHECK_AVX2_I64X4 --implicit-check-not "switch"
|
|
// RUN: %{ispc} %s --target=avx2-i32x8 -h %t.h --emit-llvm -o %t.bc
|
|
// RUN: llvm-dis %t.bc -o - | FileCheck %s -check-prefix=CHECK_AVX2_I32X8 --implicit-check-not "switch"
|
|
// RUN: %{ispc} %s --target=avx2-i32x16 -h %t.h --emit-llvm -o %t.bc
|
|
// RUN: llvm-dis %t.bc -o - | FileCheck %s -check-prefix=CHECK_AVX2_I32X16 --implicit-check-not "switch"
|
|
// RUN: %{ispc} %s --target=avx512knl-x16 -h %t.h --emit-llvm -o %t.bc
|
|
// RUN: llvm-dis %t.bc -o - | FileCheck %s -check-prefix=CHECK_AVX512KNL_X16 --implicit-check-not "switch"
|
|
// RUN: %{ispc} %s --target=avx512skx-x16 -h %t.h --emit-llvm -o %t.bc
|
|
// RUN: llvm-dis %t.bc -o - | FileCheck %s -check-prefix=CHECK_AVX512SKX_X16 --implicit-check-not "switch"
|
|
|
|
// REQUIRES: X86_ENABLED
|
|
|
|
//; CHECK_AVX2_I64X4: llvm.x86.avx2.gather.d.ps
|
|
|
|
//; CHECK_AVX2_I32X8: llvm.x86.avx2.gather.d.ps.256
|
|
|
|
//; CHECK_AVX2_I32X16: llvm.x86.avx2.gather.d.ps.256
|
|
|
|
//; CHECK_AVX512KNL_X16: llvm.x86.avx512.mask.gather.dps.512
|
|
|
|
//; CHECK_AVX512SKX_X16: llvm.x86.avx512.mask.gather.dps.512
|
|
struct Foo {
|
|
uniform float x[programCount + 1];
|
|
};
|
|
void f_fu(uniform float RET[], uniform float aFOO[]) {
|
|
float a = aFOO[programIndex];
|
|
uniform Foo foo;
|
|
uniform int i;
|
|
for (i = 0; i < programCount + 1; ++i)
|
|
foo.x[i] = i;
|
|
|
|
//; CHECK_AVX2_I64X4: llvm.x86.avx2.gather.d.ps
|
|
//; CHECK_AVX2_I64X4: llvm.x86.avx2.gather.d.ps
|
|
//; CHECK_AVX2_I64X4-NOT: llvm.x86.avx2.gather.d.ps
|
|
|
|
//; CHECK_AVX2_I32X8: llvm.x86.avx2.gather.d.ps.256
|
|
//; CHECK_AVX2_I32X8: llvm.x86.avx2.gather.d.ps.256
|
|
//; CHECK_AVX2_I32X8-NOT: llvm.x86.avx2.gather.d.ps.256
|
|
|
|
//; CHECK_AVX2_I32X16: llvm.x86.avx2.gather.d.ps.256
|
|
//; CHECK_AVX2_I32X16: llvm.x86.avx2.gather.d.ps.256
|
|
//; CHECK_AVX2_I32X16: llvm.x86.avx2.gather.d.ps.256
|
|
//; CHECK_AVX2_I32X16: llvm.x86.avx2.gather.d.ps.256
|
|
//; CHECK_AVX2_I32X16-NOT: llvm.x86.avx2.gather.d.ps.256
|
|
|
|
//; CHECK_AVX512KNL_X16: llvm.x86.avx512.mask.gather.dps.512
|
|
//; CHECK_AVX512KNL_X16-NOT: llvm.x86.avx512.mask.gather.dps.512
|
|
|
|
//; CHECK_AVX512SKX_X16: llvm.x86.avx512.mask.gather.dps.512
|
|
//; CHECK_AVX512SKX_X16-NOT: llvm.x86.avx512.mask.gather.dps.512
|
|
if ((int)a & 1) {
|
|
#pragma ignore warning(perf)
|
|
RET[programIndex] = foo.x[a - 1];
|
|
return;
|
|
}
|
|
}
|