17eafe0841
If we lower a v2i64 shuffle to PSHUFD, we currently clamp undef elements to 0, (elements 0,1 of the v4i32) which can result in the shuffle referencing more elements of the source vector than expected, affecting later shuffle combines and KnownBits/SimplifyDemanded calls. By ensuring we widen the undef mask element we allow getV4X86ShuffleImm8 to use inline elements as the default, which are more likely to fold.
72 lines
2.5 KiB
LLVM
72 lines
2.5 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
|
|
; RUN: llc < %s -mtriple=i686-unknown -mattr=+sse2 | FileCheck %s --check-prefix=X32-SSE
|
|
; RUN: llc < %s -mtriple=i686-unknown -mattr=+avx | FileCheck %s --check-prefix=X32-AVX
|
|
; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+sse2 | FileCheck %s --check-prefix=X64-SSE
|
|
; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+avx | FileCheck %s --check-prefix=X64-AVX
|
|
|
|
;PR29079
|
|
|
|
define <4 x float> @mask_ucvt_4i32_4f32(<4 x i32> %a) {
|
|
; X32-SSE-LABEL: mask_ucvt_4i32_4f32:
|
|
; X32-SSE: # %bb.0:
|
|
; X32-SSE-NEXT: andps {{\.LCPI.*}}, %xmm0
|
|
; X32-SSE-NEXT: cvtdq2ps %xmm0, %xmm0
|
|
; X32-SSE-NEXT: retl
|
|
;
|
|
; X32-AVX-LABEL: mask_ucvt_4i32_4f32:
|
|
; X32-AVX: # %bb.0:
|
|
; X32-AVX-NEXT: vandps {{\.LCPI.*}}, %xmm0, %xmm0
|
|
; X32-AVX-NEXT: vcvtdq2ps %xmm0, %xmm0
|
|
; X32-AVX-NEXT: retl
|
|
;
|
|
; X64-SSE-LABEL: mask_ucvt_4i32_4f32:
|
|
; X64-SSE: # %bb.0:
|
|
; X64-SSE-NEXT: andps {{.*}}(%rip), %xmm0
|
|
; X64-SSE-NEXT: cvtdq2ps %xmm0, %xmm0
|
|
; X64-SSE-NEXT: retq
|
|
;
|
|
; X64-AVX-LABEL: mask_ucvt_4i32_4f32:
|
|
; X64-AVX: # %bb.0:
|
|
; X64-AVX-NEXT: vandps {{.*}}(%rip), %xmm0, %xmm0
|
|
; X64-AVX-NEXT: vcvtdq2ps %xmm0, %xmm0
|
|
; X64-AVX-NEXT: retq
|
|
%and = and <4 x i32> %a, <i32 127, i32 255, i32 4095, i32 65595>
|
|
%cvt = uitofp <4 x i32> %and to <4 x float>
|
|
ret <4 x float> %cvt
|
|
}
|
|
|
|
define <4 x double> @mask_ucvt_4i32_4f64(<4 x i32> %a) {
|
|
; X32-SSE-LABEL: mask_ucvt_4i32_4f64:
|
|
; X32-SSE: # %bb.0:
|
|
; X32-SSE-NEXT: pand {{\.LCPI.*}}, %xmm0
|
|
; X32-SSE-NEXT: cvtdq2pd %xmm0, %xmm2
|
|
; X32-SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[2,3,2,3]
|
|
; X32-SSE-NEXT: cvtdq2pd %xmm0, %xmm1
|
|
; X32-SSE-NEXT: movaps %xmm2, %xmm0
|
|
; X32-SSE-NEXT: retl
|
|
;
|
|
; X32-AVX-LABEL: mask_ucvt_4i32_4f64:
|
|
; X32-AVX: # %bb.0:
|
|
; X32-AVX-NEXT: vandps {{\.LCPI.*}}, %xmm0, %xmm0
|
|
; X32-AVX-NEXT: vcvtdq2pd %xmm0, %ymm0
|
|
; X32-AVX-NEXT: retl
|
|
;
|
|
; X64-SSE-LABEL: mask_ucvt_4i32_4f64:
|
|
; X64-SSE: # %bb.0:
|
|
; X64-SSE-NEXT: pand {{.*}}(%rip), %xmm0
|
|
; X64-SSE-NEXT: cvtdq2pd %xmm0, %xmm2
|
|
; X64-SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[2,3,2,3]
|
|
; X64-SSE-NEXT: cvtdq2pd %xmm0, %xmm1
|
|
; X64-SSE-NEXT: movaps %xmm2, %xmm0
|
|
; X64-SSE-NEXT: retq
|
|
;
|
|
; X64-AVX-LABEL: mask_ucvt_4i32_4f64:
|
|
; X64-AVX: # %bb.0:
|
|
; X64-AVX-NEXT: vandps {{.*}}(%rip), %xmm0, %xmm0
|
|
; X64-AVX-NEXT: vcvtdq2pd %xmm0, %ymm0
|
|
; X64-AVX-NEXT: retq
|
|
%and = and <4 x i32> %a, <i32 127, i32 255, i32 4095, i32 65595>
|
|
%cvt = uitofp <4 x i32> %and to <4 x double>
|
|
ret <4 x double> %cvt
|
|
}
|