1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2; RUN: llc < %s -mtriple=x86_64-apple-darwin -mattr=+avx | FileCheck %s --check-prefix=AVX --check-prefix=AVX1 3; RUN: llc < %s -mtriple=x86_64-apple-darwin -mattr=+avx2 | FileCheck %s --check-prefix=AVX --check-prefix=AVX2 4 5; Prefer a blend instruction to a vinsert128 instruction because blends 6; are simpler (no lane changes) and therefore will have equal or better 7; performance. 8 9define <8 x float> @castA(<4 x float> %m) nounwind uwtable readnone ssp { 10; AVX-LABEL: castA: 11; AVX: ## BB#0: 12; AVX-NEXT: ## kill: %XMM0<def> %XMM0<kill> %YMM0<def> 13; AVX-NEXT: vxorps %ymm1, %ymm1, %ymm1 14; AVX-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7] 15; AVX-NEXT: retq 16 %shuffle.i = shufflevector <4 x float> %m, <4 x float> zeroinitializer, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 4, i32 4, i32 4> 17 ret <8 x float> %shuffle.i 18} 19 20define <4 x double> @castB(<2 x double> %m) nounwind uwtable readnone ssp { 21; AVX-LABEL: castB: 22; AVX: ## BB#0: 23; AVX-NEXT: ## kill: %XMM0<def> %XMM0<kill> %YMM0<def> 24; AVX-NEXT: vxorpd %ymm1, %ymm1, %ymm1 25; AVX-NEXT: vblendpd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3] 26; AVX-NEXT: retq 27 %shuffle.i = shufflevector <2 x double> %m, <2 x double> zeroinitializer, <4 x i32> <i32 0, i32 1, i32 2, i32 2> 28 ret <4 x double> %shuffle.i 29} 30 31; AVX2 is needed for integer types. 32 33define <4 x i64> @castC(<2 x i64> %m) nounwind uwtable readnone ssp { 34; AVX1-LABEL: castC: 35; AVX1: ## BB#0: 36; AVX1-NEXT: ## kill: %XMM0<def> %XMM0<kill> %YMM0<def> 37; AVX1-NEXT: vxorpd %ymm1, %ymm1, %ymm1 38; AVX1-NEXT: vblendpd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3] 39; AVX1-NEXT: retq 40; 41; AVX2-LABEL: castC: 42; AVX2: ## BB#0: 43; AVX2-NEXT: ## kill: %XMM0<def> %XMM0<kill> %YMM0<def> 44; AVX2-NEXT: vpxor %ymm1, %ymm1, %ymm1 45; AVX2-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7] 46; AVX2-NEXT: retq 47 %shuffle.i = shufflevector <2 x i64> %m, <2 x i64> zeroinitializer, <4 x i32> <i32 0, i32 1, i32 2, i32 2> 48 ret <4 x i64> %shuffle.i 49} 50 51; The next three tests don't need any shuffling. There may or may not be a 52; vzeroupper before the return, so just check for the absence of shuffles. 53 54define <4 x float> @castD(<8 x float> %m) nounwind uwtable readnone ssp { 55; AVX-LABEL: castD: 56; AVX: ## BB#0: 57; AVX-NEXT: ## kill: %XMM0<def> %XMM0<kill> %YMM0<kill> 58; AVX-NEXT: vzeroupper 59; AVX-NEXT: retq 60 %shuffle.i = shufflevector <8 x float> %m, <8 x float> %m, <4 x i32> <i32 0, i32 1, i32 2, i32 3> 61 ret <4 x float> %shuffle.i 62} 63 64define <2 x i64> @castE(<4 x i64> %m) nounwind uwtable readnone ssp { 65; AVX-LABEL: castE: 66; AVX: ## BB#0: 67; AVX-NEXT: ## kill: %XMM0<def> %XMM0<kill> %YMM0<kill> 68; AVX-NEXT: vzeroupper 69; AVX-NEXT: retq 70 %shuffle.i = shufflevector <4 x i64> %m, <4 x i64> %m, <2 x i32> <i32 0, i32 1> 71 ret <2 x i64> %shuffle.i 72} 73 74define <2 x double> @castF(<4 x double> %m) nounwind uwtable readnone ssp { 75; AVX-LABEL: castF: 76; AVX: ## BB#0: 77; AVX-NEXT: ## kill: %XMM0<def> %XMM0<kill> %YMM0<kill> 78; AVX-NEXT: vzeroupper 79; AVX-NEXT: retq 80 %shuffle.i = shufflevector <4 x double> %m, <4 x double> %m, <2 x i32> <i32 0, i32 1> 81 ret <2 x double> %shuffle.i 82} 83 84