110 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1 | FileCheck %s --check-prefix=SSE3; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=AVX,AVX14; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX25 6declare <16 x i8> @llvm.x86.sse2.pavg.b(<16 x i8>, <16 x i8>) nounwind readnone7declare <8 x i16> @llvm.x86.sse2.pavg.w(<8 x i16>, <8 x i16>) nounwind readnone8 9define <16 x i8> @combine_pavgb_self(<16 x i8> %a0) {10; SSE-LABEL: combine_pavgb_self:11; SSE: # %bb.0:12; SSE-NEXT: retq13;14; AVX-LABEL: combine_pavgb_self:15; AVX: # %bb.0:16; AVX-NEXT: retq17 %1 = call <16 x i8> @llvm.x86.sse2.pavg.b(<16 x i8> %a0, <16 x i8> %a0)18 ret <16 x i8> %119}20 21define <16 x i8> @combine_pavgb_zero(<16 x i8> %a0) {22; SSE-LABEL: combine_pavgb_zero:23; SSE: # %bb.0:24; SSE-NEXT: pxor %xmm1, %xmm125; SSE-NEXT: pavgb %xmm1, %xmm026; SSE-NEXT: retq27;28; AVX-LABEL: combine_pavgb_zero:29; AVX: # %bb.0:30; AVX-NEXT: vpxor %xmm1, %xmm1, %xmm131; AVX-NEXT: vpavgb %xmm1, %xmm0, %xmm032; AVX-NEXT: retq33 %1 = call <16 x i8> @llvm.x86.sse2.pavg.b(<16 x i8> zeroinitializer, <16 x i8> %a0)34 ret <16 x i8> %135}36 37define <16 x i8> @combine_pavgw_knownbits(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> %a2, <8 x i16> %a3) {38; SSE-LABEL: combine_pavgw_knownbits:39; SSE: # %bb.0:40; SSE-NEXT: pmovsxbw {{.*#+}} xmm4 = [31,31,31,31,31,31,31,31]41; SSE-NEXT: pand %xmm4, %xmm042; SSE-NEXT: pand %xmm4, %xmm143; SSE-NEXT: pavgw %xmm1, %xmm044; SSE-NEXT: pand %xmm4, %xmm245; SSE-NEXT: pand %xmm4, %xmm346; SSE-NEXT: pavgw %xmm2, %xmm347; SSE-NEXT: packuswb %xmm3, %xmm048; SSE-NEXT: retq49;50; AVX1-LABEL: combine_pavgw_knownbits:51; AVX1: # %bb.0:52; AVX1-NEXT: vbroadcastss {{.*#+}} xmm4 = [31,31,31,31,31,31,31,31]53; AVX1-NEXT: vpand %xmm4, %xmm0, %xmm054; AVX1-NEXT: vpand %xmm4, %xmm1, %xmm155; AVX1-NEXT: vpavgw %xmm1, %xmm0, %xmm056; AVX1-NEXT: vpand %xmm4, %xmm2, %xmm157; AVX1-NEXT: vpand %xmm4, %xmm3, %xmm258; AVX1-NEXT: vpavgw %xmm2, %xmm1, %xmm159; AVX1-NEXT: vpackuswb %xmm1, %xmm0, %xmm060; AVX1-NEXT: retq61;62; AVX2-LABEL: combine_pavgw_knownbits:63; AVX2: # %bb.0:64; AVX2-NEXT: vpbroadcastw {{.*#+}} xmm4 = [31,31,31,31,31,31,31,31]65; AVX2-NEXT: vpand %xmm4, %xmm0, %xmm066; AVX2-NEXT: vpand %xmm4, %xmm1, %xmm167; AVX2-NEXT: vpavgw %xmm1, %xmm0, %xmm068; AVX2-NEXT: vpand %xmm4, %xmm2, %xmm169; AVX2-NEXT: vpand %xmm4, %xmm3, %xmm270; AVX2-NEXT: vpavgw %xmm2, %xmm1, %xmm171; AVX2-NEXT: vpackuswb %xmm1, %xmm0, %xmm072; AVX2-NEXT: retq73 %m0 = and <8 x i16> %a0, <i16 31, i16 31, i16 31, i16 31, i16 31, i16 31, i16 31, i16 31>74 %m1 = and <8 x i16> %a1, <i16 31, i16 31, i16 31, i16 31, i16 31, i16 31, i16 31, i16 31>75 %m2 = and <8 x i16> %a2, <i16 31, i16 31, i16 31, i16 31, i16 31, i16 31, i16 31, i16 31>76 %m3 = and <8 x i16> %a3, <i16 31, i16 31, i16 31, i16 31, i16 31, i16 31, i16 31, i16 31>77 %avg01 = tail call <8 x i16> @llvm.x86.sse2.pavg.w(<8 x i16> %m0, <8 x i16> %m1)78 %avg23 = tail call <8 x i16> @llvm.x86.sse2.pavg.w(<8 x i16> %m2, <8 x i16> %m3)79 %shuffle = shufflevector <8 x i16> %avg01, <8 x i16> %avg23, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>80 %trunc = trunc <16 x i16> %shuffle to <16 x i8>81 ret <16 x i8> %trunc82}83 84define <8 x i16> @combine_pavgw_demandedelts(<8 x i16> %a0, <8 x i16> %a1) {85; SSE-LABEL: combine_pavgw_demandedelts:86; SSE: # %bb.0:87; SSE-NEXT: pavgw %xmm1, %xmm088; SSE-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]89; SSE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,1,0,1]90; SSE-NEXT: retq91;92; AVX1-LABEL: combine_pavgw_demandedelts:93; AVX1: # %bb.0:94; AVX1-NEXT: vpavgw %xmm1, %xmm0, %xmm095; AVX1-NEXT: vpshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]96; AVX1-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[0,1,0,1]97; AVX1-NEXT: retq98;99; AVX2-LABEL: combine_pavgw_demandedelts:100; AVX2: # %bb.0:101; AVX2-NEXT: vpavgw %xmm1, %xmm0, %xmm0102; AVX2-NEXT: vpbroadcastw %xmm0, %xmm0103; AVX2-NEXT: retq104 %s0 = shufflevector <8 x i16> %a0, <8 x i16> poison, <8 x i32> <i32 0, i32 0, i32 2, i32 2, i32 4, i32 4, i32 6, i32 6>105 %avg = tail call <8 x i16> @llvm.x86.sse2.pavg.w(<8 x i16> %s0, <8 x i16> %a1)106 %shuffle = shufflevector <8 x i16> %avg, <8 x i16> poison, <8 x i32> zeroinitializer107 ret <8 x i16> %shuffle108}109 110