1; RUN: llc < %s -mtriple=x86_64-unknown-unknown | FileCheck %s --check-prefix=ALL --check-prefix=SSE2 2; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=sse4.1 | FileCheck %s --check-prefix=ALL --check-prefix=SSE41 3 4define <4 x i32> @a(<4 x i32> %i) nounwind { 5; SSE2-LABEL: a: 6; SSE2: # BB#0: # %entry 7; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [117,117,117,117] 8; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3] 9; SSE2-NEXT: pmuludq %xmm1, %xmm0 10; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3] 11; SSE2-NEXT: pmuludq %xmm1, %xmm2 12; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm2[0,2,2,3] 13; SSE2-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1] 14; SSE2-NEXT: retq 15; 16; SSE41-LABEL: a: 17; SSE41: # BB#0: # %entry 18; SSE41-NEXT: pmulld {{.*}}(%rip), %xmm0 19; SSE41-NEXT: retq 20entry: 21 %A = mul <4 x i32> %i, < i32 117, i32 117, i32 117, i32 117 > 22 ret <4 x i32> %A 23} 24 25define <2 x i64> @b(<2 x i64> %i) nounwind { 26; ALL-LABEL: b: 27; ALL: # BB#0: # %entry 28; ALL-NEXT: movdqa {{.*#+}} xmm1 = [117,117] 29; ALL-NEXT: movdqa %xmm0, %xmm2 30; ALL-NEXT: pmuludq %xmm1, %xmm2 31; ALL-NEXT: pxor %xmm3, %xmm3 32; ALL-NEXT: pmuludq %xmm0, %xmm3 33; ALL-NEXT: psllq $32, %xmm3 34; ALL-NEXT: paddq %xmm3, %xmm2 35; ALL-NEXT: psrlq $32, %xmm0 36; ALL-NEXT: pmuludq %xmm1, %xmm0 37; ALL-NEXT: psllq $32, %xmm0 38; ALL-NEXT: paddq %xmm2, %xmm0 39; ALL-NEXT: retq 40entry: 41 %A = mul <2 x i64> %i, < i64 117, i64 117 > 42 ret <2 x i64> %A 43} 44 45define <4 x i32> @c(<4 x i32> %i, <4 x i32> %j) nounwind { 46; SSE2-LABEL: c: 47; SSE2: # BB#0: # %entry 48; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3] 49; SSE2-NEXT: pmuludq %xmm1, %xmm0 50; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3] 51; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3] 52; SSE2-NEXT: pmuludq %xmm2, %xmm1 53; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3] 54; SSE2-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1] 55; SSE2-NEXT: retq 56; 57; SSE41-LABEL: c: 58; SSE41: # BB#0: # %entry 59; SSE41-NEXT: pmulld %xmm1, %xmm0 60; SSE41-NEXT: retq 61entry: 62 %A = mul <4 x i32> %i, %j 63 ret <4 x i32> %A 64} 65 66define <2 x i64> @d(<2 x i64> %i, <2 x i64> %j) nounwind { 67; ALL-LABEL: d: 68; ALL: # BB#0: # %entry 69; ALL-NEXT: movdqa %xmm0, %xmm2 70; ALL-NEXT: pmuludq %xmm1, %xmm2 71; ALL-NEXT: movdqa %xmm1, %xmm3 72; ALL-NEXT: psrlq $32, %xmm3 73; ALL-NEXT: pmuludq %xmm0, %xmm3 74; ALL-NEXT: psllq $32, %xmm3 75; ALL-NEXT: paddq %xmm3, %xmm2 76; ALL-NEXT: psrlq $32, %xmm0 77; ALL-NEXT: pmuludq %xmm1, %xmm0 78; ALL-NEXT: psllq $32, %xmm0 79; ALL-NEXT: paddq %xmm2, %xmm0 80; ALL-NEXT: retq 81entry: 82 %A = mul <2 x i64> %i, %j 83 ret <2 x i64> %A 84} 85 86declare void @foo() 87 88define <4 x i32> @e(<4 x i32> %i, <4 x i32> %j) nounwind { 89; SSE2-LABEL: e: 90; SSE2: # BB#0: # %entry 91; SSE2-NEXT: subq $40, %rsp 92; SSE2-NEXT: movaps %xmm1, {{[0-9]+}}(%rsp) # 16-byte Spill 93; SSE2-NEXT: movaps %xmm0, (%rsp) # 16-byte Spill 94; SSE2-NEXT: callq foo 95; SSE2-NEXT: movdqa (%rsp), %xmm0 # 16-byte Reload 96; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[1,1,3,3] 97; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm2 # 16-byte Reload 98; SSE2-NEXT: pmuludq %xmm2, %xmm0 99; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3] 100; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3] 101; SSE2-NEXT: pmuludq %xmm1, %xmm2 102; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm2[0,2,2,3] 103; SSE2-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1] 104; SSE2-NEXT: addq $40, %rsp 105; SSE2-NEXT: retq 106; 107; SSE41-LABEL: e: 108; SSE41: # BB#0: # %entry 109; SSE41-NEXT: subq $40, %rsp 110; SSE41-NEXT: movaps %xmm1, {{[0-9]+}}(%rsp) # 16-byte Spill 111; SSE41-NEXT: movaps %xmm0, (%rsp) # 16-byte Spill 112; SSE41-NEXT: callq foo 113; SSE41-NEXT: movdqa (%rsp), %xmm0 # 16-byte Reload 114; SSE41-NEXT: pmulld {{[0-9]+}}(%rsp), %xmm0 # 16-byte Folded Reload 115; SSE41-NEXT: addq $40, %rsp 116; SSE41-NEXT: retq 117entry: 118 ; Use a call to force spills. 119 call void @foo() 120 %A = mul <4 x i32> %i, %j 121 ret <4 x i32> %A 122} 123 124define <2 x i64> @f(<2 x i64> %i, <2 x i64> %j) nounwind { 125; ALL-LABEL: f: 126; ALL: # BB#0: # %entry 127; ALL-NEXT: subq $40, %rsp 128; ALL-NEXT: movaps %xmm1, {{[0-9]+}}(%rsp) # 16-byte Spill 129; ALL-NEXT: movaps %xmm0, (%rsp) # 16-byte Spill 130; ALL-NEXT: callq foo 131; ALL-NEXT: movdqa (%rsp), %xmm0 # 16-byte Reload 132; ALL-NEXT: movdqa %xmm0, %xmm2 133; ALL-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm3 # 16-byte Reload 134; ALL-NEXT: pmuludq %xmm3, %xmm2 135; ALL-NEXT: movdqa %xmm3, %xmm1 136; ALL-NEXT: psrlq $32, %xmm1 137; ALL-NEXT: pmuludq %xmm0, %xmm1 138; ALL-NEXT: psllq $32, %xmm1 139; ALL-NEXT: paddq %xmm1, %xmm2 140; ALL-NEXT: psrlq $32, %xmm0 141; ALL-NEXT: pmuludq %xmm3, %xmm0 142; ALL-NEXT: psllq $32, %xmm0 143; ALL-NEXT: paddq %xmm2, %xmm0 144; ALL-NEXT: addq $40, %rsp 145; ALL-NEXT: retq 146entry: 147 ; Use a call to force spills. 148 call void @foo() 149 %A = mul <2 x i64> %i, %j 150 ret <2 x i64> %A 151} 152