1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2; RUN: llc < %s -fast-isel -mtriple=i386-unknown-unknown -mattr=+sse4a | FileCheck %s --check-prefixes=CHECK,X86 3; RUN: llc < %s -fast-isel -mtriple=i386-unknown-unknown -mattr=+sse4a,+avx | FileCheck %s --check-prefixes=CHECK,X86 4; RUN: llc < %s -fast-isel -mtriple=x86_64-unknown-unknown -mattr=+sse4a | FileCheck %s --check-prefixes=CHECK,X64 5; RUN: llc < %s -fast-isel -mtriple=x86_64-unknown-unknown -mattr=+sse4a,+avx | FileCheck %s --check-prefixes=CHECK,X64 6 7; NOTE: This should use IR equivalent to what is generated by clang/test/CodeGen/sse4a-builtins.c 8 9define <2 x i64> @test_mm_extracti_si64(<2 x i64> %x) { 10; CHECK-LABEL: test_mm_extracti_si64: 11; CHECK: # %bb.0: 12; CHECK-NEXT: extrq $2, $3, %xmm0 13; CHECK-NEXT: ret{{[l|q]}} 14 %res = call <2 x i64> @llvm.x86.sse4a.extrqi(<2 x i64> %x, i8 3, i8 2) 15 ret <2 x i64> %res 16} 17declare <2 x i64> @llvm.x86.sse4a.extrqi(<2 x i64>, i8, i8) nounwind readnone 18 19define <2 x i64> @test_mm_extract_si64(<2 x i64> %x, <2 x i64> %y) { 20; CHECK-LABEL: test_mm_extract_si64: 21; CHECK: # %bb.0: 22; CHECK-NEXT: extrq %xmm1, %xmm0 23; CHECK-NEXT: ret{{[l|q]}} 24 %bc = bitcast <2 x i64> %y to <16 x i8> 25 %res = call <2 x i64> @llvm.x86.sse4a.extrq(<2 x i64> %x, <16 x i8> %bc) 26 ret <2 x i64> %res 27} 28declare <2 x i64> @llvm.x86.sse4a.extrq(<2 x i64>, <16 x i8>) nounwind readnone 29 30define <2 x i64> @test_mm_inserti_si64(<2 x i64> %x, <2 x i64> %y) { 31; CHECK-LABEL: test_mm_inserti_si64: 32; CHECK: # %bb.0: 33; CHECK-NEXT: insertq $6, $5, %xmm1, %xmm0 34; CHECK-NEXT: ret{{[l|q]}} 35 %res = call <2 x i64> @llvm.x86.sse4a.insertqi(<2 x i64> %x, <2 x i64> %y, i8 5, i8 6) 36 ret <2 x i64> %res 37} 38declare <2 x i64> @llvm.x86.sse4a.insertqi(<2 x i64>, <2 x i64>, i8, i8) nounwind readnone 39 40define <2 x i64> @test_mm_insert_si64(<2 x i64> %x, <2 x i64> %y) { 41; CHECK-LABEL: test_mm_insert_si64: 42; CHECK: # %bb.0: 43; CHECK-NEXT: insertq %xmm1, %xmm0 44; CHECK-NEXT: ret{{[l|q]}} 45 %res = call <2 x i64> @llvm.x86.sse4a.insertq(<2 x i64> %x, <2 x i64> %y) 46 ret <2 x i64> %res 47} 48declare <2 x i64> @llvm.x86.sse4a.insertq(<2 x i64>, <2 x i64>) nounwind readnone 49 50define void @test_stream_sd(double* %p, <2 x double> %a) { 51; X86-LABEL: test_stream_sd: 52; X86: # %bb.0: 53; X86-NEXT: movl {{[0-9]+}}(%esp), %eax 54; X86-NEXT: movntsd %xmm0, (%eax) 55; X86-NEXT: retl 56; 57; X64-LABEL: test_stream_sd: 58; X64: # %bb.0: 59; X64-NEXT: movntsd %xmm0, (%rdi) 60; X64-NEXT: retq 61 %1 = extractelement <2 x double> %a, i64 0 62 store double %1, double* %p, align 1, !nontemporal !1 63 ret void 64} 65 66define void @test_mm_stream_ss(float* %p, <4 x float> %a) { 67; X86-LABEL: test_mm_stream_ss: 68; X86: # %bb.0: 69; X86-NEXT: movl {{[0-9]+}}(%esp), %eax 70; X86-NEXT: movntss %xmm0, (%eax) 71; X86-NEXT: retl 72; 73; X64-LABEL: test_mm_stream_ss: 74; X64: # %bb.0: 75; X64-NEXT: movntss %xmm0, (%rdi) 76; X64-NEXT: retq 77 %1 = extractelement <4 x float> %a, i64 0 78 store float %1, float* %p, align 1, !nontemporal !1 79 ret void 80} 81 82!1 = !{i32 1} 83