; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=icelake-server | FileCheck %s define <2 x i64> @test_extract_pclmulqdq_v4i64_v2i64(<4 x i64> %a0, <4 x i64> %a1) { ; CHECK-LABEL: test_extract_pclmulqdq_v4i64_v2i64: ; CHECK: # %bb.0: ; CHECK-NEXT: vpclmulqdq $1, %xmm1, %xmm0, %xmm0 ; CHECK-NEXT: vzeroupper ; CHECK-NEXT: retq %clmul = call <4 x i64> @llvm.x86.pclmulqdq.256(<4 x i64> %a0, <4 x i64> %a1, i8 1) %res = shufflevector <4 x i64> %clmul, <4 x i64> poison, <2 x i32> ret <2 x i64> %res } define <4 x i64> @test_extract_pclmulqdq_v8i64_v4i64(<8 x i64> %a0, <8 x i64> %a1) { ; CHECK-LABEL: test_extract_pclmulqdq_v8i64_v4i64: ; CHECK: # %bb.0: ; CHECK-NEXT: vpclmulqdq $1, %ymm1, %ymm0, %ymm0 ; CHECK-NEXT: retq %clmul = call <8 x i64> @llvm.x86.pclmulqdq.512(<8 x i64> %a0, <8 x i64> %a1, i8 1) %res = shufflevector <8 x i64> %clmul, <8 x i64> poison, <4 x i32> ret <4 x i64> %res } define <4 x i64> @test_concat_pclmulqdq_v4i64_v2i64(<4 x i64> %a0, <4 x i64> %a1) { ; CHECK-LABEL: test_concat_pclmulqdq_v4i64_v2i64: ; CHECK: # %bb.0: ; CHECK-NEXT: vpclmulqdq $0, %ymm1, %ymm0, %ymm0 ; CHECK-NEXT: retq %lo0 = shufflevector <4 x i64> %a0, <4 x i64> poison, <2 x i32> %lo1 = shufflevector <4 x i64> %a1, <4 x i64> poison, <2 x i32> %hi0 = shufflevector <4 x i64> %a0, <4 x i64> poison, <2 x i32> %hi1 = shufflevector <4 x i64> %a1, <4 x i64> poison, <2 x i32> %lo = call <2 x i64> @llvm.x86.pclmulqdq(<2 x i64> %lo0, <2 x i64> %lo1, i8 0) %hi = call <2 x i64> @llvm.x86.pclmulqdq(<2 x i64> %hi0, <2 x i64> %hi1, i8 0) %res = shufflevector <2 x i64> %lo, <2 x i64> %hi, <4 x i32> ret <4 x i64> %res } define <8 x i64> @test_concat_pclmulqdq_v8i64_v2i64(<8 x i64> %a0, <8 x i64> %a1) { ; CHECK-LABEL: test_concat_pclmulqdq_v8i64_v2i64: ; CHECK: # %bb.0: ; CHECK-NEXT: vpclmulqdq $1, %zmm1, %zmm0, %zmm0 ; CHECK-NEXT: retq %x0 = shufflevector <8 x i64> %a0, <8 x i64> poison, <2 x i32> %x1 = shufflevector <8 x i64> %a1, <8 x i64> poison, <2 x i32> %y0 = shufflevector <8 x i64> %a0, <8 x i64> poison, <2 x i32> %y1 = shufflevector <8 x i64> %a1, <8 x i64> poison, <2 x i32> %z0 = shufflevector <8 x i64> %a0, <8 x i64> poison, <2 x i32> %z1 = shufflevector <8 x i64> %a1, <8 x i64> poison, <2 x i32> %w0 = shufflevector <8 x i64> %a0, <8 x i64> poison, <2 x i32> %w1 = shufflevector <8 x i64> %a1, <8 x i64> poison, <2 x i32> %x = call <2 x i64> @llvm.x86.pclmulqdq(<2 x i64> %x0, <2 x i64> %x1, i8 1) %y = call <2 x i64> @llvm.x86.pclmulqdq(<2 x i64> %y0, <2 x i64> %y1, i8 1) %z = call <2 x i64> @llvm.x86.pclmulqdq(<2 x i64> %z0, <2 x i64> %z1, i8 1) %w = call <2 x i64> @llvm.x86.pclmulqdq(<2 x i64> %w0, <2 x i64> %w1, i8 1) %lo = shufflevector <2 x i64> %x, <2 x i64> %y, <4 x i32> %hi = shufflevector <2 x i64> %z, <2 x i64> %w, <4 x i32> %res = shufflevector <4 x i64> %lo, <4 x i64> %hi, <8 x i32> ret <8 x i64> %res } define <8 x i64> @test_concat_pclmulqdq_v8i64_v4i64(<8 x i64> %a0, <8 x i64> %a1) { ; CHECK-LABEL: test_concat_pclmulqdq_v8i64_v4i64: ; CHECK: # %bb.0: ; CHECK-NEXT: vpclmulqdq $17, %zmm1, %zmm0, %zmm0 ; CHECK-NEXT: retq %lo0 = shufflevector <8 x i64> %a0, <8 x i64> poison, <4 x i32> %lo1 = shufflevector <8 x i64> %a1, <8 x i64> poison, <4 x i32> %hi0 = shufflevector <8 x i64> %a0, <8 x i64> poison, <4 x i32> %hi1 = shufflevector <8 x i64> %a1, <8 x i64> poison, <4 x i32> %lo = call <4 x i64> @llvm.x86.pclmulqdq.256(<4 x i64> %lo0, <4 x i64> %lo1, i8 17) %hi = call <4 x i64> @llvm.x86.pclmulqdq.256(<4 x i64> %hi0, <4 x i64> %hi1, i8 17) %res = shufflevector <4 x i64> %lo, <4 x i64> %hi, <8 x i32> ret <8 x i64> %res } define <2 x i64> @test_shuffle_pclmulqdq_v2i64(<2 x i64> %a0, <2 x i64> %a1) { ; CHECK-LABEL: test_shuffle_pclmulqdq_v2i64: ; CHECK: # %bb.0: ; CHECK-NEXT: vpclmulqdq $16, %xmm1, %xmm0, %xmm0 ; CHECK-NEXT: retq %s0 = shufflevector <2 x i64> %a0, <2 x i64> poison, <2 x i32> %s1 = shufflevector <2 x i64> %a1, <2 x i64> poison, <2 x i32> %clmul = call <2 x i64> @llvm.x86.pclmulqdq(<2 x i64> %s0, <2 x i64> %s1, i8 1) ret <2 x i64> %clmul } define <4 x i64> @test_shuffle_pclmulqdq_v4i64(<4 x i64> %a0, <4 x i64> %a1) { ; CHECK-LABEL: test_shuffle_pclmulqdq_v4i64: ; CHECK: # %bb.0: ; CHECK-NEXT: vpclmulqdq $1, %ymm1, %ymm0, %ymm0 ; CHECK-NEXT: retq %s0 = shufflevector <4 x i64> %a0, <4 x i64> poison, <4 x i32> %s1 = shufflevector <4 x i64> %a1, <4 x i64> poison, <4 x i32> %clmul = call <4 x i64> @llvm.x86.pclmulqdq.256(<4 x i64> %s0, <4 x i64> %s1, i8 16) ret <4 x i64> %clmul } define <8 x i64> @test_shuffle_pclmulqdq_v8i64(<8 x i64> %a0, <8 x i64> %a1) { ; CHECK-LABEL: test_shuffle_pclmulqdq_v8i64: ; CHECK: # %bb.0: ; CHECK-NEXT: vpclmulqdq $0, %zmm1, %zmm0, %zmm0 ; CHECK-NEXT: retq %s0 = shufflevector <8 x i64> %a0, <8 x i64> poison, <8 x i32> %s1 = shufflevector <8 x i64> %a1, <8 x i64> poison, <8 x i32> %clmul = call <8 x i64> @llvm.x86.pclmulqdq.512(<8 x i64> %s0, <8 x i64> %s1, i8 17) ret <8 x i64> %clmul } define <2 x i64> @test_demanded_elts_pclmulqdq_0(<2 x i64> %a0, <2 x i64> %a1, i64 %s0, i64 %s1) { ; CHECK-LABEL: test_demanded_elts_pclmulqdq_0: ; CHECK: # %bb.0: ; CHECK-NEXT: vpclmulqdq $0, %xmm1, %xmm0, %xmm0 ; CHECK-NEXT: retq %1 = insertelement <2 x i64> %a0, i64 %s0, i64 1 %2 = insertelement <2 x i64> %a1, i64 %s1, i64 1 %3 = call <2 x i64> @llvm.x86.pclmulqdq(<2 x i64> %1, <2 x i64> %2, i8 0) ret <2 x i64> %3 } define <2 x i64> @test_demanded_elts_pclmulqdq_1(<2 x i64> %a0, <2 x i64> %a1, i64 %s0, i64 %s1) { ; CHECK-LABEL: test_demanded_elts_pclmulqdq_1: ; CHECK: # %bb.0: ; CHECK-NEXT: vpinsrq $1, %rdi, %xmm0, %xmm0 ; CHECK-NEXT: vpclmulqdq $1, %xmm1, %xmm0, %xmm0 ; CHECK-NEXT: retq %1 = insertelement <2 x i64> %a0, i64 %s0, i64 1 %2 = insertelement <2 x i64> %a1, i64 %s1, i64 1 %3 = call <2 x i64> @llvm.x86.pclmulqdq(<2 x i64> %1, <2 x i64> %2, i8 1) ret <2 x i64> %3 } define <2 x i64> @test_demanded_elts_pclmulqdq_16(<2 x i64> %a0, <2 x i64> %a1, i64 %s0, i64 %s1) { ; CHECK-LABEL: test_demanded_elts_pclmulqdq_16: ; CHECK: # %bb.0: ; CHECK-NEXT: vpinsrq $1, %rsi, %xmm0, %xmm1 ; CHECK-NEXT: vpclmulqdq $16, %xmm1, %xmm0, %xmm0 ; CHECK-NEXT: retq %1 = insertelement <2 x i64> %a0, i64 %s0, i64 1 %2 = insertelement <2 x i64> %a1, i64 %s1, i64 1 %3 = call <2 x i64> @llvm.x86.pclmulqdq(<2 x i64> %1, <2 x i64> %2, i8 16) ret <2 x i64> %3 } define <2 x i64> @test_demanded_elts_pclmulqdq_17(<2 x i64> %a0, <2 x i64> %a1, i64 %s0, i64 %s1) { ; CHECK-LABEL: test_demanded_elts_pclmulqdq_17: ; CHECK: # %bb.0: ; CHECK-NEXT: vpinsrq $1, %rdi, %xmm0, %xmm0 ; CHECK-NEXT: vpinsrq $1, %rsi, %xmm0, %xmm1 ; CHECK-NEXT: vpclmulqdq $17, %xmm1, %xmm0, %xmm0 ; CHECK-NEXT: retq %1 = insertelement <2 x i64> %a0, i64 %s0, i64 1 %2 = insertelement <2 x i64> %a1, i64 %s1, i64 1 %3 = call <2 x i64> @llvm.x86.pclmulqdq(<2 x i64> %1, <2 x i64> %2, i8 17) ret <2 x i64> %3 } define <4 x i64> @test_demanded_elts_pclmulqdq_256_0(<4 x i64> %a0, <4 x i64> %a1, i64 %s0, i64 %s1) { ; CHECK-LABEL: test_demanded_elts_pclmulqdq_256_0: ; CHECK: # %bb.0: ; CHECK-NEXT: vpclmulqdq $0, %ymm1, %ymm0, %ymm0 ; CHECK-NEXT: retq %1 = insertelement <4 x i64> %a0, i64 %s0, i64 1 %2 = insertelement <4 x i64> %a1, i64 %s1, i64 1 %3 = insertelement <4 x i64> %1, i64 %s0, i64 3 %4 = insertelement <4 x i64> %2, i64 %s1, i64 3 %res = call <4 x i64> @llvm.x86.pclmulqdq.256(<4 x i64> %3, <4 x i64> %4, i8 0) ret <4 x i64> %res } define <4 x i64> @test_demanded_elts_pclmulqdq_256_1(<4 x i64> %a0, <4 x i64> %a1, i64 %s0, i64 %s1) { ; CHECK-LABEL: test_demanded_elts_pclmulqdq_256_1: ; CHECK: # %bb.0: ; CHECK-NEXT: vpinsrq $1, %rdi, %xmm0, %xmm0 ; CHECK-NEXT: vpbroadcastq %rdi, %ymm2 ; CHECK-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm2[6,7] ; CHECK-NEXT: vpclmulqdq $1, %ymm1, %ymm0, %ymm0 ; CHECK-NEXT: retq %1 = insertelement <4 x i64> %a0, i64 %s0, i64 1 %2 = insertelement <4 x i64> %a1, i64 %s1, i64 1 %3 = insertelement <4 x i64> %1, i64 %s0, i64 3 %4 = insertelement <4 x i64> %2, i64 %s1, i64 3 %res = call <4 x i64> @llvm.x86.pclmulqdq.256(<4 x i64> %3, <4 x i64> %4, i8 1) ret <4 x i64> %res } define <4 x i64> @test_demanded_elts_pclmulqdq_256_16(<4 x i64> %a0, <4 x i64> %a1, i64 %s0, i64 %s1) { ; CHECK-LABEL: test_demanded_elts_pclmulqdq_256_16: ; CHECK: # %bb.0: ; CHECK-NEXT: vpinsrq $1, %rsi, %xmm0, %xmm1 ; CHECK-NEXT: vpbroadcastq %rsi, %ymm2 ; CHECK-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6,7] ; CHECK-NEXT: vpclmulqdq $16, %ymm1, %ymm0, %ymm0 ; CHECK-NEXT: retq %1 = insertelement <4 x i64> %a0, i64 %s0, i64 1 %2 = insertelement <4 x i64> %a1, i64 %s1, i64 1 %3 = insertelement <4 x i64> %1, i64 %s0, i64 3 %4 = insertelement <4 x i64> %2, i64 %s1, i64 3 %res = call <4 x i64> @llvm.x86.pclmulqdq.256(<4 x i64> %3, <4 x i64> %4, i8 16) ret <4 x i64> %res } define <4 x i64> @test_demanded_elts_pclmulqdq_256_17(<4 x i64> %a0, <4 x i64> %a1, i64 %s0, i64 %s1) { ; CHECK-LABEL: test_demanded_elts_pclmulqdq_256_17: ; CHECK: # %bb.0: ; CHECK-NEXT: vpinsrq $1, %rdi, %xmm0, %xmm0 ; CHECK-NEXT: vpinsrq $1, %rsi, %xmm0, %xmm1 ; CHECK-NEXT: vpbroadcastq %rdi, %ymm2 ; CHECK-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm2[6,7] ; CHECK-NEXT: vpbroadcastq %rsi, %ymm2 ; CHECK-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4,5],ymm2[6,7] ; CHECK-NEXT: vpclmulqdq $17, %ymm1, %ymm0, %ymm0 ; CHECK-NEXT: retq %1 = insertelement <4 x i64> %a0, i64 %s0, i64 1 %2 = insertelement <4 x i64> %a1, i64 %s1, i64 1 %3 = insertelement <4 x i64> %1, i64 %s0, i64 3 %4 = insertelement <4 x i64> %2, i64 %s1, i64 3 %res = call <4 x i64> @llvm.x86.pclmulqdq.256(<4 x i64> %3, <4 x i64> %4, i8 17) ret <4 x i64> %res } define <8 x i64> @test_demanded_elts_pclmulqdq_512_0(<8 x i64> %a0, <8 x i64> %a1, i64 %s0, i64 %s1) { ; CHECK-LABEL: test_demanded_elts_pclmulqdq_512_0: ; CHECK: # %bb.0: ; CHECK-NEXT: vpclmulqdq $0, %zmm1, %zmm0, %zmm0 ; CHECK-NEXT: retq %1 = insertelement <8 x i64> %a0, i64 %s0, i64 1 %2 = insertelement <8 x i64> %a1, i64 %s1, i64 1 %3 = insertelement <8 x i64> %1, i64 %s0, i64 3 %4 = insertelement <8 x i64> %2, i64 %s1, i64 3 %5 = insertelement <8 x i64> %3, i64 %s0, i64 5 %6 = insertelement <8 x i64> %4, i64 %s1, i64 5 %7 = insertelement <8 x i64> %5, i64 %s0, i64 7 %8 = insertelement <8 x i64> %6, i64 %s1, i64 7 %res = call <8 x i64> @llvm.x86.pclmulqdq.512(<8 x i64> %7, <8 x i64> %8, i8 0) ret <8 x i64> %res } define <8 x i64> @test_demanded_elts_pclmulqdq_512_1(<8 x i64> %a0, <8 x i64> %a1, i64 %s0, i64 %s1) { ; CHECK-LABEL: test_demanded_elts_pclmulqdq_512_1: ; CHECK: # %bb.0: ; CHECK-NEXT: vpinsrq $1, %rdi, %xmm0, %xmm0 ; CHECK-NEXT: vpbroadcastq %rdi, %zmm2 ; CHECK-NEXT: vmovdqa64 {{.*#+}} zmm3 = [u,1,u,11,u,13,u,15] ; CHECK-NEXT: vpermi2q %zmm2, %zmm0, %zmm3 ; CHECK-NEXT: vpclmulqdq $1, %zmm1, %zmm3, %zmm0 ; CHECK-NEXT: retq %1 = insertelement <8 x i64> %a0, i64 %s0, i64 1 %2 = insertelement <8 x i64> %a1, i64 %s1, i64 1 %3 = insertelement <8 x i64> %1, i64 %s0, i64 3 %4 = insertelement <8 x i64> %2, i64 %s1, i64 3 %5 = insertelement <8 x i64> %3, i64 %s0, i64 5 %6 = insertelement <8 x i64> %4, i64 %s1, i64 5 %7 = insertelement <8 x i64> %5, i64 %s0, i64 7 %8 = insertelement <8 x i64> %6, i64 %s1, i64 7 %res = call <8 x i64> @llvm.x86.pclmulqdq.512(<8 x i64> %7, <8 x i64> %8, i8 1) ret <8 x i64> %res } define <8 x i64> @test_demanded_elts_pclmulqdq_512_16(<8 x i64> %a0, <8 x i64> %a1, i64 %s0, i64 %s1) { ; CHECK-LABEL: test_demanded_elts_pclmulqdq_512_16: ; CHECK: # %bb.0: ; CHECK-NEXT: vpinsrq $1, %rsi, %xmm0, %xmm1 ; CHECK-NEXT: vpbroadcastq %rsi, %zmm2 ; CHECK-NEXT: vmovdqa64 {{.*#+}} zmm3 = [u,1,u,11,u,13,u,15] ; CHECK-NEXT: vpermi2q %zmm2, %zmm1, %zmm3 ; CHECK-NEXT: vpclmulqdq $16, %zmm3, %zmm0, %zmm0 ; CHECK-NEXT: retq %1 = insertelement <8 x i64> %a0, i64 %s0, i64 1 %2 = insertelement <8 x i64> %a1, i64 %s1, i64 1 %3 = insertelement <8 x i64> %1, i64 %s0, i64 3 %4 = insertelement <8 x i64> %2, i64 %s1, i64 3 %5 = insertelement <8 x i64> %3, i64 %s0, i64 5 %6 = insertelement <8 x i64> %4, i64 %s1, i64 5 %7 = insertelement <8 x i64> %5, i64 %s0, i64 7 %8 = insertelement <8 x i64> %6, i64 %s1, i64 7 %res = call <8 x i64> @llvm.x86.pclmulqdq.512(<8 x i64> %7, <8 x i64> %8, i8 16) ret <8 x i64> %res } define <8 x i64> @test_demanded_elts_pclmulqdq_512_17(<8 x i64> %a0, <8 x i64> %a1, i64 %s0, i64 %s1) { ; CHECK-LABEL: test_demanded_elts_pclmulqdq_512_17: ; CHECK: # %bb.0: ; CHECK-NEXT: vpinsrq $1, %rdi, %xmm0, %xmm0 ; CHECK-NEXT: vpinsrq $1, %rsi, %xmm0, %xmm1 ; CHECK-NEXT: vpbroadcastq %rdi, %zmm2 ; CHECK-NEXT: vmovdqa64 {{.*#+}} zmm3 = [0,1,2,11,4,13,6,15] ; CHECK-NEXT: vpermt2q %zmm2, %zmm3, %zmm0 ; CHECK-NEXT: vpbroadcastq %rsi, %zmm2 ; CHECK-NEXT: vpermt2q %zmm2, %zmm3, %zmm1 ; CHECK-NEXT: vpclmulqdq $17, %zmm1, %zmm0, %zmm0 ; CHECK-NEXT: retq %1 = insertelement <8 x i64> %a0, i64 %s0, i64 1 %2 = insertelement <8 x i64> %a1, i64 %s1, i64 1 %3 = insertelement <8 x i64> %1, i64 %s0, i64 3 %4 = insertelement <8 x i64> %2, i64 %s1, i64 3 %5 = insertelement <8 x i64> %3, i64 %s0, i64 5 %6 = insertelement <8 x i64> %4, i64 %s1, i64 5 %7 = insertelement <8 x i64> %5, i64 %s0, i64 7 %8 = insertelement <8 x i64> %6, i64 %s1, i64 7 %res = call <8 x i64> @llvm.x86.pclmulqdq.512(<8 x i64> %7, <8 x i64> %8, i8 17) ret <8 x i64> %res } ; Ensure that target shuffle extraction can handle SM_SentinelZero. define <2 x i64> @pclmul_sentinel_zero(<2 x i64> %a0, <2 x i64> %a1) { ; CHECK-LABEL: pclmul_sentinel_zero: ; CHECK: # %bb.0: ; CHECK-NEXT: vpclmulqdq $0, %xmm1, %xmm0, %xmm0 ; CHECK-NEXT: retq %and = and <2 x i64> %a0, %res = tail call <2 x i64> @llvm.x86.pclmulqdq(<2 x i64> %and, <2 x i64> %a1, i8 0) ret <2 x i64> %res }