diff options
Diffstat (limited to 'llvm/test/CodeGen/X86/avx512bw-intrinsics-fast-isel.ll')
| -rw-r--r-- | llvm/test/CodeGen/X86/avx512bw-intrinsics-fast-isel.ll | 120 |
1 files changed, 120 insertions, 0 deletions
diff --git a/llvm/test/CodeGen/X86/avx512bw-intrinsics-fast-isel.ll b/llvm/test/CodeGen/X86/avx512bw-intrinsics-fast-isel.ll index c3b9d30cda1..50a9076163e 100644 --- a/llvm/test/CodeGen/X86/avx512bw-intrinsics-fast-isel.ll +++ b/llvm/test/CodeGen/X86/avx512bw-intrinsics-fast-isel.ll @@ -4,6 +4,126 @@ ; NOTE: This should use IR equivalent to what is generated by clang/test/CodeGen/avx512bw-builtins.c +define <8 x i64> @test_mm512_broadcastb_epi8(<2 x i64> %a0) { +; X32-LABEL: test_mm512_broadcastb_epi8: +; X32: # BB#0: +; X32-NEXT: vpbroadcastb %xmm0, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_broadcastb_epi8: +; X64: # BB#0: +; X64-NEXT: vpbroadcastb %xmm0, %zmm0 +; X64-NEXT: retq + %arg0 = bitcast <2 x i64> %a0 to <16 x i8> + %res0 = shufflevector <16 x i8> %arg0, <16 x i8> undef, <64 x i32> zeroinitializer + %res1 = bitcast <64 x i8> %res0 to <8 x i64> + ret <8 x i64> %res1 +} + +define <8 x i64> @test_mm512_mask_broadcastb_epi8(<8 x i64> %a0, i64* %a1, <2 x i64> %a2) { +; X32-LABEL: test_mm512_mask_broadcastb_epi8: +; X32: # BB#0: +; X32-NEXT: movl {{[0-9]+}}(%esp), %eax +; X32-NEXT: kmovq (%eax), %k1 +; X32-NEXT: vpbroadcastb %xmm1, %zmm0 {%k1} +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask_broadcastb_epi8: +; X64: # BB#0: +; X64-NEXT: kmovq (%rdi), %k1 +; X64-NEXT: vpbroadcastb %xmm1, %zmm0 {%k1} +; X64-NEXT: retq + %arg0 = bitcast <8 x i64> %a0 to <64 x i8> + %bc1 = bitcast i64* %a1 to <64 x i1>* + %arg1 = load <64 x i1>, <64 x i1>* %bc1 + %arg2 = bitcast <2 x i64> %a2 to <16 x i8> + %res0 = shufflevector <16 x i8> %arg2, <16 x i8> undef, <64 x i32> zeroinitializer + %res1 = select <64 x i1> %arg1, <64 x i8> %res0, <64 x i8> %arg0 + %res2 = bitcast <64 x i8> %res1 to <8 x i64> + ret <8 x i64> %res2 +} + +define <8 x i64> @test_mm512_maskz_broadcastb_epi8(i64* %a0, <2 x i64> %a1) { +; X32-LABEL: test_mm512_maskz_broadcastb_epi8: +; X32: # BB#0: +; X32-NEXT: movl {{[0-9]+}}(%esp), %eax +; X32-NEXT: kmovq (%eax), %k1 +; X32-NEXT: vpbroadcastb %xmm0, %zmm0 {%k1} {z} +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_maskz_broadcastb_epi8: +; X64: # BB#0: +; X64-NEXT: kmovq (%rdi), %k1 +; X64-NEXT: vpbroadcastb %xmm0, %zmm0 {%k1} {z} +; X64-NEXT: retq + %bc0 = bitcast i64* %a0 to <64 x i1>* + %arg0 = load <64 x i1>, <64 x i1>* %bc0 + %arg1 = bitcast <2 x i64> %a1 to <16 x i8> + %res0 = shufflevector <16 x i8> %arg1, <16 x i8> undef, <64 x i32> zeroinitializer + %res1 = select <64 x i1> %arg0, <64 x i8> %res0, <64 x i8> zeroinitializer + %res2 = bitcast <64 x i8> %res1 to <8 x i64> + ret <8 x i64> %res2 +} + +define <8 x i64> @test_mm512_broadcastw_epi16(<2 x i64> %a0) { +; X32-LABEL: test_mm512_broadcastw_epi16: +; X32: # BB#0: +; X32-NEXT: vpbroadcastw %xmm0, %zmm0 +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_broadcastw_epi16: +; X64: # BB#0: +; X64-NEXT: vpbroadcastw %xmm0, %zmm0 +; X64-NEXT: retq + %arg0 = bitcast <2 x i64> %a0 to <8 x i16> + %res0 = shufflevector <8 x i16> %arg0, <8 x i16> undef, <32 x i32> zeroinitializer + %res1 = bitcast <32 x i16> %res0 to <8 x i64> + ret <8 x i64> %res1 +} + +define <8 x i64> @test_mm512_mask_broadcastw_epi16(<8 x i64> %a0, i32 %a1, <2 x i64> %a2) { +; X32-LABEL: test_mm512_mask_broadcastw_epi16: +; X32: # BB#0: +; X32-NEXT: movl {{[0-9]+}}(%esp), %eax +; X32-NEXT: kmovd %eax, %k1 +; X32-NEXT: vpbroadcastw %xmm1, %zmm0 {%k1} +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_mask_broadcastw_epi16: +; X64: # BB#0: +; X64-NEXT: kmovd %edi, %k1 +; X64-NEXT: vpbroadcastw %xmm1, %zmm0 {%k1} +; X64-NEXT: retq + %arg0 = bitcast <8 x i64> %a0 to <32 x i16> + %arg1 = bitcast i32 %a1 to <32 x i1> + %arg2 = bitcast <2 x i64> %a2 to <8 x i16> + %res0 = shufflevector <8 x i16> %arg2, <8 x i16> undef, <32 x i32> zeroinitializer + %res1 = select <32 x i1> %arg1, <32 x i16> %res0, <32 x i16> %arg0 + %res2 = bitcast <32 x i16> %res1 to <8 x i64> + ret <8 x i64> %res2 +} + +define <8 x i64> @test_mm512_maskz_broadcastw_epi16(i32 %a0, <2 x i64> %a1) { +; X32-LABEL: test_mm512_maskz_broadcastw_epi16: +; X32: # BB#0: +; X32-NEXT: movl {{[0-9]+}}(%esp), %eax +; X32-NEXT: kmovd %eax, %k1 +; X32-NEXT: vpbroadcastw %xmm0, %zmm0 {%k1} {z} +; X32-NEXT: retl +; +; X64-LABEL: test_mm512_maskz_broadcastw_epi16: +; X64: # BB#0: +; X64-NEXT: kmovd %edi, %k1 +; X64-NEXT: vpbroadcastw %xmm0, %zmm0 {%k1} {z} +; X64-NEXT: retq + %arg0 = bitcast i32 %a0 to <32 x i1> + %arg1 = bitcast <2 x i64> %a1 to <8 x i16> + %res0 = shufflevector <8 x i16> %arg1, <8 x i16> undef, <32 x i32> zeroinitializer + %res1 = select <32 x i1> %arg0, <32 x i16> %res0, <32 x i16> zeroinitializer + %res2 = bitcast <32 x i16> %res1 to <8 x i64> + ret <8 x i64> %res2 +} + define <8 x i64> @test_mm512_bslli_epi128(<8 x i64> %a0) { ; X32-LABEL: test_mm512_bslli_epi128: ; X32: # BB#0: |
