[X86] Add tests showing failure to concat sqrt intrinsics together. (llvm#170096)

Similar to fdiv, we should be trying to concat these high latency instructions together

Author: RKSimon

llvm/test/CodeGen/X86/combine-fsqrt.ll

@@ -0,0 +1,91 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64    | FileCheck %s --check-prefixes=SSE
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64-v2 | FileCheck %s --check-prefixes=SSE
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=sandybridge | FileCheck %s --check-prefixes=AVX,AVX1OR2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64-v3 | FileCheck %s --check-prefixes=AVX,AVX1OR2
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64-v4 | FileCheck %s --check-prefixes=AVX,AVX512
+
+define <8 x float> @concat_sqrt_v8f32_v4f32(<4 x float> %a0, <4 x float> %a1) {
+; SSE-LABEL: concat_sqrt_v8f32_v4f32:
+; SSE:       # %bb.0:
+; SSE-NEXT:    sqrtps %xmm0, %xmm0
+; SSE-NEXT:    sqrtps %xmm1, %xmm1
+; SSE-NEXT:    retq
+;
+; AVX-LABEL: concat_sqrt_v8f32_v4f32:
+; AVX:       # %bb.0:
+; AVX-NEXT:    vsqrtps %xmm0, %xmm0
+; AVX-NEXT:    vsqrtps %xmm1, %xmm1
+; AVX-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX-NEXT:    retq
+  %v0 = call <4 x float> @llvm.sqrt.v4f32(<4 x float> %a0)
+  %v1 = call <4 x float> @llvm.sqrt.v4f32(<4 x float> %a1)
+  %res  = shufflevector <4 x float> %v0, <4 x float> %v1, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+  ret <8 x float> %res
+}
+
+define <16 x float> @concat_sqrt_v16f32_v4f32(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, <4 x float> %a3) {
+; SSE-LABEL: concat_sqrt_v16f32_v4f32:
+; SSE:       # %bb.0:
+; SSE-NEXT:    sqrtps %xmm0, %xmm0
+; SSE-NEXT:    sqrtps %xmm1, %xmm1
+; SSE-NEXT:    sqrtps %xmm2, %xmm2
+; SSE-NEXT:    sqrtps %xmm3, %xmm3
+; SSE-NEXT:    retq
+;
+; AVX1OR2-LABEL: concat_sqrt_v16f32_v4f32:
+; AVX1OR2:       # %bb.0:
+; AVX1OR2-NEXT:    vsqrtps %xmm0, %xmm0
+; AVX1OR2-NEXT:    vsqrtps %xmm1, %xmm1
+; AVX1OR2-NEXT:    vsqrtps %xmm2, %xmm2
+; AVX1OR2-NEXT:    vsqrtps %xmm3, %xmm3
+; AVX1OR2-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX1OR2-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm1
+; AVX1OR2-NEXT:    retq
+;
+; AVX512-LABEL: concat_sqrt_v16f32_v4f32:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vsqrtps %xmm0, %xmm0
+; AVX512-NEXT:    vsqrtps %xmm1, %xmm1
+; AVX512-NEXT:    vsqrtps %xmm2, %xmm2
+; AVX512-NEXT:    vsqrtps %xmm3, %xmm3
+; AVX512-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm2
+; AVX512-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX512-NEXT:    vinsertf64x4 $1, %ymm2, %zmm0, %zmm0
+; AVX512-NEXT:    retq
+  %v0 = call <4 x float> @llvm.sqrt.v4f32(<4 x float> %a0)
+  %v1 = call <4 x float> @llvm.sqrt.v4f32(<4 x float> %a1)
+  %v2 = call <4 x float> @llvm.sqrt.v4f32(<4 x float> %a2)
+  %v3 = call <4 x float> @llvm.sqrt.v4f32(<4 x float> %a3)
+  %r01 = shufflevector <4 x float> %v0, <4 x float> %v1, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+  %r23 = shufflevector <4 x float> %v2, <4 x float> %v3, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+  %res  = shufflevector <8 x float> %r01, <8 x float> %r23, <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>
+  ret <16 x float> %res
+}
+
+define <16 x float> @concat_sqrt_v16f32_v8f32(<8 x float> %a0, <8 x float> %a1) {
+; SSE-LABEL: concat_sqrt_v16f32_v8f32:
+; SSE:       # %bb.0:
+; SSE-NEXT:    sqrtps %xmm0, %xmm0
+; SSE-NEXT:    sqrtps %xmm1, %xmm1
+; SSE-NEXT:    sqrtps %xmm2, %xmm2
+; SSE-NEXT:    sqrtps %xmm3, %xmm3
+; SSE-NEXT:    retq
+;
+; AVX1OR2-LABEL: concat_sqrt_v16f32_v8f32:
+; AVX1OR2:       # %bb.0:
+; AVX1OR2-NEXT:    vsqrtps %ymm0, %ymm0
+; AVX1OR2-NEXT:    vsqrtps %ymm1, %ymm1
+; AVX1OR2-NEXT:    retq
+;
+; AVX512-LABEL: concat_sqrt_v16f32_v8f32:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vsqrtps %ymm0, %ymm0
+; AVX512-NEXT:    vsqrtps %ymm1, %ymm1
+; AVX512-NEXT:    vinsertf64x4 $1, %ymm1, %zmm0, %zmm0
+; AVX512-NEXT:    retq
+  %v0 = call <8 x float> @llvm.sqrt.v8f32(<8 x float> %a0)
+  %v1 = call <8 x float> @llvm.sqrt.v8f32(<8 x float> %a1)
+  %res  = shufflevector <8 x float> %v0, <8 x float> %v1, <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>
+  ret <16 x float> %res
+}