From: Simon Pilgrim <llvm-dev@redking.me.uk>
Date: Sun, 7 Nov 2021 12:59:35 +0000 (+0000)
Subject: [X86][AVX] Add missing X86ISD::VBROADCAST(v4f32 -> v8f32) isel pattern for AVX1 targets
X-Git-Tag: upstream/15.0.7~26528
X-Git-Url: http://review.tizen.org/git/?a=commitdiff_plain;h=b5ef56f0bc9b51bd026a643d4a541af9993aaf89;p=platform%2Fupstream%2Fllvm.git

[X86][AVX] Add missing X86ISD::VBROADCAST(v4f32 -> v8f32) isel pattern for AVX1 targets

D109434 addressed the v2f64 -> v4f64 case, an internal test has found an equivalent crash for the v4f32 -> v8f32 case.
---

diff --git a/llvm/lib/Target/X86/X86InstrSSE.td b/llvm/lib/Target/X86/X86InstrSSE.td
index b918b57..035f139 100644
--- a/llvm/lib/Target/X86/X86InstrSSE.td
+++ b/llvm/lib/Target/X86/X86InstrSSE.td
@@ -7630,6 +7630,10 @@ let Predicates = [HasAVX1Only] in {
             (VINSERTF128rr (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)),
               (v4f32 (VPERMILPSri (v4f32 (COPY_TO_REGCLASS FR32:$src, VR128)), 0)), sub_xmm),
               (v4f32 (VPERMILPSri (v4f32 (COPY_TO_REGCLASS FR32:$src, VR128)), 0)), 1)>;
+  def : Pat<(v8f32 (X86VBroadcast v4f32:$src)),
+            (VINSERTF128rr (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)),
+              (v4f32 (VPERMILPSri VR128:$src, 0)), sub_xmm),
+              (v4f32 (VPERMILPSri VR128:$src, 0)), 1)>;
   def : Pat<(v4f64 (X86VBroadcast FR64:$src)),
             (VINSERTF128rr (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)),
               (v2f64 (VMOVDDUPrr (v2f64 (COPY_TO_REGCLASS FR64:$src, VR128)))), sub_xmm),
diff --git a/llvm/test/CodeGen/X86/combine-concatvectors.ll b/llvm/test/CodeGen/X86/combine-concatvectors.ll
index 3dffeae..c2f8827 100644
--- a/llvm/test/CodeGen/X86/combine-concatvectors.ll
+++ b/llvm/test/CodeGen/X86/combine-concatvectors.ll
@@ -85,3 +85,38 @@ ifmerge.1298:                                     ; preds = %loop.4942
   store <2 x float> <float 1.000000e+00, float 1.000000e+00>, <2 x float>* bitcast (i8* getelementptr inbounds ([49216 x i8], [49216 x i8]* @qa_, i64 0, i64 47372) to <2 x float>*), align 4
   ret void
 }
+
+define <4 x float> @concat_of_broadcast_v4f32_v8f32(<8 x float>* %a0, <8 x float>* %a1, <8 x float>* %a2) {
+; AVX1-LABEL: concat_of_broadcast_v4f32_v8f32:
+; AVX1:       # %bb.0:
+; AVX1-NEXT:    vmovaps (%rdi), %ymm0
+; AVX1-NEXT:    vmovaps (%rsi), %ymm1
+; AVX1-NEXT:    vmovaps (%rdx), %ymm2
+; AVX1-NEXT:    vpermilps {{.*#+}} xmm1 = xmm1[0,0,0,0]
+; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm1, %ymm1
+; AVX1-NEXT:    vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
+; AVX1-NEXT:    vblendps {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4],ymm2[5,6],ymm0[7]
+; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
+; AVX1-NEXT:    vshufps {{.*#+}} xmm1 = xmm2[3,0],xmm0[0,0]
+; AVX1-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[2,3],xmm1[2,0]
+; AVX1-NEXT:    vzeroupper
+; AVX1-NEXT:    retq
+;
+; AVX2-LABEL: concat_of_broadcast_v4f32_v8f32:
+; AVX2:       # %bb.0:
+; AVX2-NEXT:    vmovaps (%rdi), %ymm0
+; AVX2-NEXT:    vunpcklps {{.*#+}} ymm0 = ymm0[0],mem[0],ymm0[1],mem[1],ymm0[4],mem[4],ymm0[5],mem[5]
+; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,1,2,0]
+; AVX2-NEXT:    vmovaps {{.*#+}} xmm1 = [6,7,4,3]
+; AVX2-NEXT:    vblendps {{.*#+}} ymm0 = mem[0,1,2,3],ymm0[4],mem[5,6],ymm0[7]
+; AVX2-NEXT:    vpermps %ymm0, %ymm1, %ymm0
+; AVX2-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
+; AVX2-NEXT:    vzeroupper
+; AVX2-NEXT:    retq
+  %ld0 = load volatile <8 x float>, <8 x float>* %a0
+  %ld1 = load volatile <8 x float>, <8 x float>* %a1
+  %ld2 = load volatile <8 x float>, <8 x float>* %a2
+  %shuffle = shufflevector <8 x float> %ld0, <8 x float> %ld1, <8 x i32> <i32 undef, i32 undef, i32 undef, i32 undef, i32 4, i32 undef, i32 undef, i32 8>
+  %shuffle1 = shufflevector <8 x float> %ld2, <8 x float> %shuffle, <4 x i32> <i32 6, i32 15, i32 12, i32 3>
+  ret <4 x float> %shuffle1
+}