Merging r370592:

zmodem · zmodem · commit 5cbaa56ac5ff · 2019-09-09T09:48:38.000Z
------------------------------------------------------------------------
r370592 | rksimon | 2019-08-31 18:21:31 +0200 (Sat, 31 Aug 2019) | 3 lines

[X86] EltsFromConsecutiveLoads - Don't confuse elt count with vector element count (PR43170)

EltsFromConsecutiveLoads was assuming that the number of input elts was the same as the number of elements in the output vector type when creating a zeroing shuffle, causing an assert when subvectors were being combined instead of just scalars.
------------------------------------------------------------------------

llvm-svn: 371382
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -7650,17 +7650,22 @@ static SDValue EltsFromConsecutiveLoads(EVT VT, ArrayRef<SDValue> Elts,
     // IsConsecutiveLoadWithZeros - we need to create a shuffle of the loaded
     // vector and a zero vector to clear out the zero elements.
     if (!isAfterLegalize && VT.isVector()) {
-      SmallVector<int, 4> ClearMask(NumElems, -1);
-      for (unsigned i = 0; i < NumElems; ++i) {
-        if (ZeroMask[i])
-          ClearMask[i] = i + NumElems;
-        else if (LoadMask[i])
-          ClearMask[i] = i;
-      }
-      SDValue V = CreateLoad(VT, LDBase);
-      SDValue Z = VT.isInteger() ? DAG.getConstant(0, DL, VT)
-                                 : DAG.getConstantFP(0.0, DL, VT);
-      return DAG.getVectorShuffle(VT, DL, V, Z, ClearMask);
+      unsigned NumMaskElts = VT.getVectorNumElements();
+      if ((NumMaskElts % NumElems) == 0) {
+        unsigned Scale = NumMaskElts / NumElems;
+        SmallVector<int, 4> ClearMask(NumMaskElts, -1);
+        for (unsigned i = 0; i < NumElems; ++i) {
+          if (UndefMask[i])
+            continue;
+          int Offset = ZeroMask[i] ? NumMaskElts : 0;
+          for (unsigned j = 0; j != Scale; ++j)
+            ClearMask[(i * Scale) + j] = (i * Scale) + j + Offset;
+        }
+        SDValue V = CreateLoad(VT, LDBase);
+        SDValue Z = VT.isInteger() ? DAG.getConstant(0, DL, VT)
+                                   : DAG.getConstantFP(0.0, DL, VT);
+        return DAG.getVectorShuffle(VT, DL, V, Z, ClearMask);
+      }
     }
   }
 
diff --git a/llvm/test/CodeGen/X86/vector-shuffle-avx512.ll b/llvm/test/CodeGen/X86/vector-shuffle-avx512.ll
@@ -936,3 +936,41 @@ define <16 x float> @test_masked_permps_v16f32(<16 x float>* %vp, <16 x float> %
   %res = select <16 x i1> <i1 1, i1 1, i1 1, i1 0, i1 1, i1 1, i1 0, i1 0, i1 1, i1 1, i1 1, i1 0, i1 1, i1 0, i1 1, i1 0>, <16 x float> %shuf, <16 x float> %vec2
   ret <16 x float> %res
 }
+
+%union1= type { <16 x float> }
+@src1 = external dso_local local_unnamed_addr global %union1, align 64
+
+define void @PR43170(<16 x float>* %a0) {
+; SKX64-LABEL: PR43170:
+; SKX64:       # %bb.0: # %entry
+; SKX64-NEXT:    vmovaps {{.*}}(%rip), %ymm0
+; SKX64-NEXT:    vmovaps %zmm0, (%rdi)
+; SKX64-NEXT:    vzeroupper
+; SKX64-NEXT:    retq
+;
+; KNL64-LABEL: PR43170:
+; KNL64:       # %bb.0: # %entry
+; KNL64-NEXT:    vmovaps {{.*}}(%rip), %ymm0
+; KNL64-NEXT:    vmovaps %zmm0, (%rdi)
+; KNL64-NEXT:    retq
+;
+; SKX32-LABEL: PR43170:
+; SKX32:       # %bb.0: # %entry
+; SKX32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; SKX32-NEXT:    vmovaps src1, %ymm0
+; SKX32-NEXT:    vmovaps %zmm0, (%eax)
+; SKX32-NEXT:    vzeroupper
+; SKX32-NEXT:    retl
+;
+; KNL32-LABEL: PR43170:
+; KNL32:       # %bb.0: # %entry
+; KNL32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; KNL32-NEXT:    vmovaps src1, %ymm0
+; KNL32-NEXT:    vmovaps %zmm0, (%eax)
+; KNL32-NEXT:    retl
+entry:
+  %0 = load <8 x float>, <8 x float>* bitcast (%union1* @src1 to <8 x float>*), align 64
+  %1 = shufflevector <8 x float> %0, <8 x float> zeroinitializer, <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>
+  store <16 x float> %1, <16 x float>* %a0, align 64
+  ret void
+}
