[ConstantFold] Improve the bitcast folding logic for constant vectors.

The constant folder didn't know how to always fold bitcasts of constant integer
vectors. In particular, it was unable to handle the case where a constant vector
had some undef elements, and the resulting (i.e. bitcasted) vector type had more
elements than the original vector type.

Example:
  %cast = bitcast <2 x i64><i64 undef, i64 2> to <4 x i32>

On a little endian target, %cast could have been folded to:
  <4 x i32><i32 undef, i32 undef, i32 2, i32 0>

This patch improves the folding logic by teaching how to correctly propagate
undef elements in the folded vector.

Differential Revision: https://reviews.llvm.org/D24301

llvm-svn: 281343

3 files changed, 35 insertions, 25 deletions

diff --git a/llvm/lib/Analysis/ConstantFolding.cpp b/llvm/lib/Analysis/ConstantFolding.cpp
index 73cf336f5eb..1df724a044b 100644
--- a/llvm/lib/Analysis/ConstantFolding.cpp
+++ b/llvm/lib/Analysis/ConstantFolding.cpp
@@ -224,8 +224,19 @@ Constant *FoldBitCast(Constant *C, Type *DestTy, const DataLayout &DL) {
 
   // Loop over each source value, expanding into multiple results.
   for (unsigned i = 0; i != NumSrcElt; ++i) {
-    auto *Src = dyn_cast_or_null<ConstantInt>(C->getAggregateElement(i));
-    if (!Src)  // Reject constantexpr elements.
+    auto *Element = C->getAggregateElement(i);
+
+    if (!Element) // Reject constantexpr elements.
+      return ConstantExpr::getBitCast(C, DestTy);
+
+    if (isa<UndefValue>(Element)) {
+      // Correctly Propagate undef values.
+      Result.append(Ratio, UndefValue::get(DstEltTy));
+      continue;
+    }
+
+    auto *Src = dyn_cast<ConstantInt>(Element);
+    if (!Src)
       return ConstantExpr::getBitCast(C, DestTy);
 
     unsigned ShiftAmt = isLittleEndian ? 0 : DstBitSize*(Ratio-1);
diff --git a/llvm/test/Transforms/InstCombine/x86-sse4a.ll b/llvm/test/Transforms/InstCombine/x86-sse4a.ll
index d2714e0c630..e36a7353225 100644
--- a/llvm/test/Transforms/InstCombine/x86-sse4a.ll
+++ b/llvm/test/Transforms/InstCombine/x86-sse4a.ll
@@ -57,8 +57,7 @@ define <2 x i64> @test_extrq_constant_undef(<2 x i64> %x, <16 x i8> %y) {
 
 define <2 x i64> @test_extrq_call_constexpr(<2 x i64> %x) {
 ; CHECK-LABEL: @test_extrq_call_constexpr(
-; CHECK-NEXT:    [[TMP1:%.*]] = call <2 x i64> @llvm.x86.sse4a.extrq(<2 x i64> %x, <16 x i8> bitcast (<2 x i64> <i64 0, i64 undef> to <16 x i8>))
-; CHECK-NEXT:    ret <2 x i64> [[TMP1]]
+; CHECK-NEXT:    ret <2 x i64> %x
 ;
   %1 = call <2 x i64> @llvm.x86.sse4a.extrq(<2 x i64> %x, <16 x i8> bitcast (<2 x i64> <i64 0, i64 undef> to <16 x i8>))
   ret <2 x i64> %1
@@ -133,7 +132,7 @@ define <2 x i64> @test_extrqi_constant_undef(<2 x i64> %x) {
 
 define <2 x i64> @test_extrqi_call_constexpr() {
 ; CHECK-LABEL: @test_extrqi_call_constexpr(
-; CHECK-NEXT:    ret <2 x i64> bitcast (<16 x i8> <i8 extractelement (<16 x i8> trunc (<16 x i16> bitcast (<4 x i64> <i64 0, i64 undef, i64 2, i64 undef> to <16 x i16>) to <16 x i8>), i32 2), i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef> to <2 x i64>)
+; CHECK-NEXT:    ret <2 x i64> zeroinitializer
 ;
   %1 = tail call <2 x i64> @llvm.x86.sse4a.extrqi(<2 x i64> bitcast (<16 x i8> trunc (<16 x i16> bitcast (<4 x i64> <i64 0, i64 undef, i64 2, i64 undef> to <16 x i16>) to <16 x i8>) to <2 x i64>), i8 8, i8 16)
   ret <2 x i64> %1
@@ -179,7 +178,7 @@ define <2 x i64> @test_insertq_constant_undef(<2 x i64> %x, <2 x i64> %y) {
 
 define <2 x i64> @test_insertq_call_constexpr(<2 x i64> %x) {
 ; CHECK-LABEL: @test_insertq_call_constexpr(
-; CHECK-NEXT:    [[TMP1:%.*]] = tail call <2 x i64> @llvm.x86.sse4a.insertq(<2 x i64> %x, <2 x i64> bitcast (<16 x i8> trunc (<16 x i16> bitcast (<4 x i64> <i64 0, i64 undef, i64 2, i64 undef> to <16 x i16>) to <16 x i8>) to <2 x i64>))
+; CHECK-NEXT:    [[TMP1:%.*]] = call <2 x i64> @llvm.x86.sse4a.insertqi(<2 x i64> %x, <2 x i64> <i64 0, i64 undef>, i8 2, i8 0)
 ; CHECK-NEXT:    ret <2 x i64> [[TMP1]]
 ;
   %1 = tail call <2 x i64> @llvm.x86.sse4a.insertq(<2 x i64> %x, <2 x i64> bitcast (<16 x i8> trunc (<16 x i16> bitcast (<4 x i64> <i64 0, i64 undef, i64 2, i64 undef> to <16 x i16>) to <16 x i8>) to <2 x i64>))
@@ -224,7 +223,7 @@ define <2 x i64> @test_insertqi_constant(<2 x i64> %v, <2 x i64> %i) {
 
 define <2 x i64> @test_insertqi_call_constexpr(<2 x i64> %x) {
 ; CHECK-LABEL: @test_insertqi_call_constexpr(
-; CHECK-NEXT:    [[TMP1:%.*]] = tail call <2 x i64> @llvm.x86.sse4a.insertqi(<2 x i64> %x, <2 x i64> bitcast (<16 x i8> trunc (<16 x i16> bitcast (<4 x i64> <i64 0, i64 undef, i64 2, i64 undef> to <16 x i16>) to <16 x i8>) to <2 x i64>), i8 48, i8 3)
+; CHECK-NEXT:    [[TMP1:%.*]] = tail call <2 x i64> @llvm.x86.sse4a.insertqi(<2 x i64> %x, <2 x i64> <i64 0, i64 undef>, i8 48, i8 3)
 ; CHECK-NEXT:    ret <2 x i64> [[TMP1]]
 ;
   %1 = tail call <2 x i64> @llvm.x86.sse4a.insertqi(<2 x i64> %x, <2 x i64> bitcast (<16 x i8> trunc (<16 x i16> bitcast (<4 x i64> <i64 0, i64 undef, i64 2, i64 undef> to <16 x i16>) to <16 x i8>) to <2 x i64>), i8 48, i8 3)
diff --git a/llvm/test/Transforms/InstSimplify/bitcast-vector-fold.ll b/llvm/test/Transforms/InstSimplify/bitcast-vector-fold.ll
index 126aba63ecb..c8c390ea6be 100644
--- a/llvm/test/Transforms/InstSimplify/bitcast-vector-fold.ll
+++ b/llvm/test/Transforms/InstSimplify/bitcast-vector-fold.ll
@@ -126,7 +126,7 @@ define <2 x double> @foo6() {
 
 define <4 x i32> @bitcast_constexpr_4i32_2i64_u2() {
 ; CHECK-LABEL: @bitcast_constexpr_4i32_2i64_u2(
-; CHECK-NEXT:    ret <4 x i32> bitcast (<2 x i64> <i64 undef, i64 2> to <4 x i32>)
+; CHECK-NEXT:    ret <4 x i32> <i32 undef, i32 undef, i32 2, i32 0>
 ;
   %cast = bitcast <2 x i64><i64 undef, i64 2> to <4 x i32>
   ret <4 x i32> %cast
@@ -134,7 +134,7 @@ define <4 x i32> @bitcast_constexpr_4i32_2i64_u2() {
 
 define <4 x i32> @bitcast_constexpr_4i32_2i64_1u() {
 ; CHECK-LABEL: @bitcast_constexpr_4i32_2i64_1u(
-; CHECK-NEXT:    ret <4 x i32> bitcast (<2 x i64> <i64 1, i64 undef> to <4 x i32>)
+; CHECK-NEXT:    ret <4 x i32> <i32 1, i32 0, i32 undef, i32 undef>
 ;
   %cast = bitcast <2 x i64><i64 1, i64 undef> to <4 x i32>
   ret <4 x i32> %cast
@@ -142,7 +142,7 @@ define <4 x i32> @bitcast_constexpr_4i32_2i64_1u() {
 
 define <4 x i32> @bitcast_constexpr_4i32_2i64() {
 ; CHECK-LABEL: @bitcast_constexpr_4i32_2i64(
-; CHECK-NEXT:    ret <4 x i32> bitcast (<2 x i64> <i64 undef, i64 2> to <4 x i32>)
+; CHECK-NEXT:    ret <4 x i32> <i32 undef, i32 undef, i32 2, i32 0>
 ;
   %cast = bitcast <2 x i64><i64 undef, i64 2> to <4 x i32>
   ret <4 x i32> %cast
@@ -150,7 +150,7 @@ define <4 x i32> @bitcast_constexpr_4i32_2i64() {
 
 define <8 x i16> @bitcast_constexpr_8i16_2i64_u2() {
 ; CHECK-LABEL: @bitcast_constexpr_8i16_2i64_u2(
-; CHECK-NEXT:    ret <8 x i16> bitcast (<2 x i64> <i64 undef, i64 2> to <8 x i16>)
+; CHECK-NEXT:    ret <8 x i16> <i16 undef, i16 undef, i16 undef, i16 undef, i16 2, i16 0, i16 0, i16 0>
 ;
   %cast = bitcast <2 x i64><i64 undef, i64 2> to <8 x i16>
   ret <8 x i16> %cast
@@ -158,7 +158,7 @@ define <8 x i16> @bitcast_constexpr_8i16_2i64_u2() {
 
 define <8 x i16> @bitcast_constexpr_8i16_2i64_1u() {
 ; CHECK-LABEL: @bitcast_constexpr_8i16_2i64_1u(
-; CHECK-NEXT:    ret <8 x i16> bitcast (<2 x i64> <i64 1, i64 undef> to <8 x i16>)
+; CHECK-NEXT:    ret <8 x i16> <i16 1, i16 0, i16 0, i16 0, i16 undef, i16 undef, i16 undef, i16 undef>
 ;
   %cast = bitcast <2 x i64><i64 1, i64 undef> to <8 x i16>
   ret <8 x i16> %cast
@@ -166,7 +166,7 @@ define <8 x i16> @bitcast_constexpr_8i16_2i64_1u() {
 
 define <8 x i16> @bitcast_constexpr_8i16_2i64_u65536() {
 ; CHECK-LABEL: @bitcast_constexpr_8i16_2i64_u65536(
-; CHECK-NEXT:    ret <8 x i16> bitcast (<2 x i64> <i64 undef, i64 65536> to <8 x i16>)
+; CHECK-NEXT:    ret <8 x i16> <i16 undef, i16 undef, i16 undef, i16 undef, i16 0, i16 1, i16 0, i16 0>
 ;
   %cast = bitcast <2 x i64><i64 undef, i64 65536> to <8 x i16>
   ret <8 x i16> %cast
@@ -174,7 +174,7 @@ define <8 x i16> @bitcast_constexpr_8i16_2i64_u65536() {
 
 define <16 x i8> @bitcast_constexpr_16i8_2i64_u2() {
 ; CHECK-LABEL: @bitcast_constexpr_16i8_2i64_u2(
-; CHECK-NEXT:    ret <16 x i8> bitcast (<2 x i64> <i64 undef, i64 2> to <16 x i8>)
+; CHECK-NEXT:    ret <16 x i8> <i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 2, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0>
 ;
   %cast = bitcast <2 x i64><i64 undef, i64 2> to <16 x i8>
   ret <16 x i8> %cast
@@ -182,7 +182,7 @@ define <16 x i8> @bitcast_constexpr_16i8_2i64_u2() {
 
 define <16 x i8> @bitcast_constexpr_16i8_2i64_256u() {
 ; CHECK-LABEL: @bitcast_constexpr_16i8_2i64_256u(
-; CHECK-NEXT:    ret <16 x i8> bitcast (<2 x i64> <i64 256, i64 undef> to <16 x i8>)
+; CHECK-NEXT:    ret <16 x i8> <i8 0, i8 1, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef>
 ;
   %cast = bitcast <2 x i64><i64 256, i64 undef> to <16 x i8>
   ret <16 x i8> %cast
@@ -190,7 +190,7 @@ define <16 x i8> @bitcast_constexpr_16i8_2i64_256u() {
 
 define <16 x i8> @bitcast_constexpr_16i8_2i64_u256() {
 ; CHECK-LABEL: @bitcast_constexpr_16i8_2i64_u256(
-; CHECK-NEXT:    ret <16 x i8> bitcast (<2 x i64> <i64 undef, i64 256> to <16 x i8>)
+; CHECK-NEXT:    ret <16 x i8> <i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 0, i8 1, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0>
 ;
   %cast = bitcast <2 x i64><i64 undef, i64 256> to <16 x i8>
   ret <16 x i8> %cast
@@ -198,7 +198,7 @@ define <16 x i8> @bitcast_constexpr_16i8_2i64_u256() {
 
 define <8 x i16> @bitcast_constexpr_8i16_4i32_uu22() {
 ; CHECK-LABEL: @bitcast_constexpr_8i16_4i32_uu22(
-; CHECK-NEXT:    ret <8 x i16> bitcast (<4 x i32> <i32 undef, i32 undef, i32 2, i32 2> to <8 x i16>)
+; CHECK-NEXT:    ret <8 x i16> <i16 undef, i16 undef, i16 undef, i16 undef, i16 2, i16 0, i16 2, i16 0>
 ;
   %cast = bitcast <4 x i32><i32 undef, i32 undef, i32 2, i32 2> to <8 x i16>
   ret <8 x i16> %cast
@@ -206,7 +206,7 @@ define <8 x i16> @bitcast_constexpr_8i16_4i32_uu22() {
 
 define <8 x i16> @bitcast_constexpr_8i16_4i32_10uu() {
 ; CHECK-LABEL: @bitcast_constexpr_8i16_4i32_10uu(
-; CHECK-NEXT:    ret <8 x i16> bitcast (<4 x i32> <i32 1, i32 0, i32 undef, i32 undef> to <8 x i16>)
+; CHECK-NEXT:    ret <8 x i16> <i16 1, i16 0, i16 0, i16 0, i16 undef, i16 undef, i16 undef, i16 undef>
 ;
   %cast = bitcast <4 x i32><i32 1, i32 0, i32 undef, i32 undef> to <8 x i16>
   ret <8 x i16> %cast
@@ -214,7 +214,7 @@ define <8 x i16> @bitcast_constexpr_8i16_4i32_10uu() {
 
 define <8 x i16> @bitcast_constexpr_8i16_4i32_u257u256() {
 ; CHECK-LABEL: @bitcast_constexpr_8i16_4i32_u257u256(
-; CHECK-NEXT:    ret <8 x i16> bitcast (<4 x i32> <i32 undef, i32 65536, i32 undef, i32 65536> to <8 x i16>)
+; CHECK-NEXT:    ret <8 x i16> <i16 undef, i16 undef, i16 0, i16 1, i16 undef, i16 undef, i16 0, i16 1>
 ;
   %cast = bitcast <4 x i32><i32 undef, i32 65536, i32 undef, i32 65536> to <8 x i16>
   ret <8 x i16> %cast
@@ -222,7 +222,7 @@ define <8 x i16> @bitcast_constexpr_8i16_4i32_u257u256() {
 
 define <16 x i8> @bitcast_constexpr_16i8_4i32_u2u2() {
 ; CHECK-LABEL: @bitcast_constexpr_16i8_4i32_u2u2(
-; CHECK-NEXT:    ret <16 x i8> bitcast (<4 x i32> <i32 undef, i32 2, i32 undef, i32 2> to <16 x i8>)
+; CHECK-NEXT:    ret <16 x i8> <i8 undef, i8 undef, i8 undef, i8 undef, i8 2, i8 0, i8 0, i8 0, i8 undef, i8 undef, i8 undef, i8 undef, i8 2, i8 0, i8 0, i8 0>
 ;
   %cast = bitcast <4 x i32><i32 undef, i32 2, i32 undef, i32 2> to <16 x i8>
   ret <16 x i8> %cast
@@ -230,7 +230,7 @@ define <16 x i8> @bitcast_constexpr_16i8_4i32_u2u2() {
 
 define <16 x i8> @bitcast_constexpr_16i8_4i32_1u1u() {
 ; CHECK-LABEL: @bitcast_constexpr_16i8_4i32_1u1u(
-; CHECK-NEXT:    ret <16 x i8> bitcast (<4 x i32> <i32 1, i32 undef, i32 1, i32 undef> to <16 x i8>)
+; CHECK-NEXT:    ret <16 x i8> <i8 1, i8 0, i8 0, i8 0, i8 undef, i8 undef, i8 undef, i8 undef, i8 1, i8 0, i8 0, i8 0, i8 undef, i8 undef, i8 undef, i8 undef>
 ;
   %cast = bitcast <4 x i32><i32 1, i32 undef, i32 1, i32 undef> to <16 x i8>
   ret <16 x i8> %cast
@@ -238,7 +238,7 @@ define <16 x i8> @bitcast_constexpr_16i8_4i32_1u1u() {
 
 define <16 x i8> @bitcast_constexpr_16i8_4i32_u256uu() {
 ; CHECK-LABEL: @bitcast_constexpr_16i8_4i32_u256uu(
-; CHECK-NEXT:    ret <16 x i8> bitcast (<4 x i32> <i32 undef, i32 256, i32 undef, i32 undef> to <16 x i8>)
+; CHECK-NEXT:    ret <16 x i8> <i8 undef, i8 undef, i8 undef, i8 undef, i8 0, i8 1, i8 0, i8 0, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef>
 ;
   %cast = bitcast <4 x i32><i32 undef, i32 256, i32 undef, i32 undef> to <16 x i8>
   ret <16 x i8> %cast
@@ -246,7 +246,7 @@ define <16 x i8> @bitcast_constexpr_16i8_4i32_u256uu() {
 
 define <16 x i8> @bitcast_constexpr_16i8_8i16_u2u2u2u2() {
 ; CHECK-LABEL: @bitcast_constexpr_16i8_8i16_u2u2u2u2(
-; CHECK-NEXT:    ret <16 x i8> bitcast (<8 x i16> <i16 undef, i16 2, i16 undef, i16 2, i16 undef, i16 2, i16 undef, i16 2> to <16 x i8>)
+; CHECK-NEXT:    ret <16 x i8> <i8 undef, i8 undef, i8 2, i8 0, i8 undef, i8 undef, i8 2, i8 0, i8 undef, i8 undef, i8 2, i8 0, i8 undef, i8 undef, i8 2, i8 0>
 ;
   %cast = bitcast <8 x i16><i16 undef, i16 2, i16 undef, i16 2, i16 undef, i16 2, i16 undef, i16 2> to <16 x i8>
   ret <16 x i8> %cast
@@ -254,7 +254,7 @@ define <16 x i8> @bitcast_constexpr_16i8_8i16_u2u2u2u2() {
 
 define <16 x i8> @bitcast_constexpr_16i8_8i16_1u1u1u1u() {
 ; CHECK-LABEL: @bitcast_constexpr_16i8_8i16_1u1u1u1u(
-; CHECK-NEXT:    ret <16 x i8> bitcast (<8 x i16> <i16 1, i16 undef, i16 1, i16 undef, i16 1, i16 undef, i16 1, i16 undef> to <16 x i8>)
+; CHECK-NEXT:    ret <16 x i8> <i8 1, i8 0, i8 undef, i8 undef, i8 1, i8 0, i8 undef, i8 undef, i8 1, i8 0, i8 undef, i8 undef, i8 1, i8 0, i8 undef, i8 undef>
 ;
   %cast = bitcast <8 x i16><i16 1, i16 undef, i16 1, i16 undef, i16 1, i16 undef, i16 1, i16 undef> to <16 x i8>
   ret <16 x i8> %cast
@@ -262,7 +262,7 @@ define <16 x i8> @bitcast_constexpr_16i8_8i16_1u1u1u1u() {
 
 define <16 x i8> @bitcast_constexpr_16i8_8i16_u256uuu256uu() {
 ; CHECK-LABEL: @bitcast_constexpr_16i8_8i16_u256uuu256uu(
-; CHECK-NEXT:    ret <16 x i8> bitcast (<8 x i16> <i16 undef, i16 256, i16 undef, i16 undef, i16 undef, i16 256, i16 undef, i16 undef> to <16 x i8>)
+; CHECK-NEXT:    ret <16 x i8> <i8 undef, i8 undef, i8 0, i8 1, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 undef, i8 0, i8 1, i8 undef, i8 undef, i8 undef, i8 undef>
 ;
   %cast = bitcast <8 x i16><i16 undef, i16 256, i16 undef, i16 undef, i16 undef, i16 256, i16 undef, i16 undef> to <16 x i8>
   ret <16 x i8> %cast