[PATCH] D66965: [InstCombine] recognize bswap disguised as shufflevector

[Sanjay Patel via Phabricator via llvm-commits]

spatel updated this revision to Diff 217969.
spatel marked 2 inline comments as done.
spatel added a comment.

Patch updated:

1. Optimization: rearranged comparisons for better perf.
2. Added widening shuffle negative test.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D66965/new/

https://reviews.llvm.org/D66965

Files:
  llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp
  llvm/test/Transforms/InstCombine/bswap.ll


Index: llvm/test/Transforms/InstCombine/bswap.ll
===================================================================
--- llvm/test/Transforms/InstCombine/bswap.ll
+++ llvm/test/Transforms/InstCombine/bswap.ll
@@ -233,8 +233,8 @@
 
 define i32 @shuf_4bytes(<4 x i8> %x) {
 ; CHECK-LABEL: @shuf_4bytes(
-; CHECK-NEXT:    [[BSWAP:%.*]] = shufflevector <4 x i8> [[X:%.*]], <4 x i8> undef, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
-; CHECK-NEXT:    [[CAST:%.*]] = bitcast <4 x i8> [[BSWAP]] to i32
+; CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i8> [[X:%.*]] to i32
+; CHECK-NEXT:    [[CAST:%.*]] = call i32 @llvm.bswap.i32(i32 [[TMP1]])
 ; CHECK-NEXT:    ret i32 [[CAST]]
 ;
   %bswap = shufflevector <4 x i8> %x, <4 x i8> undef, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
@@ -244,9 +244,9 @@
 
 define i32 @shuf_load_4bytes(<4 x i8>* %p) {
 ; CHECK-LABEL: @shuf_load_4bytes(
-; CHECK-NEXT:    [[X:%.*]] = load <4 x i8>, <4 x i8>* [[P:%.*]], align 4
-; CHECK-NEXT:    [[BSWAP:%.*]] = shufflevector <4 x i8> [[X]], <4 x i8> undef, <4 x i32> <i32 3, i32 2, i32 undef, i32 0>
-; CHECK-NEXT:    [[CAST:%.*]] = bitcast <4 x i8> [[BSWAP]] to i32
+; CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i8>* [[P:%.*]] to i32*
+; CHECK-NEXT:    [[X1:%.*]] = load i32, i32* [[TMP1]], align 4
+; CHECK-NEXT:    [[CAST:%.*]] = call i32 @llvm.bswap.i32(i32 [[X1]])
 ; CHECK-NEXT:    ret i32 [[CAST]]
 ;
   %x = load <4 x i8>, <4 x i8>* %p
@@ -257,9 +257,7 @@
 
 define i32 @shuf_bitcast_twice_4bytes(i32 %x) {
 ; CHECK-LABEL: @shuf_bitcast_twice_4bytes(
-; CHECK-NEXT:    [[CAST1:%.*]] = bitcast i32 [[X:%.*]] to <4 x i8>
-; CHECK-NEXT:    [[BSWAP:%.*]] = shufflevector <4 x i8> [[CAST1]], <4 x i8> undef, <4 x i32> <i32 undef, i32 2, i32 1, i32 0>
-; CHECK-NEXT:    [[CAST2:%.*]] = bitcast <4 x i8> [[BSWAP]] to i32
+; CHECK-NEXT:    [[CAST2:%.*]] = call i32 @llvm.bswap.i32(i32 [[X:%.*]])
 ; CHECK-NEXT:    ret i32 [[CAST2]]
 ;
   %cast1 = bitcast i32 %x to <4 x i8>
@@ -268,6 +266,7 @@
   ret i32 %cast2
 }
 
+; Negative test - extra use
 declare void @use(<4 x i8>)
 
 define i32 @shuf_4bytes_extra_use(<4 x i8> %x) {
@@ -283,6 +282,8 @@
   ret i32 %cast
 }
 
+; Negative test - scalar type is not in the data layout
+
 define i128 @shuf_16bytes(<16 x i8> %x) {
 ; CHECK-LABEL: @shuf_16bytes(
 ; CHECK-NEXT:    [[BSWAP:%.*]] = shufflevector <16 x i8> [[X:%.*]], <16 x i8> undef, <16 x i32> <i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
@@ -294,6 +295,8 @@
   ret i128 %cast
 }
 
+; Negative test - don't touch widening shuffles (for now)
+
 define i32 @shuf_2bytes_widening(<2 x i8> %x) {
 ; CHECK-LABEL: @shuf_2bytes_widening(
 ; CHECK-NEXT:    [[BSWAP:%.*]] = shufflevector <2 x i8> [[X:%.*]], <2 x i8> undef, <4 x i32> <i32 1, i32 0, i32 undef, i32 undef>
Index: llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp
===================================================================
--- llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp
+++ llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp
@@ -2416,6 +2416,22 @@
         return new ShuffleVectorInst(LHS, RHS, Shuf->getOperand(2));
       }
     }
+
+    // A bitcasted-to-scalar and byte-reversing shuffle is better recognized as
+    // a byte-swap:
+    // bitcast <N x i8> (shuf X, undef, <N, N-1,...0>) --> bswap (bitcast X)
+    // TODO: We should match the related pattern for bitreverse.
+    if (DestTy->isIntegerTy() &&
+        DL.isLegalInteger(DestTy->getScalarSizeInBits()) &&
+        SrcTy->getScalarSizeInBits() == 8 && NumShufElts % 2 == 0 &&
+        Shuf->hasOneUse() && Shuf->isReverse()) {
+      assert(ShufOp0->getType() == SrcTy && "Unexpected shuffle mask");
+      assert(isa<UndefValue>(ShufOp1) && "Unexpected shuffle op");
+      Function *Bswap =
+          Intrinsic::getDeclaration(CI.getModule(), Intrinsic::bswap, DestTy);
+      Value *ScalarX = Builder.CreateBitCast(ShufOp0, DestTy);
+      return IntrinsicInst::Create(Bswap, { ScalarX });
+    }
   }
 
   // Handle the A->B->A cast, and there is an intervening PHI node.


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D66965.217969.patch
Type: text/x-patch
Size: 4068 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20190829/c089bbaa/attachment.bin>