[PATCH] D53659: [ValueTracking] peek through shuffles in ComputeNumSignBits (PR37549)

Sanjay Patel via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Wed Oct 24 10:24:05 PDT 2018


spatel created this revision.
spatel added reviewers: lebedev.ri, craig.topper, RKSimon.
Herald added a subscriber: mcrosier.

The motivating case is from PR37549:
https://bugs.llvm.org/show_bug.cgi?id=37549

The analysis improvement allows us to form a vector 'select' out of bitwise logic (the use of ComputeNumSignBits was added at https://reviews.llvm.org/rL345149).

The smaller test shows another InstCombine improvement - we use ComputeNumSignBits to add 'nsw' to shift-left.


https://reviews.llvm.org/D53659

Files:
  lib/Analysis/ValueTracking.cpp
  test/Transforms/InstCombine/logical-select.ll
  test/Transforms/InstCombine/nsw.ll


Index: test/Transforms/InstCombine/nsw.ll
===================================================================
--- test/Transforms/InstCombine/nsw.ll
+++ test/Transforms/InstCombine/nsw.ll
@@ -99,13 +99,13 @@
   ret i8 %add
 }
 
-; TODO: ComputeNumSignBits()/computeKnownBits() should look through a shufflevector.
+; TODO: computeKnownBits() should look through a shufflevector.
 
 define <3 x i32> @shl_nuw_nsw_shuffle_splat_vec(<2 x i8> %x) {
 ; CHECK-LABEL: @shl_nuw_nsw_shuffle_splat_vec(
 ; CHECK-NEXT:    [[T2:%.*]] = zext <2 x i8> [[X:%.*]] to <2 x i32>
 ; CHECK-NEXT:    [[SHUF:%.*]] = shufflevector <2 x i32> [[T2]], <2 x i32> undef, <3 x i32> <i32 1, i32 0, i32 1>
-; CHECK-NEXT:    [[T3:%.*]] = shl <3 x i32> [[SHUF]], <i32 17, i32 17, i32 17>
+; CHECK-NEXT:    [[T3:%.*]] = shl nsw <3 x i32> [[SHUF]], <i32 17, i32 17, i32 17>
 ; CHECK-NEXT:    ret <3 x i32> [[T3]]
 ;
   %t2 = zext <2 x i8> %x to <2 x i32>
Index: test/Transforms/InstCombine/logical-select.ll
===================================================================
--- test/Transforms/InstCombine/logical-select.ll
+++ test/Transforms/InstCombine/logical-select.ll
@@ -594,13 +594,10 @@
 ; CHECK-NEXT:    [[S3:%.*]] = shufflevector <4 x i32> [[SHUF_OR1]], <4 x i32> undef, <4 x i32> <i32 0, i32 0, i32 1, i32 1>
 ; CHECK-NEXT:    [[S4:%.*]] = shufflevector <4 x i32> [[SHUF_OR1]], <4 x i32> undef, <4 x i32> <i32 2, i32 2, i32 3, i32 3>
 ; CHECK-NEXT:    [[SHUF_OR2:%.*]] = or <4 x i32> [[S3]], [[S4]]
-; CHECK-NEXT:    [[NOT_OR2:%.*]] = xor <4 x i32> [[SHUF_OR2]], <i32 -1, i32 -1, i32 -1, i32 -1>
-; CHECK-NEXT:    [[XBC:%.*]] = bitcast <4 x float> [[X]] to <4 x i32>
-; CHECK-NEXT:    [[ZBC:%.*]] = bitcast <4 x float> [[Z:%.*]] to <4 x i32>
-; CHECK-NEXT:    [[AND1:%.*]] = and <4 x i32> [[NOT_OR2]], [[XBC]]
-; CHECK-NEXT:    [[AND2:%.*]] = and <4 x i32> [[SHUF_OR2]], [[ZBC]]
-; CHECK-NEXT:    [[SEL:%.*]] = or <4 x i32> [[AND1]], [[AND2]]
-; CHECK-NEXT:    ret <4 x i32> [[SEL]]
+; CHECK-NEXT:    [[TMP1:%.*]] = trunc <4 x i32> [[SHUF_OR2]] to <4 x i1>
+; CHECK-NEXT:    [[DOTV:%.*]] = select <4 x i1> [[TMP1]], <4 x float> [[Z:%.*]], <4 x float> [[X]]
+; CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x float> [[DOTV]] to <4 x i32>
+; CHECK-NEXT:    ret <4 x i32> [[TMP2]]
 ;
   %cmp = fcmp ole <4 x float> %x, %y
   %sext = sext <4 x i1> %cmp to <4 x i32>
Index: lib/Analysis/ValueTracking.cpp
===================================================================
--- lib/Analysis/ValueTracking.cpp
+++ lib/Analysis/ValueTracking.cpp
@@ -2510,6 +2510,17 @@
     // valid for all elements of the vector (for example if vector is sign
     // extended, shifted, etc).
     return ComputeNumSignBits(U->getOperand(0), Depth + 1, Q);
+
+  case Instruction::ShuffleVector:
+    // Look through shuffle of 1 source vector.
+    if (isa<UndefValue>(U->getOperand(0)))
+      return ComputeNumSignBits(U->getOperand(1), Depth + 1, Q);
+    if (isa<UndefValue>(U->getOperand(1)))
+      return ComputeNumSignBits(U->getOperand(0), Depth + 1, Q);
+
+    // TODO: We can look through shuffles of 2 sources by computing the minimum
+    // sign bits for each operand (similar to what we do for binops).
+    break;
   }
 
   // Finally, if we can prove that the top bits of the result are 0's or 1's,


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D53659.170928.patch
Type: text/x-patch
Size: 3267 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20181024/8e5667c7/attachment.bin>


More information about the llvm-commits mailing list