[llvm] 4c44b02 - [InstCombine] fold ctpop-of-select with 1 or more constant arms

[Sanjay Patel via llvm-commits]

Author: Sanjay Patel
Date: 2021-06-20T11:28:45-04:00
New Revision: 4c44b02d87bce42452594096cc0e9245d94596a8

URL: https://github.com/llvm/llvm-project/commit/4c44b02d87bce42452594096cc0e9245d94596a8
DIFF: https://github.com/llvm/llvm-project/commit/4c44b02d87bce42452594096cc0e9245d94596a8.diff

LOG: [InstCombine] fold ctpop-of-select with 1 or more constant arms

The general pattern is mentioned in:
https://llvm.org/PR50140
...but we need to do a bit more to handle intrinsics with extra operands
like ctlz/cttz.

Added: 
    

Modified: 
    llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
    llvm/lib/Transforms/InstCombine/InstructionCombining.cpp
    llvm/test/Transforms/InstCombine/intrinsic-select.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
index c5911fc507d9..4059a723a785 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
@@ -1085,6 +1085,11 @@ Instruction *InstCombinerImpl::visitCallInst(CallInst &CI) {
   case Intrinsic::ctpop:
     if (auto *I = foldCtpop(*II, *this))
       return I;
+
+    // If the operand is a select with constant arm(s), try to hoist ctpop.
+    if (auto *Sel = dyn_cast<SelectInst>(II->getArgOperand(0)))
+      if (Instruction *R = FoldOpIntoSelect(*II, Sel))
+        return R;
     break;
 
   case Intrinsic::fshl:

diff  --git a/llvm/lib/Transforms/InstCombine/InstructionCombining.cpp b/llvm/lib/Transforms/InstCombine/InstructionCombining.cpp
index d93a45253dfd..30b96d38fd69 100644
--- a/llvm/lib/Transforms/InstCombine/InstructionCombining.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstructionCombining.cpp
@@ -938,6 +938,13 @@ static Value *foldOperationIntoSelectOperand(Instruction &I, Value *SO,
   if (auto *Cast = dyn_cast<CastInst>(&I))
     return Builder.CreateCast(Cast->getOpcode(), SO, I.getType());
 
+  if (auto *II = dyn_cast<IntrinsicInst>(&I)) {
+    assert(canConstantFoldCallTo(II, cast<Function>(II->getCalledOperand())) &&
+           "Expected constant-foldable intrinsic");
+
+    return Builder.CreateIntrinsic(II->getIntrinsicID(), I.getType(), SO);
+  }
+
   assert(I.isBinaryOp() && "Unexpected opcode for select folding");
 
   // Figure out if the constant is the left or the right argument.

diff  --git a/llvm/test/Transforms/InstCombine/intrinsic-select.ll b/llvm/test/Transforms/InstCombine/intrinsic-select.ll
index 2b2522f5aae5..1ee792ded641 100644
--- a/llvm/test/Transforms/InstCombine/intrinsic-select.ll
+++ b/llvm/test/Transforms/InstCombine/intrinsic-select.ll
@@ -106,8 +106,7 @@ define i32 @cttz_sel_const_true_false_extra_use(i1 %b) {
 
 define i32 @ctpop_sel_const_true_false(i1 %b) {
 ; CHECK-LABEL: @ctpop_sel_const_true_false(
-; CHECK-NEXT:    [[S:%.*]] = select i1 [[B:%.*]], i32 5, i32 -7
-; CHECK-NEXT:    [[C:%.*]] = call i32 @llvm.ctpop.i32(i32 [[S]]), !range [[RNG3:![0-9]+]]
+; CHECK-NEXT:    [[C:%.*]] = select i1 [[B:%.*]], i32 2, i32 30
 ; CHECK-NEXT:    ret i32 [[C]]
 ;
   %s = select i1 %b, i32 5, i32 -7
@@ -117,8 +116,8 @@ define i32 @ctpop_sel_const_true_false(i1 %b) {
 
 define i32 @ctpop_sel_const_true(i1 %b, i32 %x) {
 ; CHECK-LABEL: @ctpop_sel_const_true(
-; CHECK-NEXT:    [[S:%.*]] = select i1 [[B:%.*]], i32 5, i32 [[X:%.*]]
-; CHECK-NEXT:    [[C:%.*]] = call i32 @llvm.ctpop.i32(i32 [[S]]), !range [[RNG1]]
+; CHECK-NEXT:    [[TMP1:%.*]] = call i32 @llvm.ctpop.i32(i32 [[X:%.*]]), !range [[RNG1]]
+; CHECK-NEXT:    [[C:%.*]] = select i1 [[B:%.*]], i32 2, i32 [[TMP1]]
 ; CHECK-NEXT:    ret i32 [[C]]
 ;
   %s = select i1 %b, i32 5, i32 %x
@@ -128,8 +127,8 @@ define i32 @ctpop_sel_const_true(i1 %b, i32 %x) {
 
 define <3 x i7> @ctpop_sel_const_false(<3 x i1> %b, <3 x i7> %x) {
 ; CHECK-LABEL: @ctpop_sel_const_false(
-; CHECK-NEXT:    [[S:%.*]] = select <3 x i1> [[B:%.*]], <3 x i7> [[X:%.*]], <3 x i7> <i7 7, i7 -1, i7 0>
-; CHECK-NEXT:    [[C:%.*]] = call <3 x i7> @llvm.ctpop.v3i7(<3 x i7> [[S]])
+; CHECK-NEXT:    [[TMP1:%.*]] = call <3 x i7> @llvm.ctpop.v3i7(<3 x i7> [[X:%.*]])
+; CHECK-NEXT:    [[C:%.*]] = select <3 x i1> [[B:%.*]], <3 x i7> [[TMP1]], <3 x i7> <i7 3, i7 7, i7 0>
 ; CHECK-NEXT:    ret <3 x i7> [[C]]
 ;
   %s = select <3 x i1> %b, <3 x i7> %x, <3 x i7> <i7 7, i7 -1, i7 0>
@@ -141,7 +140,7 @@ define i32 @ctpop_sel_const_true_false_extra_use(i1 %b) {
 ; CHECK-LABEL: @ctpop_sel_const_true_false_extra_use(
 ; CHECK-NEXT:    [[S:%.*]] = select i1 [[B:%.*]], i32 5, i32 7
 ; CHECK-NEXT:    call void @use(i32 [[S]])
-; CHECK-NEXT:    [[C:%.*]] = call i32 @llvm.ctpop.i32(i32 [[S]]), !range [[RNG4:![0-9]+]]
+; CHECK-NEXT:    [[C:%.*]] = call i32 @llvm.ctpop.i32(i32 [[S]]), !range [[RNG3:![0-9]+]]
 ; CHECK-NEXT:    ret i32 [[C]]
 ;
   %s = select i1 %b, i32 5, i32 7