-
Notifications
You must be signed in to change notification settings - Fork 15.2k
[VPlan] Handle VPWidenSelectRecipe in tryToFoldLiveIns #150357
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[VPlan] Handle VPWidenSelectRecipe in tryToFoldLiveIns #150357
Conversation
This helps simplify VPBlendRecipes that are expanded to selects in another patch.
|
@llvm/pr-subscribers-vectorizers @llvm/pr-subscribers-llvm-transforms Author: Luke Lau (lukel97) ChangesThis helps simplify VPBlendRecipes that are expanded to selects in another patch. Full diff: https://github.com/llvm/llvm-project/pull/150357.diff 2 Files Affected:
diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
index 5da43b61c672e..be5bc6fdbd1e8 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
@@ -1079,6 +1079,12 @@ static void simplifyRecipe(VPRecipeBase &R, VPTypeAnalysis &TypeInfo) {
return;
}
+ if (match(Def, m_Select(m_True(), m_VPValue(X), m_VPValue())))
+ return Def->replaceAllUsesWith(X);
+
+ if (match(Def, m_Select(m_False(), m_VPValue(), m_VPValue(X))))
+ return Def->replaceAllUsesWith(X);
+
if (match(Def, m_Select(m_VPValue(), m_VPValue(X), m_Deferred(X))))
return Def->replaceAllUsesWith(X);
diff --git a/llvm/test/Transforms/LoopVectorize/X86/constant-fold.ll b/llvm/test/Transforms/LoopVectorize/X86/constant-fold.ll
index 05a495d51c458..86b28c3183166 100644
--- a/llvm/test/Transforms/LoopVectorize/X86/constant-fold.ll
+++ b/llvm/test/Transforms/LoopVectorize/X86/constant-fold.ll
@@ -67,17 +67,16 @@ define void @redundant_or_1(ptr %dst, i1 %c.0, i1 %c.1) {
; CHECK: vector.ph:
; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i1> poison, i1 [[C_0:%.*]], i64 0
; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i1> [[BROADCAST_SPLATINSERT]], <4 x i1> poison, <4 x i32> zeroinitializer
+; CHECK-NEXT: [[TMP0:%.*]] = xor <4 x i1> [[BROADCAST_SPLAT]], splat (i1 true)
; CHECK-NEXT: [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <4 x i1> poison, i1 [[C_1:%.*]], i64 0
; CHECK-NEXT: [[BROADCAST_SPLAT2:%.*]] = shufflevector <4 x i1> [[BROADCAST_SPLATINSERT1]], <4 x i1> poison, <4 x i32> zeroinitializer
-; CHECK-NEXT: [[TMP0:%.*]] = xor <4 x i1> [[BROADCAST_SPLAT2]], splat (i1 true)
-; CHECK-NEXT: [[TMP4:%.*]] = select i1 true, <4 x i1> [[BROADCAST_SPLAT]], <4 x i1> zeroinitializer
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK: vector.body:
; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_STORE_CONTINUE8:%.*]] ]
; CHECK-NEXT: [[VEC_IND:%.*]] = phi <4 x i8> [ <i8 0, i8 1, i8 2, i8 3>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[PRED_STORE_CONTINUE8]] ]
; CHECK-NEXT: [[TMP1:%.*]] = icmp ule <4 x i8> [[VEC_IND]], splat (i8 2)
; CHECK-NEXT: [[TMP2:%.*]] = select <4 x i1> [[TMP1]], <4 x i1> [[TMP0]], <4 x i1> zeroinitializer
-; CHECK-NEXT: [[TMP5:%.*]] = select <4 x i1> [[TMP2]], <4 x i1> [[TMP4]], <4 x i1> zeroinitializer
+; CHECK-NEXT: [[TMP5:%.*]] = select <4 x i1> [[TMP2]], <4 x i1> [[BROADCAST_SPLAT2]], <4 x i1> zeroinitializer
; CHECK-NEXT: [[TMP6:%.*]] = extractelement <4 x i1> [[TMP5]], i32 0
; CHECK-NEXT: br i1 [[TMP6]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE:%.*]]
; CHECK: pred.store.if:
@@ -120,11 +119,11 @@ define void @redundant_or_1(ptr %dst, i1 %c.0, i1 %c.1) {
; CHECK-NEXT: br label [[LOOP_HEADER:%.*]]
; CHECK: loop.header:
; CHECK-NEXT: [[IV:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP_LATCH:%.*]] ]
-; CHECK-NEXT: br i1 [[C_1]], label [[LOOP_LATCH]], label [[THEN_1:%.*]]
+; CHECK-NEXT: br i1 [[C_0]], label [[LOOP_LATCH]], label [[THEN_1:%.*]]
; CHECK: then.1:
; CHECK-NEXT: [[CMP:%.*]] = icmp eq i32 [[IV]], 2
; CHECK-NEXT: [[OR:%.*]] = or i1 [[CMP]], true
-; CHECK-NEXT: [[COND:%.*]] = select i1 [[OR]], i1 [[C_0]], i1 false
+; CHECK-NEXT: [[COND:%.*]] = select i1 [[OR]], i1 [[C_1]], i1 false
; CHECK-NEXT: br i1 [[COND]], label [[THEN_2:%.*]], label [[LOOP_LATCH]]
; CHECK: then.2:
; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds i32, ptr [[DST]], i32 [[IV]]
@@ -171,17 +170,16 @@ define void @redundant_or_2(ptr %dst, i1 %c.0, i1 %c.1) {
; CHECK: vector.ph:
; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i1> poison, i1 [[C_1:%.*]], i64 0
; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i1> [[BROADCAST_SPLATINSERT]], <4 x i1> poison, <4 x i32> zeroinitializer
+; CHECK-NEXT: [[TMP0:%.*]] = xor <4 x i1> [[BROADCAST_SPLAT]], splat (i1 true)
; CHECK-NEXT: [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <4 x i1> poison, i1 [[C_0:%.*]], i64 0
; CHECK-NEXT: [[BROADCAST_SPLAT2:%.*]] = shufflevector <4 x i1> [[BROADCAST_SPLATINSERT1]], <4 x i1> poison, <4 x i32> zeroinitializer
-; CHECK-NEXT: [[TMP0:%.*]] = xor <4 x i1> [[BROADCAST_SPLAT2]], splat (i1 true)
-; CHECK-NEXT: [[TMP1:%.*]] = select i1 true, <4 x i1> [[BROADCAST_SPLAT]], <4 x i1> zeroinitializer
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK: vector.body:
; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_STORE_CONTINUE8:%.*]] ]
; CHECK-NEXT: [[VEC_IND:%.*]] = phi <4 x i8> [ <i8 0, i8 1, i8 2, i8 3>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[PRED_STORE_CONTINUE8]] ]
; CHECK-NEXT: [[TMP2:%.*]] = icmp ule <4 x i8> [[VEC_IND]], splat (i8 2)
; CHECK-NEXT: [[TMP3:%.*]] = select <4 x i1> [[TMP2]], <4 x i1> [[TMP0]], <4 x i1> zeroinitializer
-; CHECK-NEXT: [[TMP4:%.*]] = select <4 x i1> [[TMP3]], <4 x i1> [[TMP1]], <4 x i1> zeroinitializer
+; CHECK-NEXT: [[TMP4:%.*]] = select <4 x i1> [[TMP3]], <4 x i1> [[BROADCAST_SPLAT2]], <4 x i1> zeroinitializer
; CHECK-NEXT: [[TMP5:%.*]] = extractelement <4 x i1> [[TMP4]], i32 0
; CHECK-NEXT: br i1 [[TMP5]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE:%.*]]
; CHECK: pred.store.if:
@@ -224,11 +222,11 @@ define void @redundant_or_2(ptr %dst, i1 %c.0, i1 %c.1) {
; CHECK-NEXT: br label [[LOOP_HEADER:%.*]]
; CHECK: loop.header:
; CHECK-NEXT: [[IV:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP_LATCH:%.*]] ]
-; CHECK-NEXT: br i1 [[C_0]], label [[LOOP_LATCH]], label [[THEN_1:%.*]]
+; CHECK-NEXT: br i1 [[C_1]], label [[LOOP_LATCH]], label [[THEN_1:%.*]]
; CHECK: then.1:
; CHECK-NEXT: [[CMP:%.*]] = icmp eq i32 [[IV]], 2
; CHECK-NEXT: [[OR:%.*]] = or i1 true, [[CMP]]
-; CHECK-NEXT: [[COND:%.*]] = select i1 [[OR]], i1 [[C_1]], i1 false
+; CHECK-NEXT: [[COND:%.*]] = select i1 [[OR]], i1 [[C_0]], i1 false
; CHECK-NEXT: br i1 [[COND]], label [[THEN_2:%.*]], label [[LOOP_LATCH]]
; CHECK: then.2:
; CHECK-NEXT: [[GEP:%.*]] = getelementptr inbounds i32, ptr [[DST]], i32 [[IV]]
|
artagnon
left a comment
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Seems like a good straight-forward improvement, but I wonder if a nested match m_Select(m_VPValue(Cond) ...) and a match(Cond, m_True()) would be worth it.
artagnon
left a comment
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I wonder why InstSimplifyFolder didn't kick in here?
I think some of these are exposed by other simplifications in simplifyRecipe, which tend to just mutate the operands in place rather than creating new instructions via VPBuilder |
The InstSimplifyFolder is used independently of the VPBuilder, by matching live-ins in simplifyRecipe. I think the broadcast_splat operand in the changed tests aren't live-ins, and that's the reason the folder fails. Any opinions on extending tryToFoldLiveIns to handle the case when some operands are live-ins? We're already doing a pretty extensive match there, and true/false in the select cond should be constant live-ins. |
Oh sorry you're right, I tried this out and it's because we don't handle VPWidenSelectRecipes with the InstSimplifyFolder. I'll rework this PR to support it |
artagnon
left a comment
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM, thanks!
fhahn
left a comment
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM, thanks
This helps simplify VPBlendRecipes that are expanded to selects in another patch.
This helps simplify VPBlendRecipes that are expanded to selects in another patch.