LoopIdiom/test: add negative tests for fpowi #68985
Closed
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Compiling the following program under -ffast-math results in very poor
codegen due to LoopVectorize:
float fpowi(float base, int exp) {
float result = 1.0;
while (exp != 0) {
result *= base;
--exp;
}
return result;
}
In preparation to get LoopIdiomRecognize to convert this idiom to a
llvm.powi, add negative tests corresponding to variations of this
program.
|
@llvm/pr-subscribers-llvm-transforms Author: Ramkumar Ramachandra (artagnon) ChangesCompiling the following program under -ffast-math results in very poor codegen due to LoopVectorize: float fpowi(float base, int exp) {
float result = 1.0;
while (exp != 0) {
result *= base;
--exp;
}
return result;
}In preparation to get LoopIdiomRecognize to convert this idiom to a llvm.powi, add negative tests corresponding to variations of this program. Full diff: https://github.com/llvm/llvm-project/pull/68985.diff 1 Files Affected:
diff --git a/llvm/test/Transforms/LoopIdiom/fpowi.ll b/llvm/test/Transforms/LoopIdiom/fpowi.ll
new file mode 100644
index 000000000000000..640921cb8c367f3
--- /dev/null
+++ b/llvm/test/Transforms/LoopIdiom/fpowi.ll
@@ -0,0 +1,122 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt -passes=loop-idiom < %s -S | FileCheck %s
+
+; TODO: Currently, LIR doesn't recognize the fpowi idiom, but it could in the
+; future, generating llvm.powi intrinsics for the positive tests.
+
+define float @fpowi_f32(float %base, i32 %exp) {
+; CHECK-LABEL: @fpowi_f32(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: [[CMP_NOT:%.*]] = icmp eq i32 [[EXP:%.*]], 0
+; CHECK-NEXT: br i1 [[CMP_NOT]], label [[WHILE_END:%.*]], label [[WHILE_BODY_PREHEADER:%.*]]
+; CHECK: while.body.preheader:
+; CHECK-NEXT: br label [[WHILE_BODY:%.*]]
+; CHECK: while.body:
+; CHECK-NEXT: [[RESULT:%.*]] = phi float [ [[MUL:%.*]], [[WHILE_BODY]] ], [ 1.000000e+00, [[WHILE_BODY_PREHEADER]] ]
+; CHECK-NEXT: [[MERGE_DEC:%.*]] = phi i32 [ [[DEC:%.*]], [[WHILE_BODY]] ], [ [[EXP]], [[WHILE_BODY_PREHEADER]] ]
+; CHECK-NEXT: [[MUL]] = fmul fast float [[RESULT]], [[BASE:%.*]]
+; CHECK-NEXT: [[DEC]] = add nsw i32 [[MERGE_DEC]], -1
+; CHECK-NEXT: [[CMP_EQ:%.*]] = icmp eq i32 [[DEC]], 0
+; CHECK-NEXT: br i1 [[CMP_EQ]], label [[WHILE_END_LOOPEXIT:%.*]], label [[WHILE_BODY]]
+; CHECK: while.end.loopexit:
+; CHECK-NEXT: [[MUL_LCSSA:%.*]] = phi float [ [[MUL]], [[WHILE_BODY]] ]
+; CHECK-NEXT: br label [[WHILE_END]]
+; CHECK: while.end:
+; CHECK-NEXT: [[RESULT_LCSSA:%.*]] = phi float [ 1.000000e+00, [[ENTRY:%.*]] ], [ [[MUL_LCSSA]], [[WHILE_END_LOOPEXIT]] ]
+; CHECK-NEXT: ret float [[RESULT_LCSSA]]
+;
+entry:
+ %cmp.not = icmp eq i32 %exp, 0
+ br i1 %cmp.not, label %while.end, label %while.body
+
+while.body: ; preds = %entry, %while.body
+ %result = phi float [ %mul, %while.body ], [ 1.000000e+00, %entry ]
+ %merge.dec = phi i32 [ %dec, %while.body ], [ %exp, %entry ]
+ %mul = fmul fast float %result, %base
+ %dec = add nsw i32 %merge.dec, -1
+ %cmp.eq = icmp eq i32 %dec, 0
+ br i1 %cmp.eq, label %while.end, label %while.body
+
+while.end: ; preds = %while.body, %entry
+ %result.lcssa = phi float [ 1.000000e+00, %entry ], [ %mul, %while.body ]
+ ret float %result.lcssa
+}
+
+define double @fpowi_f64(double %base, i32 %exp) {
+; CHECK-LABEL: @fpowi_f64(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: [[CMP_NOT:%.*]] = icmp eq i32 [[EXP:%.*]], 0
+; CHECK-NEXT: br i1 [[CMP_NOT]], label [[WHILE_END:%.*]], label [[WHILE_BODY_PREHEADER:%.*]]
+; CHECK: while.body.preheader:
+; CHECK-NEXT: br label [[WHILE_BODY:%.*]]
+; CHECK: while.body:
+; CHECK-NEXT: [[RESULT:%.*]] = phi double [ [[MUL:%.*]], [[WHILE_BODY]] ], [ 1.000000e+00, [[WHILE_BODY_PREHEADER]] ]
+; CHECK-NEXT: [[MERGE_DEC:%.*]] = phi i32 [ [[DEC:%.*]], [[WHILE_BODY]] ], [ [[EXP]], [[WHILE_BODY_PREHEADER]] ]
+; CHECK-NEXT: [[MUL]] = fmul fast double [[RESULT]], [[BASE:%.*]]
+; CHECK-NEXT: [[DEC]] = add nsw i32 [[MERGE_DEC]], -1
+; CHECK-NEXT: [[CMP_EQ:%.*]] = icmp eq i32 [[DEC]], 0
+; CHECK-NEXT: br i1 [[CMP_EQ]], label [[WHILE_END_LOOPEXIT:%.*]], label [[WHILE_BODY]]
+; CHECK: while.end.loopexit:
+; CHECK-NEXT: [[MUL_LCSSA:%.*]] = phi double [ [[MUL]], [[WHILE_BODY]] ]
+; CHECK-NEXT: br label [[WHILE_END]]
+; CHECK: while.end:
+; CHECK-NEXT: [[RESULT_LCSSA:%.*]] = phi double [ 1.000000e+00, [[ENTRY:%.*]] ], [ [[MUL_LCSSA]], [[WHILE_END_LOOPEXIT]] ]
+; CHECK-NEXT: ret double [[RESULT_LCSSA]]
+;
+entry:
+ %cmp.not = icmp eq i32 %exp, 0
+ br i1 %cmp.not, label %while.end, label %while.body
+
+while.body: ; preds = %entry, %while.body
+ %result = phi double [ %mul, %while.body ], [ 1.000000e+00, %entry ]
+ %merge.dec = phi i32 [ %dec, %while.body ], [ %exp, %entry ]
+ %mul = fmul fast double %result, %base
+ %dec = add nsw i32 %merge.dec, -1
+ %cmp.eq = icmp eq i32 %dec, 0
+ br i1 %cmp.eq, label %while.end, label %while.body
+
+while.end: ; preds = %while.body, %entry
+ %result.lcssa = phi double [ 1.000000e+00, %entry ], [ %mul, %while.body ]
+ ret double %result.lcssa
+}
+
+; Negative tests
+
+; The loop idiom is only legal in -ffast-math mode
+define float @fpowi_nofast(float %base, i32 %exp) {
+; CHECK-LABEL: @fpowi_nofast(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: [[CMP_NOT:%.*]] = icmp eq i32 [[EXP:%.*]], 0
+; CHECK-NEXT: br i1 [[CMP_NOT]], label [[WHILE_END:%.*]], label [[WHILE_BODY_PREHEADER:%.*]]
+; CHECK: while.body.preheader:
+; CHECK-NEXT: br label [[WHILE_BODY:%.*]]
+; CHECK: while.body:
+; CHECK-NEXT: [[RESULT:%.*]] = phi float [ [[MUL:%.*]], [[WHILE_BODY]] ], [ 1.000000e+00, [[WHILE_BODY_PREHEADER]] ]
+; CHECK-NEXT: [[MERGE_DEC:%.*]] = phi i32 [ [[DEC:%.*]], [[WHILE_BODY]] ], [ [[EXP]], [[WHILE_BODY_PREHEADER]] ]
+; CHECK-NEXT: [[MUL]] = fmul float [[RESULT]], [[BASE:%.*]]
+; CHECK-NEXT: [[DEC]] = add nsw i32 [[MERGE_DEC]], -1
+; CHECK-NEXT: [[CMP_EQ:%.*]] = icmp eq i32 [[DEC]], 0
+; CHECK-NEXT: br i1 [[CMP_EQ]], label [[WHILE_END_LOOPEXIT:%.*]], label [[WHILE_BODY]]
+; CHECK: while.end.loopexit:
+; CHECK-NEXT: [[MUL_LCSSA:%.*]] = phi float [ [[MUL]], [[WHILE_BODY]] ]
+; CHECK-NEXT: br label [[WHILE_END]]
+; CHECK: while.end:
+; CHECK-NEXT: [[RESULT_LCSSA:%.*]] = phi float [ 1.000000e+00, [[ENTRY:%.*]] ], [ [[MUL_LCSSA]], [[WHILE_END_LOOPEXIT]] ]
+; CHECK-NEXT: ret float [[RESULT_LCSSA]]
+;
+entry:
+ %cmp.not = icmp eq i32 %exp, 0
+ br i1 %cmp.not, label %while.end, label %while.body
+
+while.body: ; preds = %entry, %while.body
+ %result = phi float [ %mul, %while.body ], [ 1.000000e+00, %entry ]
+ %merge.dec = phi i32 [ %dec, %while.body ], [ %exp, %entry ]
+ %mul = fmul float %result, %base
+ %dec = add nsw i32 %merge.dec, -1
+ %cmp.eq = icmp eq i32 %dec, 0
+ br i1 %cmp.eq, label %while.end, label %while.body
+
+while.end: ; preds = %while.body, %entry
+ %result.lcssa = phi float [ 1.000000e+00, %entry ], [ %mul, %while.body ]
+ ret float %result.lcssa
+}
|
|
What are we supposed to do for negative exp value? |
|
Will open a new PR, once I've finished implementing the idiom in LIR. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Compiling the following program under -ffast-math results in very poor codegen due to LoopVectorize:
In preparation to get LoopIdiomRecognize to convert this idiom to a llvm.powi, add negative tests corresponding to variations of this program.