[TTI] Provide a cost for memset_pattern which matches the libcall #139978
+43
−0
Conversation
This file contains hidden or 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
The motivation is that differences in unrolling were noticed when trying to swich from the libcall to the intrinsic. Neither cost is a good, well considered cost, but for the moment, let's have them be equal to simplify migration.
|
@llvm/pr-subscribers-backend-x86 @llvm/pr-subscribers-llvm-analysis Author: Philip Reames (preames) ChangesThe motivation is that differences in unrolling were noticed when trying to switch from the libcall to the intrinsic. There are likely also differences not yet noticed in other cost based decisions - such as inlining, and possibly vectorization. Neither cost is a good, well considered, cost but for the moment, let's have them be equal to simplify migration. We can come back and refine this once we have it being exercised by default. Full diff: https://github.com/llvm/llvm-project/pull/139978.diff 2 Files Affected:
diff --git a/llvm/include/llvm/CodeGen/BasicTTIImpl.h b/llvm/include/llvm/CodeGen/BasicTTIImpl.h
index ff8778168686d..449e9e8f85561 100644
--- a/llvm/include/llvm/CodeGen/BasicTTIImpl.h
+++ b/llvm/include/llvm/CodeGen/BasicTTIImpl.h
@@ -2408,6 +2408,9 @@ class BasicTTIImplBase : public TargetTransformInfoImplCRTPBase<T> {
CmpInst::ICMP_ULT, CostKind);
return Cost;
}
+ case Intrinsic::experimental_memset_pattern:
+ // This cost is set to match the cost of the memset_pattern16 libcall
+ return TTI::TCC_Basic * 4;
case Intrinsic::abs:
ISD = ISD::ABS;
break;
diff --git a/llvm/test/Analysis/CostModel/X86/memset-pattern.ll b/llvm/test/Analysis/CostModel/X86/memset-pattern.ll
new file mode 100644
index 0000000000000..aa0c6efdf34fa
--- /dev/null
+++ b/llvm/test/Analysis/CostModel/X86/memset-pattern.ll
@@ -0,0 +1,40 @@
+; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py UTC_ARGS: --version 4
+; RUN: opt < %s -mtriple=x86_64-apple-darwin10.0.0 -passes="print<cost-model>" 2>&1 -disable-output | FileCheck %s
+
+target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64"
+
+target triple = "x86_64-apple-darwin10.0.0"
+
+@.memset_pattern = private unnamed_addr constant [4 x i32] [i32 2, i32 2, i32 2, i32 2], align 16
+
+define void @via_libcall(ptr %p) nounwind ssp {
+; CHECK-LABEL: 'via_libcall'
+; CHECK-NEXT: Cost Model: Found an estimated cost of 4 for instruction: call void @memset_pattern4(ptr %p, ptr @.memset_pattern, i64 200)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 4 for instruction: call void @memset_pattern8(ptr %p, ptr @.memset_pattern, i64 200)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 4 for instruction: call void @memset_pattern16(ptr %p, ptr @.memset_pattern, i64 200)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+ call void @memset_pattern4(ptr %p, ptr @.memset_pattern, i64 200)
+ call void @memset_pattern8(ptr %p, ptr @.memset_pattern, i64 200)
+ call void @memset_pattern16(ptr %p, ptr @.memset_pattern, i64 200)
+ ret void
+}
+
+declare void @memset_pattern4(ptr, ptr, i64)
+declare void @memset_pattern8(ptr, ptr, i64)
+declare void @memset_pattern16(ptr, ptr, i64)
+
+define void @via_intrinsic(ptr %p) {
+; CHECK-LABEL: 'via_intrinsic'
+; CHECK-NEXT: Cost Model: Found an estimated cost of 4 for instruction: call void @llvm.experimental.memset.pattern.p0.i16.i64(ptr align 4 %p, i16 2, i64 100, i1 false)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 4 for instruction: call void @llvm.experimental.memset.pattern.p0.i32.i64(ptr align 4 %p, i32 2, i64 50, i1 false)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 4 for instruction: call void @llvm.experimental.memset.pattern.p0.i64.i64(ptr align 4 %p, i64 2, i64 25, i1 false)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 4 for instruction: call void @llvm.experimental.memset.pattern.p0.i128.i64(ptr align 4 %p, i128 2, i64 12, i1 false)
+; CHECK-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+;
+ call void @llvm.experimental.memset.pattern(ptr align 4 %p, i16 2, i64 100, i1 false)
+ call void @llvm.experimental.memset.pattern(ptr align 4 %p, i32 2, i64 50, i1 false)
+ call void @llvm.experimental.memset.pattern(ptr align 4 %p, i64 2, i64 25, i1 false)
+ call void @llvm.experimental.memset.pattern(ptr align 4 %p, i128 2, i64 12, i1 false)
+ ret void
+}
|
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.
The motivation is that differences in unrolling were noticed when trying to switch from the libcall to the intrinsic. There are likely also differences not yet noticed in other cost based decisions - such as inlining, and possibly vectorization.
Neither cost is a good, well considered, cost but for the moment, let's have them be equal to simplify migration. We can come back and refine this once we have it being exercised by default.