[AArch64][GlobalISel] Add better basic legalization for llround. #168427
+23
−23
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
Member
|
@llvm/pr-subscribers-llvm-globalisel @llvm/pr-subscribers-backend-aarch64 Author: David Green (davemgreen) ChangesThis adds handling for f16 and f128 lround/llround under LP64 targets, promoting the f16 where needed and using a libcall for f128. This codegen is now identical to the selection dag version. Full diff: https://github.com/llvm/llvm-project/pull/168427.diff 6 Files Affected:
diff --git a/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp b/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
index cacb292acee18..2290e45bd999e 100644
--- a/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
@@ -3439,6 +3439,19 @@ LegalizerHelper::widenScalar(MachineInstr &MI, unsigned TypeIdx, LLT WideTy) {
Observer.changedInstr(MI);
return Legalized;
}
+ case TargetOpcode::G_LROUND:
+ case TargetOpcode::G_LLROUND:
+ Observer.changingInstr(MI);
+
+ if (TypeIdx == 0) {
+ widenScalarDst(MI, WideTy);
+ } else {
+ widenScalarSrc(MI, WideTy, 1, TargetOpcode::G_FPEXT);
+ }
+
+ Observer.changedInstr(MI);
+ return Legalized;
+
case TargetOpcode::G_INTTOPTR:
if (TypeIdx != 1)
return UnableToLegalize;
diff --git a/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp b/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
index a88817c9d2d19..fdf69b04bf676 100644
--- a/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
+++ b/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
@@ -449,10 +449,12 @@ AArch64LegalizerInfo::AArch64LegalizerInfo(const AArch64Subtarget &ST)
.minScalar(0, s32)
.libcallFor({{s32, s32}, {s64, s32}, {s128, s32}});
- // TODO: Libcall support for s128.
- // TODO: s16 should be legal with full FP16 support.
getActionDefinitionsBuilder({G_LROUND, G_LLROUND})
- .legalFor({{s64, s32}, {s64, s64}});
+ .legalFor({{s64, s32}, {s64, s64}})
+ .legalFor(HasFP16, {{s64, s16}})
+ .minScalar(0, s64)
+ .minScalar(1, s32)
+ .libcallFor({{s64, s128}});
// TODO: Custom legalization for mismatched types.
getActionDefinitionsBuilder(G_FCOPYSIGN)
diff --git a/llvm/test/CodeGen/AArch64/llround-conv-fp16.ll b/llvm/test/CodeGen/AArch64/llround-conv-fp16.ll
index cb042757a4a42..3a4be1bda7cd6 100644
--- a/llvm/test/CodeGen/AArch64/llround-conv-fp16.ll
+++ b/llvm/test/CodeGen/AArch64/llround-conv-fp16.ll
@@ -1,12 +1,8 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
; RUN: llc < %s -mtriple=aarch64 | FileCheck %s --check-prefixes=CHECK-NOFP16
; RUN: llc < %s -mtriple=aarch64 -mattr=+fullfp16 | FileCheck %s --check-prefixes=CHECK-FP16
-; RUN: llc < %s -mtriple=aarch64 -global-isel -global-isel-abort=2 2>&1 | FileCheck %s --check-prefixes=CHECK-NOFP16,CHECK-GI
-; RUN: llc < %s -mtriple=aarch64 -mattr=+fullfp16 -global-isel -global-isel-abort=2 2>&1 | FileCheck %s --check-prefixes=CHECK-FP16,CHECK-GI
-
-; CHECK-GI: warning: Instruction selection used fallback path for testmhhs
-; CHECK-GI-NEXT: warning: Instruction selection used fallback path for testmhws
-; CHECK-GI-NEXT: warning: Instruction selection used fallback path for testmhxs
+; RUN: llc < %s -mtriple=aarch64 -global-isel | FileCheck %s --check-prefixes=CHECK-NOFP16
+; RUN: llc < %s -mtriple=aarch64 -mattr=+fullfp16 -global-isel | FileCheck %s --check-prefixes=CHECK-FP16
define i16 @testmhhs(half %x) {
; CHECK-NOFP16-LABEL: testmhhs:
diff --git a/llvm/test/CodeGen/AArch64/llround-conv.ll b/llvm/test/CodeGen/AArch64/llround-conv.ll
index 4cc089804ce97..bdee73076347a 100644
--- a/llvm/test/CodeGen/AArch64/llround-conv.ll
+++ b/llvm/test/CodeGen/AArch64/llround-conv.ll
@@ -1,9 +1,6 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
; RUN: llc < %s -mtriple=aarch64 -mattr=+neon | FileCheck %s
-; RUN: llc < %s -mtriple=aarch64 -mattr=+neon -global-isel -global-isel-abort=2 2>&1 | FileCheck %s --check-prefixes=CHECK,CHECK-GI
-
-; CHECK-GI: warning: Instruction selection used fallback path for testmswl
-; CHECK-GI-NEXT: warning: Instruction selection used fallback path for testmsll
+; RUN: llc < %s -mtriple=aarch64 -mattr=+neon -global-isel | FileCheck %s
define i32 @testmsws(float %x) {
; CHECK-LABEL: testmsws:
diff --git a/llvm/test/CodeGen/AArch64/lround-conv-fp16.ll b/llvm/test/CodeGen/AArch64/lround-conv-fp16.ll
index a29dea0eb9f9f..0b18f220067ca 100644
--- a/llvm/test/CodeGen/AArch64/lround-conv-fp16.ll
+++ b/llvm/test/CodeGen/AArch64/lround-conv-fp16.ll
@@ -1,12 +1,8 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
; RUN: llc < %s -mtriple=aarch64 | FileCheck %s --check-prefixes=CHECK-NOFP16
; RUN: llc < %s -mtriple=aarch64 -mattr=+fullfp16 | FileCheck %s --check-prefixes=CHECK-FP16
-; RUN: llc < %s -mtriple=aarch64 -global-isel -global-isel-abort=2 2>&1 | FileCheck %s --check-prefixes=CHECK-NOFP16,CHECK-GI
-; RUN: llc < %s -mtriple=aarch64 -mattr=+fullfp16 -global-isel -global-isel-abort=2 2>&1 | FileCheck %s --check-prefixes=CHECK-FP16,CHECK-GI
-
-; CHECK-GI: warning: Instruction selection used fallback path for testmhhs
-; CHECK-GI-NEXT: warning: Instruction selection used fallback path for testmhws
-; CHECK-GI-NEXT: warning: Instruction selection used fallback path for testmhxs
+; RUN: llc < %s -mtriple=aarch64 -global-isel | FileCheck %s --check-prefixes=CHECK-NOFP16
+; RUN: llc < %s -mtriple=aarch64 -mattr=+fullfp16 -global-isel | FileCheck %s --check-prefixes=CHECK-FP16
define i16 @testmhhs(half %x) {
; CHECK-NOFP16-LABEL: testmhhs:
diff --git a/llvm/test/CodeGen/AArch64/lround-conv.ll b/llvm/test/CodeGen/AArch64/lround-conv.ll
index 0bf82b538e70c..4b1782457cc10 100644
--- a/llvm/test/CodeGen/AArch64/lround-conv.ll
+++ b/llvm/test/CodeGen/AArch64/lround-conv.ll
@@ -1,9 +1,6 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
; RUN: llc < %s -mtriple=aarch64 -mattr=+neon | FileCheck %s
-; RUN: llc < %s -mtriple=aarch64 -mattr=+neon -global-isel -global-isel-abort=2 2>&1 | FileCheck %s --check-prefixes=CHECK,CHECK-GI
-
-; CHECK-GI: warning: Instruction selection used fallback path for testmswl
-; CHECK-GI-NEXT: warning: Instruction selection used fallback path for testmsll
+; RUN: llc < %s -mtriple=aarch64 -mattr=+neon -global-isel | FileCheck %s
define i32 @testmsws(float %x) {
; CHECK-LABEL: testmsws:
|
🐧 Linux x64 Test Results
|
aemerson
approved these changes
Nov 18, 2025
| case TargetOpcode::G_LLROUND: | ||
| Observer.changingInstr(MI); | ||
|
|
||
| if (TypeIdx == 0) { |
This adds handling for f16 and f128 lround/llround under LP64 targets, promoting the f16 where needed and using a libcall for f128. This codegen is now identical to the selection dag version.
davemgreen
force-pushed
the
gh-gi-llround
branch
from
5aa8484 to
20b127f
Compare
Lukacma
reviewed
Nov 19, 2025
| getActionDefinitionsBuilder({G_LROUND, G_LLROUND}) | ||
| .legalFor({{s64, s32}, {s64, s64}}); | ||
| .legalFor({{s64, s32}, {s64, s64}}) | ||
| .legalFor(HasFP16, {{s64, s16}}) |
Contributor
There was a problem hiding this comment.
Is there a specific reason why {s32, s16} and {s32, s64} is not legal for LROUND here? If I understand correctly, that's how it is for SelectionDAG
Collaborator
Author
There was a problem hiding this comment.
Are you meaning the windows variants? The l and ll in lround/llround stand for long and long-long AFAIU. (You could argue that info shouldn't be present in IR intrinsics). So on linux where long is 64bit you would only expect a i64, on windows i32 would be used for lround. We haven't tried to support windows for them yet, there is a separate test case. If we can generate instructions for them then hopefully we can just always support them, irrespective of whether it is windows or not, providing we do not need to generate a libcall.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
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.
This adds handling for f16 and f128 lround/llround under LP64 targets, promoting the f16 where needed and using a libcall for f128. This codegen is now identical to the selection dag version.