[AArch64] Remove wrong processor feature #151289
Open
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
tomershafir
force-pushed
the
aarch64-remove-wrong-processor-feature
branch
from
f76a5c5 to
30f404e
Compare
|
@llvm/pr-subscribers-backend-aarch64 Author: Tomer Shafir (tomershafir) ChangesFull diff: https://github.com/llvm/llvm-project/pull/151289.diff 2 Files Affected:
diff --git a/llvm/lib/Target/AArch64/AArch64Processors.td b/llvm/lib/Target/AArch64/AArch64Processors.td
index adc984ad795af..fa037df531028 100644
--- a/llvm/lib/Target/AArch64/AArch64Processors.td
+++ b/llvm/lib/Target/AArch64/AArch64Processors.td
@@ -318,7 +318,6 @@ def TuneAppleA7 : SubtargetFeature<"apple-a7", "ARMProcFamily", "AppleA7",
FeatureFuseAES, FeatureFuseCryptoEOR,
FeatureStorePairSuppress,
FeatureZCRegMoveGPR64,
- FeatureZCRegMoveFPR64,
FeatureZCZeroing,
FeatureZCZeroingFPWorkaround]>;
@@ -332,7 +331,6 @@ def TuneAppleA10 : SubtargetFeature<"apple-a10", "ARMProcFamily", "AppleA10",
FeatureFuseCryptoEOR,
FeatureStorePairSuppress,
FeatureZCRegMoveGPR64,
- FeatureZCRegMoveFPR64,
FeatureZCZeroing]>;
def TuneAppleA11 : SubtargetFeature<"apple-a11", "ARMProcFamily", "AppleA11",
@@ -345,7 +343,6 @@ def TuneAppleA11 : SubtargetFeature<"apple-a11", "ARMProcFamily", "AppleA11",
FeatureFuseCryptoEOR,
FeatureStorePairSuppress,
FeatureZCRegMoveGPR64,
- FeatureZCRegMoveFPR64,
FeatureZCZeroing]>;
def TuneAppleA12 : SubtargetFeature<"apple-a12", "ARMProcFamily", "AppleA12",
@@ -358,7 +355,6 @@ def TuneAppleA12 : SubtargetFeature<"apple-a12", "ARMProcFamily", "AppleA12",
FeatureFuseCryptoEOR,
FeatureStorePairSuppress,
FeatureZCRegMoveGPR64,
- FeatureZCRegMoveFPR64,
FeatureZCZeroing]>;
def TuneAppleA13 : SubtargetFeature<"apple-a13", "ARMProcFamily", "AppleA13",
@@ -371,7 +367,6 @@ def TuneAppleA13 : SubtargetFeature<"apple-a13", "ARMProcFamily", "AppleA13",
FeatureFuseCryptoEOR,
FeatureStorePairSuppress,
FeatureZCRegMoveGPR64,
- FeatureZCRegMoveFPR64,
FeatureZCZeroing]>;
def TuneAppleA14 : SubtargetFeature<"apple-a14", "ARMProcFamily", "AppleA14",
@@ -389,7 +384,6 @@ def TuneAppleA14 : SubtargetFeature<"apple-a14", "ARMProcFamily", "AppleA14",
FeatureFuseLiterals,
FeatureStorePairSuppress,
FeatureZCRegMoveGPR64,
- FeatureZCRegMoveFPR64,
FeatureZCZeroing]>;
def TuneAppleA15 : SubtargetFeature<"apple-a15", "ARMProcFamily", "AppleA15",
@@ -407,7 +401,6 @@ def TuneAppleA15 : SubtargetFeature<"apple-a15", "ARMProcFamily", "AppleA15",
FeatureFuseLiterals,
FeatureStorePairSuppress,
FeatureZCRegMoveGPR64,
- FeatureZCRegMoveFPR64,
FeatureZCZeroing]>;
def TuneAppleA16 : SubtargetFeature<"apple-a16", "ARMProcFamily", "AppleA16",
@@ -425,7 +418,6 @@ def TuneAppleA16 : SubtargetFeature<"apple-a16", "ARMProcFamily", "AppleA16",
FeatureFuseLiterals,
FeatureStorePairSuppress,
FeatureZCRegMoveGPR64,
- FeatureZCRegMoveFPR64,
FeatureZCZeroing]>;
def TuneAppleA17 : SubtargetFeature<"apple-a17", "ARMProcFamily", "AppleA17",
@@ -443,7 +435,6 @@ def TuneAppleA17 : SubtargetFeature<"apple-a17", "ARMProcFamily", "AppleA17",
FeatureFuseLiterals,
FeatureStorePairSuppress,
FeatureZCRegMoveGPR64,
- FeatureZCRegMoveFPR64,
FeatureZCZeroing]>;
def TuneAppleM4 : SubtargetFeature<"apple-m4", "ARMProcFamily", "AppleM4",
@@ -460,7 +451,6 @@ def TuneAppleM4 : SubtargetFeature<"apple-m4", "ARMProcFamily", "AppleM4",
FeatureFuseCryptoEOR,
FeatureFuseLiterals,
FeatureZCRegMoveGPR64,
- FeatureZCRegMoveFPR64,
FeatureZCZeroing
]>;
diff --git a/llvm/test/CodeGen/AArch64/arm64-zero-cycle-regmov-fpr.ll b/llvm/test/CodeGen/AArch64/arm64-zero-cycle-regmov-fpr.ll
deleted file mode 100644
index f422f96f33495..0000000000000
--- a/llvm/test/CodeGen/AArch64/arm64-zero-cycle-regmov-fpr.ll
+++ /dev/null
@@ -1,103 +0,0 @@
-; RUN: llc < %s -mtriple=arm64-linux-gnu | FileCheck %s -check-prefixes=NOTCPU-LINUX --match-full-lines
-; RUN: llc < %s -mtriple=arm64-apple-macosx -mcpu=generic | FileCheck %s -check-prefixes=NOTCPU-APPLE --match-full-lines
-; RUN: llc < %s -mtriple=arm64-apple-macosx -mcpu=apple-m1 | FileCheck %s -check-prefixes=CPU --match-full-lines
-; RUN: llc < %s -mtriple=arm64-apple-macosx -mcpu=apple-m1 -mattr=-zcm-fpr64 | FileCheck %s -check-prefixes=NOTATTR --match-full-lines
-; RUN: llc < %s -mtriple=arm64-apple-macosx -mattr=+zcm-fpr64 | FileCheck %s -check-prefixes=ATTR --match-full-lines
-
-define void @zero_cycle_regmov_FPR32(float %a, float %b, float %c, float %d) {
-entry:
-; CHECK-LABEL: t:
-; NOTCPU-LINUX: fmov s0, s2
-; NOTCPU-LINUX: fmov s1, s3
-; NOTCPU-LINUX: fmov [[REG2:s[0-9]+]], s3
-; NOTCPU-LINUX: fmov [[REG1:s[0-9]+]], s2
-; NOTCPU-LINUX-NEXT: bl {{_?foo_float}}
-; NOTCPU-LINUX: fmov s0, [[REG1]]
-; NOTCPU-LINUX: fmov s1, [[REG2]]
-
-; NOTCPU-APPLE: fmov s0, s2
-; NOTCPU-APPLE: fmov s1, s3
-; NOTCPU-APPLE: fmov [[REG2:s[0-9]+]], s3
-; NOTCPU-APPLE: fmov [[REG1:s[0-9]+]], s2
-; NOTCPU-APPLE-NEXT: bl {{_?foo_float}}
-; NOTCPU-APPLE: fmov s0, [[REG1]]
-; NOTCPU-APPLE: fmov s1, [[REG2]]
-
-; CPU: fmov [[REG2:d[0-9]+]], d3
-; CPU: fmov [[REG1:d[0-9]+]], d2
-; CPU: fmov d0, d2
-; CPU: fmov d1, d3
-; CPU-NEXT: bl {{_?foo_float}}
-; CPU: fmov d0, [[REG1]]
-; CPU: fmov d1, [[REG2]]
-
-; NOTATTR: fmov [[REG2:s[0-9]+]], s3
-; NOTATTR: fmov [[REG1:s[0-9]+]], s2
-; NOTATTR: fmov s0, s2
-; NOTATTR: fmov s1, s3
-; NOTATTR-NEXT: bl {{_?foo_float}}
-; NOTATTR: fmov s0, [[REG1]]
-; NOTATTR: fmov s1, [[REG2]]
-
-; ATTR: fmov d0, d2
-; ATTR: fmov d1, d3
-; ATTR: fmov [[REG2:d[0-9]+]], d3
-; ATTR: fmov [[REG1:d[0-9]+]], d2
-; ATTR-NEXT: bl {{_?foo_float}}
-; ATTR: fmov d0, [[REG1]]
-; ATTR: fmov d1, [[REG2]]
- %call = call float @foo_float(float %c, float %d)
- %call1 = call float @foo_float(float %c, float %d)
- unreachable
-}
-
-declare float @foo_float(float, float)
-
-define void @zero_cycle_regmov_FPR16(half %a, half %b, half %c, half %d) {
-entry:
-; CHECK-LABEL: t:
-; NOTCPU-LINUX: fmov s0, s2
-; NOTCPU-LINUX: fmov s1, s3
-; NOTCPU-LINUX: fmov [[REG2:s[0-9]+]], s3
-; NOTCPU-LINUX: fmov [[REG1:s[0-9]+]], s2
-; NOTCPU-LINUX-NEXT: bl {{_?foo_half}}
-; NOTCPU-LINUX: fmov s0, [[REG1]]
-; NOTCPU-LINUX: fmov s1, [[REG2]]
-
-; NOTCPU-APPLE: fmov s0, s2
-; NOTCPU-APPLE: fmov s1, s3
-; NOTCPU-APPLE: fmov [[REG2:s[0-9]+]], s3
-; NOTCPU-APPLE: fmov [[REG1:s[0-9]+]], s2
-; NOTCPU-APPLE-NEXT: bl {{_?foo_half}}
-; NOTCPU-APPLE: fmov s0, [[REG1]]
-; NOTCPU-APPLE: fmov s1, [[REG2]]
-
-; CPU: fmov [[REG2:d[0-9]+]], d3
-; CPU: fmov [[REG1:d[0-9]+]], d2
-; CPU: fmov d0, d2
-; CPU: fmov d1, d3
-; CPU-NEXT: bl {{_?foo_half}}
-; CPU: fmov d0, [[REG1]]
-; CPU: fmov d1, [[REG2]]
-
-; NOTATTR: fmov [[REG2:s[0-9]+]], s3
-; NOTATTR: fmov [[REG1:s[0-9]+]], s2
-; NOTATTR: fmov s0, s2
-; NOTATTR: fmov s1, s3
-; NOTATTR-NEXT: bl {{_?foo_half}}
-; NOTATTR: fmov s0, [[REG1]]
-; NOTATTR: fmov s1, [[REG2]]
-
-; ATTR: fmov d0, d2
-; ATTR: fmov d1, d3
-; ATTR: fmov [[REG2:d[0-9]+]], d3
-; ATTR: fmov [[REG1:d[0-9]+]], d2
-; ATTR-NEXT: bl {{_?foo_half}}
-; ATTR: fmov d0, [[REG1]]
-; ATTR: fmov d1, [[REG2]]
- %call = call half @foo_half(half %c, half %d)
- %call1 = call half @foo_half(half %c, half %d)
- unreachable
-}
-
-declare half @foo_half(half, half)
|
|
No opinion on the change itself but your PR description does need more detail. A link to the change that added it would be good too. |
tomershafir
force-pushed
the
aarch64-remove-wrong-processor-feature
branch
from
30f404e to
31d6c6f
Compare
|
@DavidSpickett Done |
|
Ok, better, but why is the feature wrong? If I do a downstream merge and see that this change made my benchmarks go one way or the other, that's what I'd be looking for in the commit message. Perhaps the intent was good you just want to implement it differently, perhaps you found it made zero difference and doesn't need to be there, whatever the reason is. |
|
Done @jroelofs Ping |
davemgreen
reviewed
Aug 10, 2025
| @@ -1,103 +0,0 @@ | |||
| ; RUN: llc < %s -mtriple=arm64-linux-gnu | FileCheck %s -check-prefixes=NOTCPU-LINUX --match-full-lines | |||
There was a problem hiding this comment.
Removing this loses all testing for the feature. Can we keep just the testing for default and +zcm-fpr64?
There was a problem hiding this comment.
Ok, gonna keep NOTCPU-APPLE, NOTCPU-LINUX, ATTR
`fmov dX, dY` is not a preferred instruction. Previously introduced by: llvm#144152
tomershafir
force-pushed
the
aarch64-remove-wrong-processor-feature
branch
from
31d6c6f to
ae5084a
Compare
I agree with David here, even after the edits. The description in the commit message needs to say "why" we are making the change, as in: why is this feature wrong and not preferred? That's something that often cannot be inferred from the "what" of the rest of the patch. |
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.
fmov dX, dYis not a preferred instruction.Previously introduced by: #144152