Always honor fp-contract pragmas on PlayStation #162549

wjristow · 2025-10-08T20:57:10Z

The pragma:
#pragma clang fp contract (off)

can be used to disable FMA, but when cross-statement FMA is enabled in fast fp contract mode (via -ffast-math, for example), the effect of that pragma is suppressed.

To have Clang honor that pragma in 'fast' fp contract mode in C/C++, an additional switch:
-ffp-contract=fast-honor-pragmas

must be applied. On PlayStation, we want to always honor that pragma, without requiring the additional switch to be specified. This commit does that.

llvmbot · 2025-10-08T20:57:41Z

@llvm/pr-subscribers-clang-codegen
@llvm/pr-subscribers-clang

@llvm/pr-subscribers-backend-x86

Author: None (wjristow)

Changes

The pragma:
#pragma clang fp contract (off)

can be used to disable FMA, but when cross-statement FMA is enabled in fast fp contract mode (via -ffast-math, for example), the effect of that pragma is suppressed.

To have Clang honor that pragma in 'fast' fp contract mode in C/C++, an additional switch:
-ffp-contract=fast-honor-pragmas

must be applied. On PlayStation, we want to always honor that pragma, without requiring the additional switch to be specified. This commit does that.

Full diff: https://github.com/llvm/llvm-project/pull/162549.diff

2 Files Affected:

(modified) clang/lib/CodeGen/BackendUtil.cpp (+5-1)
(added) clang/test/CodeGen/X86/fma-fast-pragma.cpp (+53)

diff --git a/clang/lib/CodeGen/BackendUtil.cpp b/clang/lib/CodeGen/BackendUtil.cpp
index 64f1917739e12..52bf702cdadb0 100644
--- a/clang/lib/CodeGen/BackendUtil.cpp
+++ b/clang/lib/CodeGen/BackendUtil.cpp
@@ -396,7 +396,11 @@ static bool initTargetOptions(const CompilerInstance &CI,
     Options.AllowFPOpFusion = llvm::FPOpFusion::Standard;
     break;
   case LangOptions::FPM_Fast:
-    Options.AllowFPOpFusion = llvm::FPOpFusion::Fast;
+    // We always honor fp-contract pragmas for PlayStation.
+    if (CI.getASTContext().getTargetInfo().getTriple().isPS())
+      Options.AllowFPOpFusion = llvm::FPOpFusion::Standard;
+    else
+      Options.AllowFPOpFusion = llvm::FPOpFusion::Fast;
     break;
   }
 
diff --git a/clang/test/CodeGen/X86/fma-fast-pragma.cpp b/clang/test/CodeGen/X86/fma-fast-pragma.cpp
new file mode 100644
index 0000000000000..9d38e245b5f64
--- /dev/null
+++ b/clang/test/CodeGen/X86/fma-fast-pragma.cpp
@@ -0,0 +1,53 @@
+// REQUIRES: x86-registered-target
+
+// With the pragma in place, generic targets leave FMA enabled unless the
+// switch '-ffp-contract=fast-honor-pragmas' is used to disable it; whereas
+// for PlayStation, the pragma is always honored, so FMA is disabled even in
+// plain 'fast' mode:
+// RUN: %clang_cc1 -S -triple x86_64-unknown-unknown -target-feature +fma \
+// RUN:   -O2 -ffp-contract=fast -o - %s | \
+// RUN:   FileCheck --check-prefix=CHECK-YES-FMA %s
+// RUN: %clang_cc1 -S -triple x86_64-unknown-unknown -target-feature +fma \
+// RUN:   -O2 -ffp-contract=fast-honor-pragmas -o - %s | \
+// RUN:   FileCheck --check-prefix=CHECK-NO-FMA %s
+// RUN: %clang_cc1 -S -triple x86_64-unknown-unknown -target-feature +fma \
+// RUN:   -O2 -ffp-contract=fast -ffp-contract=fast-honor-pragmas -o - %s | \
+// RUN:   FileCheck --check-prefix=CHECK-NO-FMA %s
+// RUN: %clang_cc1 -S -triple x86_64-sie-ps5 -target-feature +fma \
+// RUN:   -O2 -ffp-contract=fast -o - %s | \
+// RUN:   FileCheck --check-prefix=CHECK-NO-FMA %s
+// RUN: %clang_cc1 -S -triple x86_64-sie-ps5 -target-feature +fma \
+// RUN:   -O2 -ffp-contract=fast-honor-pragmas -o - %s | \
+// RUN:   FileCheck --check-prefix=CHECK-NO-FMA %s
+//
+// With the pragma suppressed, FMA happens in 'fast' or 'fast-honor-pragmas'
+// modes (for generic targets and for PlayStation):
+// RUN: %clang_cc1 -S -DSUPPRESS_PRAGMA \
+// RUN:   -triple x86_64-unknown-unknown -target-feature +fma \
+// RUN:   -O2 -ffp-contract=fast -o - %s | \
+// RUN:   FileCheck --check-prefix=CHECK-YES-FMA %s
+// RUN: %clang_cc1 -S -DSUPPRESS_PRAGMA \
+// RUN:   -triple x86_64-unknown-unknown -target-feature +fma \
+// RUN:   -O2 -ffp-contract=fast-honor-pragmas -o - %s | \
+// RUN:   FileCheck --check-prefix=CHECK-YES-FMA %s
+// RUN: %clang_cc1 -S -DSUPPRESS_PRAGMA \
+// RUN:   -triple x86_64-sie-ps5 -target-feature +fma \
+// RUN:   -O2 -ffp-contract=fast -o - %s | \
+// RUN:   FileCheck --check-prefix=CHECK-YES-FMA %s
+// RUN: %clang_cc1 -S -DSUPPRESS_PRAGMA \
+// RUN:   -triple x86_64-sie-ps5 -target-feature +fma \
+// RUN:   -O2 -ffp-contract=fast-honor-pragmas -o - %s | \
+// RUN:   FileCheck --check-prefix=CHECK-YES-FMA %s
+//
+float compute(float a, float b, float c) {
+#if !defined(SUPPRESS_PRAGMA)
+#pragma clang fp contract (off)
+#endif
+  float product = a * b;
+  return product + c;
+}
+
+// CHECK-NO-FMA: vmulss
+// CHECK-NO-FMA-NEXT: vaddss
+
+// CHECK-YES-FMA: vfmadd213ss

wjristow · 2025-10-08T21:07:39Z

The test-case of this patch generates X86 assembly code, rather than the usual approach of generating generic IR. The reason I did that is because the impact of the change isn't visible in the IR. That's due to a bug, which I've just filed: #162550

If this patch is accepted, we will want to update the documentation of -ffp-contract=<value>:
https://clang.llvm.org/docs/UsersManual.html#cmdoption-ffp-contract

For reference, here are some points of discussion about the -ffp-contract=fast-honor-pragmas concept:
https://discourse.llvm.org/t/fp-contract-fast-and-pragmas/58529
https://discourse.llvm.org/t/rfc-honor-pragmas-with-ffp-contract-fast/80797/14

andykaylor · 2025-10-08T22:41:15Z

I continue to believe that ignoring pragmas when fast-math is enabled is a bug, and I'd like to see a more general change such as I proposed in https://github.com/llvm/llvm-project/pull/105746.

The pragma: #pragma clang fp contract (off) can be used to disable FMA, but when cross-statement FMA is enabled in `fast` fp contract mode (via `-ffast-math`, for example), the effect of that pragma is suppressed. To have Clang honor that pragma in 'fast' fp contract mode in C/C++, an additional switch: -ffp-contract=fast-honor-pragmas must be applied. On PlayStation, we want to always honor that pragma, without requiring the additional switch to be specified. This commit does that.

wjristow · 2025-10-08T23:31:53Z

I continue to believe that ignoring pragmas when fast-math is enabled is a bug, and I'd like to see a more general change such as I proposed in https://github.com/llvm/llvm-project/pull/105746.

FWIW, that makes good sense to me. I admit I only looked at that PR of yours carefully today, as I was preparing this one.

efriedma-quic · 2025-10-09T00:42:33Z

Maybe time to revisit #105746; the situation in the backends looks a lot better. Looks like Arm and Mips still have some problematic isel patterns; everything else looks okay at first glance?

We could blacklist specific backends which still have issues.

wjristow · 2025-10-09T03:23:38Z

Maybe time to revisit #105746; the situation in the backends looks a lot better. Looks like Arm and Mips still have some problematic isel patterns; everything else looks okay at first glance?

We could blacklist specific backends which still have issues.

That sounds good to me. Like Andy, I also view the current behavior as a bug. And I view this PR of mine as "fix that bug for PlayStation", whereas I view #105746 as "fix that bug, for all". So I'd be happy to have that done.

And as an aside, if we do Andy's full fix, then the bug I reported as related to this (#162550) becomes moot. That bug is essentially: In LTO mode the pragma is always honored. Or more directly, the switch -ffp-contract=fast-honor-pragmas is completely ineffective in LTO mode; so if that feature is removed, there is nothing there to be fixed.

wjristow · 2025-10-20T18:09:27Z

Given @andykaylor's comment:

I continue to believe that ignoring pragmas when fast-math is enabled is a bug, and I'd like to see a more general change such as I proposed in https://github.com/llvm/llvm-project/pull/105746.

and @efriedma-quic's comment:

Maybe time to revisit #105746; the situation in the backends looks a lot better. Looks like Arm and Mips still have some problematic isel patterns; everything else looks okay at first glance?

We could blacklist specific backends which still have issues.

I think revisiting that is the right thing to do. That said, if complications on some specific backends make that difficult, we could move forward with this PR, fixing this on PlayStation (in our downstream version, we're already doing that).

Regarding blacklisting specific backends, if we were to do this PR, we could also do the dual of that blacklist approach: That is, others that want the pragmas to always be honored could opt-in to the approach proposed here in this PR. I've just added a note in Andy's PR #105746, suggesting these two possibilities.

ojhunt · 2025-11-13T22:58:35Z

clang/lib/CodeGen/BackendUtil.cpp

    break;
  case LangOptions::FPM_Fast:
-    Options.AllowFPOpFusion = llvm::FPOpFusion::Fast;
+    // We always honor fp-contract pragmas for PlayStation.


I'm not sure if we should simply do this silently (which may be surprising) or if we should at least issue a warning that we are doing this? Perhaps with a deprecation notice in the vague hope that we could turn it into an error in future?

As it happens, for PlayStation, we have had a private patch in place to honor the fp-contract pragma since our llvm11-based release (this was before the concept of -ffp-contract=fast-honor-pragmas was created). So this isn't a change in behavior for PlayStation developers, and so no need for a warning, from that perspective.

That said, if #105746 is done in some form (and there is activity there now), this PR can be abandoned. And as I've said above, I'm very much in favor of that approach being taken.

llvmbot added clang Clang issues not falling into any other category backend:X86 clang:codegen IR generation bugs: mangling, exceptions, etc. labels Oct 8, 2025

wjristow requested review from fhahn, ojhunt and tarunprabhu October 8, 2025 20:57

wjristow requested a review from andykaylor October 8, 2025 20:57

wjristow mentioned this pull request Oct 8, 2025

A pragma incorrectly suppresses FMA in the IR without -ffp-contract=fast-honor-pragmas #162550

Open

andykaylor requested review from arsenm, efriedma-quic and jcranmer-intel October 8, 2025 22:33

andykaylor requested a review from MaskRay October 8, 2025 22:41

wjristow force-pushed the ps-honor-fma-pragmas branch from 9e1b410 to 123ddd9 Compare October 8, 2025 23:28

arsenm requested a review from yxsamliu October 10, 2025 03:49

wjristow mentioned this pull request Oct 20, 2025

Honor pragmas with -ffp-contract=fast, deprecate fast-honor-pragmas #105746

Open

ojhunt reviewed Nov 13, 2025

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Always honor fp-contract pragmas on PlayStation #162549

Always honor fp-contract pragmas on PlayStation #162549

Uh oh!

wjristow commented Oct 8, 2025

Uh oh!

llvmbot commented Oct 8, 2025 •

edited

Loading

Uh oh!

wjristow commented Oct 8, 2025

Uh oh!

andykaylor commented Oct 8, 2025

Uh oh!

wjristow commented Oct 8, 2025

Uh oh!

efriedma-quic commented Oct 9, 2025

Uh oh!

wjristow commented Oct 9, 2025 •

edited

Loading

Uh oh!

wjristow commented Oct 20, 2025

Uh oh!

ojhunt Nov 13, 2025

Uh oh!

wjristow Nov 14, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

Always honor fp-contract pragmas on PlayStation #162549

Are you sure you want to change the base?

Always honor fp-contract pragmas on PlayStation #162549

Uh oh!

Conversation

wjristow commented Oct 8, 2025

Uh oh!

llvmbot commented Oct 8, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

wjristow commented Oct 8, 2025

Uh oh!

andykaylor commented Oct 8, 2025

Uh oh!

wjristow commented Oct 8, 2025

Uh oh!

efriedma-quic commented Oct 9, 2025

Uh oh!

wjristow commented Oct 9, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

wjristow commented Oct 20, 2025

Uh oh!

ojhunt Nov 13, 2025

Choose a reason for hiding this comment

Uh oh!

wjristow Nov 14, 2025

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

llvmbot commented Oct 8, 2025 •

edited

Loading

wjristow commented Oct 9, 2025 •

edited

Loading