[flang] IEEE underflow control for Arm #124170
Merged
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
Update IEEE_SUPPORT_UNDERFLOW_CONTROL, IEEE_GET_UNDERFLOW_MODE, and IEEE_SET_UNDERFLOW_MODE code for Arm.
llvmbot
added
flang:runtime
flang
|
@llvm/pr-subscribers-flang-runtime Author: None (vdonaldson) ChangesUpdate IEEE_SUPPORT_UNDERFLOW_CONTROL, IEEE_GET_UNDERFLOW_MODE, and IEEE_SET_UNDERFLOW_MODE code for Arm. Full diff: https://github.com/llvm/llvm-project/pull/124170.diff 2 Files Affected:
diff --git a/flang/include/flang/Tools/TargetSetup.h b/flang/include/flang/Tools/TargetSetup.h
index d1b0da3a42c897..5d23df6823a948 100644
--- a/flang/include/flang/Tools/TargetSetup.h
+++ b/flang/include/flang/Tools/TargetSetup.h
@@ -24,34 +24,35 @@ namespace Fortran::tools {
const std::string &compilerVersion, const std::string &compilerOptions) {
const llvm::Triple &targetTriple{targetMachine.getTargetTriple()};
- // FIXME: Handle real(3) ?
- if (targetTriple.getArch() != llvm::Triple::ArchType::x86_64) {
- targetCharacteristics.DisableType(
- Fortran::common::TypeCategory::Real, /*kind=*/10);
- }
+
+ targetCharacteristics.set_ieeeFeature(evaluate::IeeeFeature::Halting, true);
+
if (targetTriple.getArch() == llvm::Triple::ArchType::x86_64) {
targetCharacteristics.set_hasSubnormalFlushingControl(/*kind=*/3);
targetCharacteristics.set_hasSubnormalFlushingControl(/*kind=*/4);
targetCharacteristics.set_hasSubnormalFlushingControl(/*kind=*/8);
}
+
if (targetTriple.isARM() || targetTriple.isAArch64()) {
targetCharacteristics.set_haltingSupportIsUnknownAtCompileTime();
targetCharacteristics.set_ieeeFeature(
evaluate::IeeeFeature::Halting, false);
- } else {
- targetCharacteristics.set_ieeeFeature(evaluate::IeeeFeature::Halting);
+ targetCharacteristics.set_hasSubnormalFlushingControl(/*kind=*/3);
+ targetCharacteristics.set_hasSubnormalFlushingControl(/*kind=*/4);
+ targetCharacteristics.set_hasSubnormalFlushingControl(/*kind=*/8);
+ }
+
+ if (targetTriple.getArch() != llvm::Triple::ArchType::x86_64) {
+ targetCharacteristics.DisableType(
+ Fortran::common::TypeCategory::Real, /*kind=*/10);
}
- // Figure out if we can support F128: see
- // flang/runtime/Float128Math/math-entries.h
- // TODO: this should be taken from TargetInfo::getLongDoubleFormat to support
- // cross-compilation
+ // Check for kind=16 support. See flang/runtime/Float128Math/math-entries.h.
+ // TODO: Take this from TargetInfo::getLongDoubleFormat for cross compilation.
#ifdef FLANG_RUNTIME_F128_MATH_LIB
- // we can use libquadmath wrappers
- constexpr bool f128Support = true;
+ constexpr bool f128Support = true; // use libquadmath wrappers
#elif HAS_LDBL128
- // we can use libm wrappers
- constexpr bool f128Support = true;
+ constexpr bool f128Support = true; // use libm wrappers
#else
constexpr bool f128Support = false;
#endif
diff --git a/flang/runtime/exceptions.cpp b/flang/runtime/exceptions.cpp
index f541b8e844aded..7fca0c431f8cd0 100644
--- a/flang/runtime/exceptions.cpp
+++ b/flang/runtime/exceptions.cpp
@@ -11,7 +11,9 @@
#include "flang/Runtime/exceptions.h"
#include "terminator.h"
#include <cfenv>
-#if __x86_64__
+#if __aarch64__
+#include <fpu_control.h>
+#elif __x86_64__
#include <xmmintrin.h>
#endif
@@ -90,20 +92,40 @@ bool RTNAME(SupportHalting)([[maybe_unused]] uint32_t except) {
#endif
}
+// A hardware FZ (flush to zero) bit is the negation of the
+// ieee_[get|set]_underflow_mode GRADUAL argument.
+#if defined(_MM_FLUSH_ZERO_MASK)
+// The MXCSR FZ bit affects computations of real kinds 3, 4, and 8.
+#elif defined(_FPU_GETCW)
+// The FPCR FZ bit affects computations of real kinds 3, 4, and 8.
+// bit 24: FZ -- single, double precision flush to zero bit
+// bit 19: FZ16 -- half precision flush to zero bit [not currently relevant]
+#define _FPU_FPCR_FZ_MASK_ 0x01080000
+#endif
+
bool RTNAME(GetUnderflowMode)(void) {
-#if _MM_FLUSH_ZERO_MASK
- // The MXCSR Flush to Zero flag is the negation of the ieee_get_underflow_mode
- // GRADUAL argument. It affects real computations of kinds 3, 4, and 8.
+#if defined(_MM_FLUSH_ZERO_MASK)
return _MM_GET_FLUSH_ZERO_MODE() == _MM_FLUSH_ZERO_OFF;
+#elif defined(_FPU_GETCW)
+ uint32_t fpcr;
+ _FPU_GETCW(fpcr);
+ return (fpcr & _FPU_FPCR_FZ_MASK_) != _FPU_FPCR_FZ_MASK_;
#else
return false;
#endif
}
void RTNAME(SetUnderflowMode)(bool flag) {
-#if _MM_FLUSH_ZERO_MASK
- // The MXCSR Flush to Zero flag is the negation of the ieee_set_underflow_mode
- // GRADUAL argument. It affects real computations of kinds 3, 4, and 8.
+#if defined(_MM_FLUSH_ZERO_MASK)
_MM_SET_FLUSH_ZERO_MODE(flag ? _MM_FLUSH_ZERO_OFF : _MM_FLUSH_ZERO_ON);
+#elif defined(_FPU_GETCW)
+ uint32_t fpcr;
+ _FPU_GETCW(fpcr);
+ if (flag) {
+ fpcr &= ~_FPU_FPCR_FZ_MASK_;
+ } else {
+ fpcr |= _FPU_FPCR_FZ_MASK_;
+ }
+ _FPU_SETCW(fpcr);
#endif
}
|
jeanPerier
reviewed
Jan 24, 2025
There was a problem hiding this comment.
Not the expert here, but code looks good to me.
One question looking at ARM doc for FPCR.FZ, does FPCR.AH also needs to be set to 0 or 1 to get the expected IEEE behavior (it looks like it impacts the flushing of fp operation inputs in the FZ doc)?
FPCR.AH seems to also impacts the rounding (RMode), so maybe modifying it here is a bad idea.
Thanks @jeanPerier for looking into this. My reading is that the AH bit affects whether denormal operands are treated as zero or not, which is distinct from whether or not denormal results are flushed to zero. My interpretation of the Fortran standard is that these ieee functions apply to results, not operands, and the AH bit is not relevant here. If at some point we implement (standard-external) compilation options to control denormal operand handling, that would involve the AH bit. |
jeanPerier
approved these changes
Jan 27, 2025
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/89/builds/15262 Here is the relevant piece of the build log for the reference |
vdonaldson
added a commit
that referenced
this pull request
Jan 27, 2025
This reverts commit 3684ec4.
vdonaldson
added a commit
that referenced
this pull request
Jan 27, 2025
github-actions bot
pushed a commit
to arm/arm-toolchain
that referenced
this pull request
Jan 27, 2025
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.
Update IEEE_SUPPORT_UNDERFLOW_CONTROL, IEEE_GET_UNDERFLOW_MODE, and IEEE_SET_UNDERFLOW_MODE code for Arm.