ppc64(le): Add an option to use IEEE long double ABI on Linux #4833

liushuyu · 2025-02-01T23:48:29Z

This pull request adds an option to use IEEE long double ABI on Linux (if the host environment supports it).

Some adjustments are made to accommodate the new ABI (glibc uses dual-ABI in this case, where new IEEE long double-capable functions are suffixed with __ieee128; older functions using IBM long double are also kept for compatibility reasons).

... the system is using the new ABI that supports IEEE 754R long double instead of the legacy IBM double double format

... and then add a long double rewrite to convert it to ppc_f128 when lowering

... if -mabi=ieeelongdouble is specified on ppc64

... if IEEE long double ABI is selected

... to avoid a LLVM bug on multiple platforms

liushuyu · 2025-02-01T23:49:17Z

During the porting process, I discovered what seems to be an LLVM bug when targeting ppc64 using the IEEE long double ABI. Consider the following D code:

extern (C++) real test1(real arg0)
{
  if (!(arg0 == 0.0))
  {
    return 0.0;
  }
  else
  {
    return 0.0;
  }
}

LDC will generate the following LLVM IR (reduced, no optimization):

  target datalayout = "e-m:e-Fn32-i64:64-i128:128-n32:64-S128-v256:256:256-v512:512:512"
  target triple = "powerpc64le-unknown-linux-gnu"

  define fp128 @_Z5test1u9__ieee128(fp128 %0) {
    %2 = fcmp ogt fp128 %0, 0xL00000000000000000000000000000000
    %3 = icmp i1 %2, false
    br i1 %3, label %5, label %4

  4:                                                ; preds = %1
    ret fp128 0xL00000000000000000000000000000000

  5:                                                ; preds = %1
    ret fp128 0xL00000000000000000000000000000000
  }

... which will lead to what seems to be an infinite ppc-isel expansion loop inside LLVM.

I have added a workaround solution in this pull request to replace icmp .., false with xor .., true when lowering a "not" operator.

JohanEngelen · 2025-02-02T18:02:15Z

During the porting process, I discovered what seems to be an LLVM bug when targeting ppc64 using the IEEE long double ABI.

Please also report the bug in LLVM's bug tracker.

Nice that you are working on this btw !

JohanEngelen · 2025-02-02T18:05:36Z

driver/targetmachine.cpp

@@ -108,6 +108,8 @@ const char *getABI(const llvm::Triple &triple, const llvm::SmallVectorImpl<llvm:
        return "elfv1";
      if (ABIName.startswith("elfv2"))
        return "elfv2";
+      if (ABIName.startswith("ieeelongdouble"))


Would it make sense for a user to want to use ieeelongdouble together with elfv1 or 2? (seems logical, but I have no knowledge here)
If so, then this has to be rewritten to accept multiple options in mabi.

It seems like both Clang and GCC do support accepting multiple -mabi= values (only on ppc64 + big-endian systems, as ppc64le only supports elfv2).

However, GCC upstream wiki does mention ieeelongdouble + ppc64 + elfv1 is an unsupported configuration: https://gcc.gnu.org/wiki/Ieee128PowerPC#A7.0_Transition_to_IEEE_128-bit_floating_point_as_the_default_for_long_double

Addressed now, you can use -mabi=elfv1 -mabi=ieeelongdouble to enable IEEE long double on ppc64 (big-endian) systems. I have also added a logic where LDC will warn if this unsupported configuration is encountered.

JohanEngelen · 2025-02-02T18:08:07Z

driver/main.cpp

@@ -683,6 +683,10 @@ void registerPredefinedTargetVersions() {
    if (triple.getOS() == llvm::Triple::Linux) {
      VersionCondition::addPredefinedGlobalIdent(
          triple.getArch() == llvm::Triple::ppc64 ? "ELFv1" : "ELFv2");
+      if (opts::mABI == "ieeelongdouble" && triple.isOSGlibc()) {
+        // Only GLibc needs this for IEEELongDouble
+        VersionCondition::addPredefinedGlobalIdent("D_PPCUseIEEE128");


We should add D_PPCUseIEEE128 as reserved version identifier upstream (dmd). Helps compatibility with GDC too.

Edit: example PR: dlang/dmd#16773

Thanks! However, I am not entirely sure if identifiers starting with D_ are already classified as somewhat reserved.

All identifiers starting with D_ are already reserved for the compiler.

I think it's good for documentation, and then you can sync/check with gdc too

I think maybe we should also forward the phobos changes to the D Language upstream? That way, GDC can also benefit from this fix.

Yep, good idea

Sorry, which Phobos changes? And IMO real.mant_dig is all we need, as used in e.g. https://github.com/dlang/phobos/blob/c5d074b12e1038389401baa8c48c332fac5521e0/std/math/traits.d#L898-L1050

JohanEngelen · 2025-02-02T18:11:02Z

runtime/druntime/src/core/thread/fiber.d

@@ -570,6 +570,8 @@ version (LDC)

    version (AArch64) version = CheckFiberMigration;

+    version (PPC64)   version = CheckFiberMigration;


intended change?

Yes, this is intended change. I fixed this after running the basic unit tests

To add to my last comment, that issue was not discovered because the D Phobos library needs IEEE 128-bit arithmetic to work, which was not implemented before this pull request.

JohanEngelen · 2025-02-02T18:11:28Z

would be good to add a lit testcase for setting this abi param

liushuyu · 2025-02-02T20:02:23Z

would be good to add a lit testcase for setting this abi param

Will do

ABI switching

liushuyu · 2025-02-04T23:45:15Z

During the porting process, I discovered what seems to be an LLVM bug when targeting ppc64 using the IEEE long double ABI.

Please also report the bug in LLVM's bug tracker.

Nice that you are working on this btw !

Thanks! I have proposed a fix to the LLVM upstream regarding this issue: llvm/llvm-project#125776

This will allow the user to specify -mabi=ieeelongdouble + -mabi=elfv1 together without changing how LDC parses mabi options too much

liushuyu · 2025-02-06T01:15:47Z

Corresponding DMD pull request: dlang/dmd#20826

liushuyu added 11 commits January 29, 2025 12:57

cmake: do not turn off long double on ppc64 if ...

ebf89e5

... the system is using the new ABI that supports IEEE 754R long double instead of the legacy IBM double double format

gen: use IEEE128 on ppc64le internally ...

47a5851

... and then add a long double rewrite to convert it to ppc_f128 when lowering

gen: use ppc_f128 or fp128 on ppc64le depending on the ABI switch

d20b1ec

driver: add D_PPCUseIEEE128 pre-defined version ...

b4eec59

... if -mabi=ieeelongdouble is specified on ppc64

druntime: redirect dual-ABI functions on glibc to IEEE128 version ...

9ac212c

... if IEEE long double ABI is selected

gen/modules.cpp: add PowerPC float-abi metadata marking

6517fa3

gen/toir.cpp: use xor operator to emulate unary not operation ...

ec5f54f

... to avoid a LLVM bug on multiple platforms

gen/target.cpp: special C++ mangling for ppc64

c4aa527

gen/abi/ppc64le.cpp: fix zext/sext for vector types

1b6b836

runtime/fiber: enable CheckFiberMigration for ppc64

7892922

runtime/atomic: enable 128BitCAS for ppc64 when using IEEE 128 ABI

9bcee7c

JohanEngelen reviewed Feb 2, 2025

View reviewed changes

liushuyu added 2 commits February 4, 2025 16:40

tests/driver/ppc_float_abi.d: add a test to test ppc64 real type

79d4e43

ABI switching

driver/main + gen/target: fix environment detection for ppc64 real type

45d2676

liushuyu added 3 commits February 4, 2025 16:59

CHANGELOG.md: document ppc64 and ppc64le IEEE 128 support changes

46c7b42

gen+driver: use a different approach to parse and pass IEEE 128 options

32c5aec

This will allow the user to specify -mabi=ieeelongdouble + -mabi=elfv1 together without changing how LDC parses mabi options too much

driver/main.cpp: fix elfv1 and elfv2 version identifier definition logic

2361ed6

liushuyu force-pushed the ppc64-d-ieee754-fix-new branch from 9b8fd56 to 2361ed6 Compare February 5, 2025 21:55

liushuyu mentioned this pull request Feb 6, 2025

druntime: redirect dual-ABI functions on glibc to IEEE128 version dlang/dmd#20826

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

ppc64(le): Add an option to use IEEE long double ABI on Linux #4833

ppc64(le): Add an option to use IEEE long double ABI on Linux #4833

liushuyu commented Feb 1, 2025

liushuyu commented Feb 1, 2025 •

edited

Loading

JohanEngelen commented Feb 2, 2025

JohanEngelen Feb 2, 2025

liushuyu Feb 4, 2025 •

edited

Loading

liushuyu Feb 5, 2025

JohanEngelen Feb 2, 2025 •

edited

Loading

liushuyu Feb 2, 2025 •

edited

Loading

thewilsonator Feb 2, 2025

JohanEngelen Feb 2, 2025

liushuyu Feb 4, 2025

JohanEngelen Feb 4, 2025

kinke Feb 6, 2025

JohanEngelen Feb 2, 2025

liushuyu Feb 2, 2025

liushuyu Feb 4, 2025

JohanEngelen commented Feb 2, 2025

liushuyu commented Feb 2, 2025

liushuyu commented Feb 4, 2025

liushuyu commented Feb 6, 2025

		@@ -570,6 +570,8 @@ version (LDC)

		version (AArch64) version = CheckFiberMigration;

		version (PPC64) version = CheckFiberMigration;

ppc64(le): Add an option to use IEEE long double ABI on Linux #4833

Are you sure you want to change the base?

ppc64(le): Add an option to use IEEE long double ABI on Linux #4833

Conversation

liushuyu commented Feb 1, 2025

liushuyu commented Feb 1, 2025 • edited Loading

JohanEngelen commented Feb 2, 2025

Choose a reason for hiding this comment

liushuyu Feb 4, 2025 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

JohanEngelen Feb 2, 2025 • edited Loading

Choose a reason for hiding this comment

liushuyu Feb 2, 2025 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

JohanEngelen commented Feb 2, 2025

liushuyu commented Feb 2, 2025

liushuyu commented Feb 4, 2025

liushuyu commented Feb 6, 2025

liushuyu commented Feb 1, 2025 •

edited

Loading

liushuyu Feb 4, 2025 •

edited

Loading

JohanEngelen Feb 2, 2025 •

edited

Loading

liushuyu Feb 2, 2025 •

edited

Loading