PowerPC: Fix using long double libm functions for f128 intrinsics #144382

arsenm · 2025-06-16T16:11:34Z

This wasn't setting the correct libcall names, which default to the
l suffixed libm names.

arsenm · 2025-06-16T16:11:47Z

This stack of pull requests is managed by Graphite. Learn more about stacking.

llvmbot · 2025-06-16T16:14:13Z

@llvm/pr-subscribers-llvm-ir

@llvm/pr-subscribers-backend-powerpc

Author: Matt Arsenault (arsenm)

Changes

This wasn't setting the correct libcall names, which default to the
l suffixed libm names.

Full diff: https://github.com/llvm/llvm-project/pull/144382.diff

2 Files Affected:

(modified) llvm/lib/IR/RuntimeLibcalls.cpp (+25)
(modified) llvm/test/CodeGen/PowerPC/f128-arith.ll (+24-24)

diff --git a/llvm/lib/IR/RuntimeLibcalls.cpp b/llvm/lib/IR/RuntimeLibcalls.cpp
index d63d398e243f9..51d7738151237 100644
--- a/llvm/lib/IR/RuntimeLibcalls.cpp
+++ b/llvm/lib/IR/RuntimeLibcalls.cpp
@@ -321,6 +321,24 @@ void RuntimeLibcallsInfo::initLibcalls(const Triple &TT) {
     setLibcallName(RTLIB::OGT_F128, "__gtkf2");
     setLibcallName(RTLIB::UO_F128, "__unordkf2");
 
+    setLibcallName(RTLIB::ACOS_F128, "acosf128");
+    setLibcallName(RTLIB::ASIN_F128, "asinf128");
+    setLibcallName(RTLIB::ATAN2_F128, "atan2f128");
+    setLibcallName(RTLIB::ATAN_F128, "atanf128");
+    setLibcallName(RTLIB::CBRT_F128, "cbrtf128");
+    setLibcallName(RTLIB::COPYSIGN_F128, "copysignf128");
+    setLibcallName(RTLIB::COSH_F128, "coshf128");
+    setLibcallName(RTLIB::EXP10_F128, "exp10f128");
+    setLibcallName(RTLIB::FMAXIMUM_F128, "fmaximumf128");
+    setLibcallName(RTLIB::FMAXIMUM_NUM_F128, "fmaximum_numf128");
+    setLibcallName(RTLIB::FMINIMUM_F128, "fminimumf128");
+    setLibcallName(RTLIB::FMINIMUM_NUM_F128, "fminimum_numf128");
+    setLibcallName(RTLIB::LDEXP_F128, "ldexpf128");
+    setLibcallName(RTLIB::MODF_F128, "modff128");
+    setLibcallName(RTLIB::ROUNDEVEN_F128, "roundevenf128");
+    setLibcallName(RTLIB::SINH_F128, "sinhf128");
+    setLibcallName(RTLIB::TANH_F128, "tanhf128");
+    setLibcallName(RTLIB::TAN_F128, "tanf128");
     setLibcallName(RTLIB::LOG_F128, "logf128");
     setLibcallName(RTLIB::LOG2_F128, "log2f128");
     setLibcallName(RTLIB::LOG10_F128, "log10f128");
@@ -347,6 +365,13 @@ void RuntimeLibcallsInfo::initLibcalls(const Triple &TT) {
     setLibcallName(RTLIB::FMA_F128, "fmaf128");
     setLibcallName(RTLIB::FREXP_F128, "frexpf128");
 
+    // TODO: Do these exist?
+    setLibcallName(RTLIB::EXP_FINITE_F128, nullptr);
+    setLibcallName(RTLIB::EXP2_FINITE_F128, nullptr);
+    setLibcallName(RTLIB::LOG_FINITE_F128, nullptr);
+    setLibcallName(RTLIB::LOG2_FINITE_F128, nullptr);
+    setLibcallName(RTLIB::LOG10_FINITE_F128, nullptr);
+
     if (TT.isOSAIX()) {
       bool isPPC64 = TT.isPPC64();
       setLibcallName(RTLIB::MEMCPY, isPPC64 ? "___memmove64" : "___memmove");
diff --git a/llvm/test/CodeGen/PowerPC/f128-arith.ll b/llvm/test/CodeGen/PowerPC/f128-arith.ll
index ffa7ac6cb0078..f9c953d483ff2 100644
--- a/llvm/test/CodeGen/PowerPC/f128-arith.ll
+++ b/llvm/test/CodeGen/PowerPC/f128-arith.ll
@@ -1413,7 +1413,7 @@ define dso_local fp128 @acos_f128(fp128 %x) {
 ; CHECK-NEXT:    std r0, 48(r1)
 ; CHECK-NEXT:    .cfi_def_cfa_offset 32
 ; CHECK-NEXT:    .cfi_offset lr, 16
-; CHECK-NEXT:    bl acosl
+; CHECK-NEXT:    bl acosf128
 ; CHECK-NEXT:    nop
 ; CHECK-NEXT:    addi r1, r1, 32
 ; CHECK-NEXT:    ld r0, 16(r1)
@@ -1427,7 +1427,7 @@ define dso_local fp128 @acos_f128(fp128 %x) {
 ; CHECK-P8-NEXT:    std r0, 48(r1)
 ; CHECK-P8-NEXT:    .cfi_def_cfa_offset 32
 ; CHECK-P8-NEXT:    .cfi_offset lr, 16
-; CHECK-P8-NEXT:    bl acosl
+; CHECK-P8-NEXT:    bl acosf128
 ; CHECK-P8-NEXT:    nop
 ; CHECK-P8-NEXT:    addi r1, r1, 32
 ; CHECK-P8-NEXT:    ld r0, 16(r1)
@@ -1445,7 +1445,7 @@ define dso_local fp128 @asin_f128(fp128 %x) {
 ; CHECK-NEXT:    std r0, 48(r1)
 ; CHECK-NEXT:    .cfi_def_cfa_offset 32
 ; CHECK-NEXT:    .cfi_offset lr, 16
-; CHECK-NEXT:    bl asinl
+; CHECK-NEXT:    bl asinf128
 ; CHECK-NEXT:    nop
 ; CHECK-NEXT:    addi r1, r1, 32
 ; CHECK-NEXT:    ld r0, 16(r1)
@@ -1459,7 +1459,7 @@ define dso_local fp128 @asin_f128(fp128 %x) {
 ; CHECK-P8-NEXT:    std r0, 48(r1)
 ; CHECK-P8-NEXT:    .cfi_def_cfa_offset 32
 ; CHECK-P8-NEXT:    .cfi_offset lr, 16
-; CHECK-P8-NEXT:    bl asinl
+; CHECK-P8-NEXT:    bl asinf128
 ; CHECK-P8-NEXT:    nop
 ; CHECK-P8-NEXT:    addi r1, r1, 32
 ; CHECK-P8-NEXT:    ld r0, 16(r1)
@@ -1477,7 +1477,7 @@ define dso_local fp128 @atan_f128(fp128 %x) {
 ; CHECK-NEXT:    std r0, 48(r1)
 ; CHECK-NEXT:    .cfi_def_cfa_offset 32
 ; CHECK-NEXT:    .cfi_offset lr, 16
-; CHECK-NEXT:    bl atanl
+; CHECK-NEXT:    bl atanf128
 ; CHECK-NEXT:    nop
 ; CHECK-NEXT:    addi r1, r1, 32
 ; CHECK-NEXT:    ld r0, 16(r1)
@@ -1491,7 +1491,7 @@ define dso_local fp128 @atan_f128(fp128 %x) {
 ; CHECK-P8-NEXT:    std r0, 48(r1)
 ; CHECK-P8-NEXT:    .cfi_def_cfa_offset 32
 ; CHECK-P8-NEXT:    .cfi_offset lr, 16
-; CHECK-P8-NEXT:    bl atanl
+; CHECK-P8-NEXT:    bl atanf128
 ; CHECK-P8-NEXT:    nop
 ; CHECK-P8-NEXT:    addi r1, r1, 32
 ; CHECK-P8-NEXT:    ld r0, 16(r1)
@@ -1509,7 +1509,7 @@ define dso_local fp128 @atan2_f128(fp128 %x, fp128 %y) {
 ; CHECK-NEXT:    std r0, 48(r1)
 ; CHECK-NEXT:    .cfi_def_cfa_offset 32
 ; CHECK-NEXT:    .cfi_offset lr, 16
-; CHECK-NEXT:    bl atan2l
+; CHECK-NEXT:    bl atan2f128
 ; CHECK-NEXT:    nop
 ; CHECK-NEXT:    addi r1, r1, 32
 ; CHECK-NEXT:    ld r0, 16(r1)
@@ -1523,7 +1523,7 @@ define dso_local fp128 @atan2_f128(fp128 %x, fp128 %y) {
 ; CHECK-P8-NEXT:    std r0, 48(r1)
 ; CHECK-P8-NEXT:    .cfi_def_cfa_offset 32
 ; CHECK-P8-NEXT:    .cfi_offset lr, 16
-; CHECK-P8-NEXT:    bl atan2l
+; CHECK-P8-NEXT:    bl atan2f128
 ; CHECK-P8-NEXT:    nop
 ; CHECK-P8-NEXT:    addi r1, r1, 32
 ; CHECK-P8-NEXT:    ld r0, 16(r1)
@@ -1566,7 +1566,7 @@ define dso_local fp128 @cosh_f128(fp128 %x) {
 ; CHECK-NEXT:    std r0, 48(r1)
 ; CHECK-NEXT:    .cfi_def_cfa_offset 32
 ; CHECK-NEXT:    .cfi_offset lr, 16
-; CHECK-NEXT:    bl coshl
+; CHECK-NEXT:    bl coshf128
 ; CHECK-NEXT:    nop
 ; CHECK-NEXT:    addi r1, r1, 32
 ; CHECK-NEXT:    ld r0, 16(r1)
@@ -1580,7 +1580,7 @@ define dso_local fp128 @cosh_f128(fp128 %x) {
 ; CHECK-P8-NEXT:    std r0, 48(r1)
 ; CHECK-P8-NEXT:    .cfi_def_cfa_offset 32
 ; CHECK-P8-NEXT:    .cfi_offset lr, 16
-; CHECK-P8-NEXT:    bl coshl
+; CHECK-P8-NEXT:    bl coshf128
 ; CHECK-P8-NEXT:    nop
 ; CHECK-P8-NEXT:    addi r1, r1, 32
 ; CHECK-P8-NEXT:    ld r0, 16(r1)
@@ -1598,7 +1598,7 @@ define dso_local fp128 @exp10_f128(fp128 %x) {
 ; CHECK-NEXT:    std r0, 48(r1)
 ; CHECK-NEXT:    .cfi_def_cfa_offset 32
 ; CHECK-NEXT:    .cfi_offset lr, 16
-; CHECK-NEXT:    bl exp10l
+; CHECK-NEXT:    bl exp10f128
 ; CHECK-NEXT:    nop
 ; CHECK-NEXT:    addi r1, r1, 32
 ; CHECK-NEXT:    ld r0, 16(r1)
@@ -1612,7 +1612,7 @@ define dso_local fp128 @exp10_f128(fp128 %x) {
 ; CHECK-P8-NEXT:    std r0, 48(r1)
 ; CHECK-P8-NEXT:    .cfi_def_cfa_offset 32
 ; CHECK-P8-NEXT:    .cfi_offset lr, 16
-; CHECK-P8-NEXT:    bl exp10l
+; CHECK-P8-NEXT:    bl exp10f128
 ; CHECK-P8-NEXT:    nop
 ; CHECK-P8-NEXT:    addi r1, r1, 32
 ; CHECK-P8-NEXT:    ld r0, 16(r1)
@@ -1655,7 +1655,7 @@ define dso_local fp128 @ldexp_f128(fp128 %x, i32 %y) {
 ; CHECK-NEXT:    .cfi_def_cfa_offset 32
 ; CHECK-NEXT:    .cfi_offset lr, 16
 ; CHECK-NEXT:    extsw r5, r5
-; CHECK-NEXT:    bl ldexpl
+; CHECK-NEXT:    bl ldexpf128
 ; CHECK-NEXT:    nop
 ; CHECK-NEXT:    addi r1, r1, 32
 ; CHECK-NEXT:    ld r0, 16(r1)
@@ -1670,7 +1670,7 @@ define dso_local fp128 @ldexp_f128(fp128 %x, i32 %y) {
 ; CHECK-P8-NEXT:    .cfi_def_cfa_offset 32
 ; CHECK-P8-NEXT:    .cfi_offset lr, 16
 ; CHECK-P8-NEXT:    extsw r5, r5
-; CHECK-P8-NEXT:    bl ldexpl
+; CHECK-P8-NEXT:    bl ldexpf128
 ; CHECK-P8-NEXT:    nop
 ; CHECK-P8-NEXT:    addi r1, r1, 32
 ; CHECK-P8-NEXT:    ld r0, 16(r1)
@@ -1689,7 +1689,7 @@ define dso_local { fp128, fp128 } @modf_f128(fp128 %x) {
 ; CHECK-NEXT:    .cfi_def_cfa_offset 48
 ; CHECK-NEXT:    .cfi_offset lr, 16
 ; CHECK-NEXT:    addi r5, r1, 32
-; CHECK-NEXT:    bl modfl
+; CHECK-NEXT:    bl modff128
 ; CHECK-NEXT:    nop
 ; CHECK-NEXT:    lxv v3, 32(r1)
 ; CHECK-NEXT:    addi r1, r1, 48
@@ -1708,7 +1708,7 @@ define dso_local { fp128, fp128 } @modf_f128(fp128 %x) {
 ; CHECK-P8-NEXT:    addi r30, r1, 32
 ; CHECK-P8-NEXT:    std r0, 80(r1)
 ; CHECK-P8-NEXT:    mr r5, r30
-; CHECK-P8-NEXT:    bl modfl
+; CHECK-P8-NEXT:    bl modff128
 ; CHECK-P8-NEXT:    nop
 ; CHECK-P8-NEXT:    lxvd2x vs0, 0, r30
 ; CHECK-P8-NEXT:    xxswapd v3, vs0
@@ -1729,7 +1729,7 @@ define dso_local fp128 @roundeven_f128(fp128 %x) {
 ; CHECK-NEXT:    std r0, 48(r1)
 ; CHECK-NEXT:    .cfi_def_cfa_offset 32
 ; CHECK-NEXT:    .cfi_offset lr, 16
-; CHECK-NEXT:    bl roundevenl
+; CHECK-NEXT:    bl roundevenf128
 ; CHECK-NEXT:    nop
 ; CHECK-NEXT:    addi r1, r1, 32
 ; CHECK-NEXT:    ld r0, 16(r1)
@@ -1743,7 +1743,7 @@ define dso_local fp128 @roundeven_f128(fp128 %x) {
 ; CHECK-P8-NEXT:    std r0, 48(r1)
 ; CHECK-P8-NEXT:    .cfi_def_cfa_offset 32
 ; CHECK-P8-NEXT:    .cfi_offset lr, 16
-; CHECK-P8-NEXT:    bl roundevenl
+; CHECK-P8-NEXT:    bl roundevenf128
 ; CHECK-P8-NEXT:    nop
 ; CHECK-P8-NEXT:    addi r1, r1, 32
 ; CHECK-P8-NEXT:    ld r0, 16(r1)
@@ -1761,7 +1761,7 @@ define dso_local fp128 @sinh_f128(fp128 %x) {
 ; CHECK-NEXT:    std r0, 48(r1)
 ; CHECK-NEXT:    .cfi_def_cfa_offset 32
 ; CHECK-NEXT:    .cfi_offset lr, 16
-; CHECK-NEXT:    bl sinhl
+; CHECK-NEXT:    bl sinhf128
 ; CHECK-NEXT:    nop
 ; CHECK-NEXT:    addi r1, r1, 32
 ; CHECK-NEXT:    ld r0, 16(r1)
@@ -1775,7 +1775,7 @@ define dso_local fp128 @sinh_f128(fp128 %x) {
 ; CHECK-P8-NEXT:    std r0, 48(r1)
 ; CHECK-P8-NEXT:    .cfi_def_cfa_offset 32
 ; CHECK-P8-NEXT:    .cfi_offset lr, 16
-; CHECK-P8-NEXT:    bl sinhl
+; CHECK-P8-NEXT:    bl sinhf128
 ; CHECK-P8-NEXT:    nop
 ; CHECK-P8-NEXT:    addi r1, r1, 32
 ; CHECK-P8-NEXT:    ld r0, 16(r1)
@@ -1793,7 +1793,7 @@ define dso_local fp128 @tanh_f128(fp128 %x) {
 ; CHECK-NEXT:    std r0, 48(r1)
 ; CHECK-NEXT:    .cfi_def_cfa_offset 32
 ; CHECK-NEXT:    .cfi_offset lr, 16
-; CHECK-NEXT:    bl tanhl
+; CHECK-NEXT:    bl tanhf128
 ; CHECK-NEXT:    nop
 ; CHECK-NEXT:    addi r1, r1, 32
 ; CHECK-NEXT:    ld r0, 16(r1)
@@ -1807,7 +1807,7 @@ define dso_local fp128 @tanh_f128(fp128 %x) {
 ; CHECK-P8-NEXT:    std r0, 48(r1)
 ; CHECK-P8-NEXT:    .cfi_def_cfa_offset 32
 ; CHECK-P8-NEXT:    .cfi_offset lr, 16
-; CHECK-P8-NEXT:    bl tanhl
+; CHECK-P8-NEXT:    bl tanhf128
 ; CHECK-P8-NEXT:    nop
 ; CHECK-P8-NEXT:    addi r1, r1, 32
 ; CHECK-P8-NEXT:    ld r0, 16(r1)
@@ -1825,7 +1825,7 @@ define dso_local fp128 @tan_f128(fp128 %x) {
 ; CHECK-NEXT:    std r0, 48(r1)
 ; CHECK-NEXT:    .cfi_def_cfa_offset 32
 ; CHECK-NEXT:    .cfi_offset lr, 16
-; CHECK-NEXT:    bl tanl
+; CHECK-NEXT:    bl tanf128
 ; CHECK-NEXT:    nop
 ; CHECK-NEXT:    addi r1, r1, 32
 ; CHECK-NEXT:    ld r0, 16(r1)
@@ -1839,7 +1839,7 @@ define dso_local fp128 @tan_f128(fp128 %x) {
 ; CHECK-P8-NEXT:    std r0, 48(r1)
 ; CHECK-P8-NEXT:    .cfi_def_cfa_offset 32
 ; CHECK-P8-NEXT:    .cfi_offset lr, 16
-; CHECK-P8-NEXT:    bl tanl
+; CHECK-P8-NEXT:    bl tanf128
 ; CHECK-P8-NEXT:    nop
 ; CHECK-P8-NEXT:    addi r1, r1, 32
 ; CHECK-P8-NEXT:    ld r0, 16(r1)

nikic · 2025-06-16T16:50:10Z

llvm/lib/IR/RuntimeLibcalls.cpp

+    setLibcallName(RTLIB::ROUNDEVEN_F128, "roundevenf128");
+    setLibcallName(RTLIB::SINH_F128, "sinhf128");
+    setLibcallName(RTLIB::TANH_F128, "tanhf128");
+    setLibcallName(RTLIB::TAN_F128, "tanf128");


We should share this list with the x86 code above.

(Side note: This is probably only correct for gnu environments, similar to x86.)

This is throwaway code that extracts the functional change out of a future patch that replaces this with shared tablegen

llvm/lib/IR/RuntimeLibcalls.cpp

nemanjai

LGTM.
I believe that the finite functions are provided by GLIBC on PPC for F128, but perhaps someone from IBM can confirm (@lei137 @w2yehia @RolandF77).

arsenm · 2025-06-18T04:19:53Z

Merge activity

Jun 18, 4:19 AM UTC: A user started a stack merge that includes this pull request via Graphite.
Jun 18, 4:24 AM UTC: Graphite rebased this pull request as part of a merge.
Jun 18, 4:26 AM UTC: @arsenm merged this pull request with Graphite.

This wasn't setting the correct libcall names, which default to the l suffixed libm names.

…vm#144382) This wasn't setting the correct libcall names, which default to the l suffixed libm names.

arsenm mentioned this pull request Jun 16, 2025

PowerPC: Add baseline tests for more f128 libcall handling #144381

Merged

arsenm added the backend:PowerPC label Jun 16, 2025 — with Graphite App

arsenm requested review from chenzheng1030, ecnelises, lei137, nemanjai and RolandF77 June 16, 2025 16:13

arsenm marked this pull request as ready for review June 16, 2025 16:13

llvmbot added the llvm:ir label Jun 16, 2025

nikic reviewed Jun 16, 2025

View reviewed changes

arsenm force-pushed the users/arsenm/powerpc/add-baseline-tests-f128-libcalls branch from 56b9d1e to 5f1fe5c Compare June 17, 2025 01:23

arsenm force-pushed the users/arsenm/powerpc/fix-using-long-double-libm-calls-f128 branch from 2c41a21 to 019434b Compare June 17, 2025 01:23

nikic reviewed Jun 17, 2025

View reviewed changes

llvm/lib/IR/RuntimeLibcalls.cpp Outdated Show resolved Hide resolved

nemanjai approved these changes Jun 18, 2025

View reviewed changes

arsenm force-pushed the users/arsenm/powerpc/add-baseline-tests-f128-libcalls branch from 5f1fe5c to c2058ef Compare June 18, 2025 04:21

Base automatically changed from users/arsenm/powerpc/add-baseline-tests-f128-libcalls to main June 18, 2025 04:23

arsenm added 3 commits June 18, 2025 04:23

PowerPC: Fix using long double libm functions for f128 intrinsics

35e7a80

This wasn't setting the correct libcall names, which default to the l suffixed libm names.

Split into helper function

f5cb84c

Fix double call

7e02808

arsenm force-pushed the users/arsenm/powerpc/fix-using-long-double-libm-calls-f128 branch from 77aceea to 7e02808 Compare June 18, 2025 04:23

arsenm merged commit 7b9d10d into main Jun 18, 2025
5 of 7 checks passed

arsenm deleted the users/arsenm/powerpc/fix-using-long-double-libm-calls-f128 branch June 18, 2025 04:26

fschlimb pushed a commit to fschlimb/llvm-project that referenced this pull request Jun 18, 2025

PowerPC: Fix using long double libm functions for f128 intrinsics (ll…

cfdad9b

…vm#144382) This wasn't setting the correct libcall names, which default to the l suffixed libm names.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

PowerPC: Fix using long double libm functions for f128 intrinsics #144382

PowerPC: Fix using long double libm functions for f128 intrinsics #144382

Uh oh!

arsenm commented Jun 16, 2025

Uh oh!

arsenm commented Jun 16, 2025 •

edited

Loading

Uh oh!

llvmbot commented Jun 16, 2025 •

edited

Loading

Uh oh!

nikic Jun 16, 2025

Uh oh!

arsenm Jun 16, 2025

Uh oh!

Uh oh!

nemanjai left a comment

Uh oh!

arsenm commented Jun 18, 2025 •

edited

Loading

Uh oh!

Uh oh!

Uh oh!

PowerPC: Fix using long double libm functions for f128 intrinsics #144382

PowerPC: Fix using long double libm functions for f128 intrinsics #144382

Uh oh!

Conversation

arsenm commented Jun 16, 2025

Uh oh!

arsenm commented Jun 16, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

llvmbot commented Jun 16, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

nikic Jun 16, 2025

Choose a reason for hiding this comment

Uh oh!

arsenm Jun 16, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

nemanjai left a comment

Choose a reason for hiding this comment

Uh oh!

arsenm commented Jun 18, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Merge activity

Uh oh!

Uh oh!

Uh oh!

arsenm commented Jun 16, 2025 •

edited

Loading

llvmbot commented Jun 16, 2025 •

edited

Loading

arsenm commented Jun 18, 2025 •

edited

Loading