https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121394
Bug ID: 121394
Summary: [16 Regression][gcn] Since r16-2595-gf1c80147641783:
link-time error: libm_a-e_atan2.o):(.rodata.cst32):
SHF_MERGE section size (56) must be a multiple of
sh_entsize (32)
Product: gcc
Version: 16.0
Status: UNCONFIRMED
Keywords: wrong-code
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: burnus at gcc dot gnu.org
CC: ams at gcc dot gnu.org, pinskia at gcc dot gnu.org
Target Milestone: ---
Target: gcn
When compiling + linking libgomp.fortran/fortran-torture_execute_math.f90
with AMD GCN offloading enabled ("-fopenmp -foffload=amdgcn-amdhsa" + in-tree
build of newlib), it fails with:
ld: error:
.../x86_64-pc-linux-gnu/16.0.0/accel/amdgcn-amdhsa/../../../../../../amdgcn-amdhsa/lib/gfx900/libm.a(libm_a-e_atan2.o):(.rodata.cst32):
SHF_MERGE section size (56) must be a multiple of sh_entsize (32)
collect2: error: ld returned 1 exit status
gcn mkoffload: fatal error: .../bin/x86_64-pc-linux-gnu-accel-amdgcn-amdhsa-gcc
returned 1 exit status
compilation terminated.
* * *
Bisecting points to:
commit r16-2595-gf1c8014764178335e3b949e06b894ff5775beae5
Author: Andrew Pinski
AuthorDate: Fri Jul 25 16:16:36 2025 -0700
Commit: Andrew Pinski
CommitDate: Mon Jul 28 23:43:32 2025 -0700
Improve mergability of CSWTCH [PR120523]
* * *
The file $(build)/amdgcn-amdhsa/newlib/libm/math/libm_a-e_atan2.s has:
--- libm_a-e_atan2.s-old 2025-08-04 12:53:47.395611975 +0200
+++ libm_a-e_atan2.s-new 2025-08-04 13:11:52.569728506 +0200
@@ -544,19 +544,19 @@
.LFE14:
.size __ieee754_atan2, .-__ieee754_atan2
- .section .rodata
- .align 16
+ .section .rodata.cst32,"aM",@progbits,32
+ .align 32
.type CSWTCH.7, @object
.size CSWTCH.7, 24
CSWTCH.7:
.long 0
.long -2147483648
.long 1413754136
.long 1074340347
.long 1413754136
.long -1073143301
- .align 16
+ .align 32
.type CSWTCH.6, @object
.size CSWTCH.6, 24
CSWTCH.6:
.long 1413754136
.long -1075240453
