Hi! On 2025-02-27T21:51:11+0100, I wrote: > With '-mfake-ptx-alloca' enabled, the user-visible behavior changes only > for configurations where PTX 'alloca' is not available. Rather than a > compile-time 'sorry, unimplemented: dynamic stack allocation not supported' > in presence of dynamic stack allocation, compilation and assembly then > succeeds. However, attempting to link in such '*.o' files then fails due > to unresolved symbol '__GCC_nvptx__PTX_alloca_not_supported'. > > This is meant to be used in scenarios where large volumes of code are > compiled, a small fraction of which runs into dynamic stack allocation, but > these parts are not important for specific use cases, and we'd thus like the > build to succeed, and error out just upon actual, very rare use of the > offending '*.o' files.
This can be tuned further; I've pushed to trunk branch commit 199f1abeef579912b4c40c42519825cedca6530f "nvptx: Support '-mfake-ptx-alloca': defer failure to run-time 'alloca' usage", see attached. Similar to the implicit FAIL -> UNSUPPORTED due to compile-time detection of 'sorry, unimplemented: dynamic stack allocation not supported' in 'gcc/testsuite/lib/gcc-dg.exp:gcc-dg-prune', we should also be able to implement a corrsponding thing for this run-time failure, via new machinery in the replacement 'gcc/testsuite/lib/gcc-dg.exp:${tool}_load', via scanning and modifying 'result'. (..., but I'm not working on that right now.) Grüße Thomas
>From 199f1abeef579912b4c40c42519825cedca6530f Mon Sep 17 00:00:00 2001 From: Thomas Schwinge <tschwi...@baylibre.com> Date: Sun, 6 Apr 2025 17:44:18 +0200 Subject: [PATCH] nvptx: Support '-mfake-ptx-alloca': defer failure to run-time 'alloca' usage Follow-up to commit 1146410c0feb0e82c689b1333fdf530a2b34dc2b "nvptx: Support '-mfake-ptx-alloca'". '-mfake-ptx-alloca' is applicable only for configurations where PTX 'alloca' is not supported, where target libraries are built with it enabled (that is, libstdc++, libgfortran). This change progresses: [-FAIL:-]{+PASS:+} g++.dg/tree-ssa/pr20458.C -std=gnu++17 (test for excess errors) [-UNRESOLVED:-]{+PASS:+} g++.dg/tree-ssa/pr20458.C -std=gnu++17 [-compilation failed to produce executable-]{+execution test+} [-FAIL:-]{+PASS:+} g++.dg/tree-ssa/pr20458.C -std=gnu++26 (test for excess errors) [-UNRESOLVED:-]{+PASS:+} g++.dg/tree-ssa/pr20458.C -std=gnu++26 [-compilation failed to produce executable-]{+execution test+} UNSUPPORTED: g++.dg/tree-ssa/pr20458.C -std=gnu++98: exception handling not supported ..., and "enables" a few test cases: FAIL: g++.old-deja/g++.other/sibcall1.C -std=gnu++17 (test for excess errors) [Etc.] FAIL: g++.old-deja/g++.other/unchanging1.C -std=gnu++17 (test for excess errors) [Etc.] ..., which now (unrelatedly to 'alloca', and in the same way as configurations where PTX 'alloca' is supported) FAIL due to: unresolved symbol _Unwind_DeleteException collect2: error: ld returned 1 exit status Most importantly, it progresses ~830 libstdc++ test cases: [-FAIL:-]{+PASS:+} [...] (test for excess errors) ..., with (if applicable, for most of them): [-UNRESOLVED:-]{+PASS:+} [...] [-compilation failed to produce executable-]{+execution test+} ..., or just a few 'FAIL: [...] execution test' where these test cases also FAIL in configurations where PTX 'alloca' is supported, or ~120 instances of 'FAIL: [...] execution test' due to run-time 'GCC/nvptx: sorry, unimplemented: dynamic stack allocation not supported'. This change also resolves the cases noted in commit bac2d8a246892334e24dfa7d62be0cd0648c5606 "nvptx: Build libgfortran with '-mfake-ptx-alloca' [PR107635]": | With '-mfake-ptx-alloca', libgfortran again succeeds to build, and compared | to before, we've got only a small number of regressions due to nvptx 'ld' | complaining about 'unresolved symbol __GCC_nvptx__PTX_alloca_not_supported': | | [-PASS:-]{+FAIL:+} gfortran.dg/coarray/codimension_2.f90 -fcoarray=lib -O2 -lcaf_single (test for excess errors) [-FAIL:-]{+PASS:+} gfortran.dg/coarray/codimension_2.f90 -fcoarray=lib -O2 -lcaf_single (test for excess errors) | [-PASS:-]{+FAIL:+} gfortran.dg/coarray/event_4.f08 -fcoarray=lib -O2 -lcaf_single (test for excess errors) | [-PASS:-]{+UNRESOLVED:+} gfortran.dg/coarray/event_4.f08 -fcoarray=lib -O2 -lcaf_single [-execution test-]{+compilation failed to produce executable+} [-FAIL:-]{+PASS:+} gfortran.dg/coarray/event_4.f08 -fcoarray=lib -O2 -lcaf_single (test for excess errors) [-UNRESOLVED:-]{+PASS:+} gfortran.dg/coarray/event_4.f08 -fcoarray=lib -O2 -lcaf_single [-compilation failed to produce executable-]{+execution test+} | [-PASS:-]{+FAIL:+} gfortran.dg/coarray/fail_image_2.f08 -fcoarray=lib -O2 -lcaf_single (test for excess errors) | [-PASS:-]{+UNRESOLVED:+} gfortran.dg/coarray/fail_image_2.f08 -fcoarray=lib -O2 -lcaf_single [-execution test-]{+compilation failed to produce executable+} [-FAIL:-]{+PASS:+} gfortran.dg/coarray/fail_image_2.f08 -fcoarray=lib -O2 -lcaf_single (test for excess errors) [-UNRESOLVED:-]{+PASS:+} gfortran.dg/coarray/fail_image_2.f08 -fcoarray=lib -O2 -lcaf_single [-compilation failed to produce executable-]{+execution test+} | [-PASS:-]{+FAIL:+} gfortran.dg/coarray/proc_pointer_assign_1.f90 -fcoarray=lib -O2 -lcaf_single (test for excess errors) | [-PASS:-]{+UNRESOLVED:+} gfortran.dg/coarray/proc_pointer_assign_1.f90 -fcoarray=lib -O2 -lcaf_single [-execution test-]{+compilation failed to produce executable+} [-FAIL:-]{+PASS:+} gfortran.dg/coarray/proc_pointer_assign_1.f90 -fcoarray=lib -O2 -lcaf_single (test for excess errors) [-UNRESOLVED:-]{+PASS:+} gfortran.dg/coarray/proc_pointer_assign_1.f90 -fcoarray=lib -O2 -lcaf_single [-compilation failed to produce executable-]{+execution test+} | [-PASS:-]{+FAIL:+} gfortran.dg/coarray_43.f90 -O (test for excess errors) [-FAIL:-]{+PASS:+} gfortran.dg/coarray_43.f90 -O (test for excess errors) ..., and further progresses: [-FAIL:-]{+PASS:+} gfortran.dg/coarray_lib_comm_1.f90 -O0 (test for excess errors) [-UNRESOLVED:-]{+FAIL:+} gfortran.dg/coarray_lib_comm_1.f90 -O0 [-compilation failed to produce executable-]{+execution test+} [Etc.] ..., which now (unrelatedly to 'alloca', and in the same way as configurations where PTX 'alloca' is supported) FAILs due to: error : Prototype doesn't match for '_gfortran_caf_transfer_between_remotes' in 'input file 9 at offset 159897', first defined in 'input file 9 at offset 159897' error : Prototype doesn't match for '_gfortran_caf_stop_numeric' in 'input file 9 at offset 159897', first defined in 'input file 9 at offset 159897' nvptx-run: cuLinkAddData failed: device kernel image is invalid (CUDA_ERROR_INVALID_SOURCE, 300) gcc/ * config/nvptx/nvptx.opt (-mfake-ptx-alloca): Update. gcc/testsuite/ * gcc.target/nvptx/alloca-2-O0_-mfake-ptx-alloca.c: Adjust. libgcc/ * config/nvptx/alloca.c: New. * config/nvptx/t-nvptx (LIB2ADD): Add it. --- gcc/config/nvptx/nvptx.opt | 10 ++--- .../nvptx/alloca-2-O0_-mfake-ptx-alloca.c | 7 ++-- libgcc/config/nvptx/alloca.c | 38 +++++++++++++++++++ libgcc/config/nvptx/t-nvptx | 3 +- 4 files changed, 49 insertions(+), 9 deletions(-) create mode 100644 libgcc/config/nvptx/alloca.c diff --git a/gcc/config/nvptx/nvptx.opt b/gcc/config/nvptx/nvptx.opt index 02d36b33025b..9be81aecdddf 100644 --- a/gcc/config/nvptx/nvptx.opt +++ b/gcc/config/nvptx/nvptx.opt @@ -173,12 +173,12 @@ Target Var(nvptx_fake_ptx_alloca) Init(0) Undocumented ; With '-mfake-ptx-alloca' enabled, the user-visible behavior changes only ; for configurations where PTX 'alloca' is not available. Rather than a ; compile-time 'sorry, unimplemented: dynamic stack allocation not supported' -; in presence of dynamic stack allocation, compilation and assembly then -; succeeds. However, attempting to link in such '*.o' files then fails due -; to unresolved symbol '__GCC_nvptx__PTX_alloca_not_supported'. +; in presence of dynamic stack allocation, with '-mfake-ptx-alloca' enabled, +; compilation, assembly, and linking succeeds, as does execution, in case that +; 'alloca' is not attempted (if only used in error code paths, for example), +; and a run-time failure only in case that 'alloca' is actually attempted. ; ; This is meant to be used in scenarios where large volumes of code are ; compiled, a small fraction of which runs into dynamic stack allocation, but ; these parts are not important for specific use cases, and we'd thus like the -; build to succeed, and error out just upon actual, very rare use of the -; offending '*.o' files. +; build to succeed, and error out just upon actual, very rare use of 'alloca'. diff --git a/gcc/testsuite/gcc.target/nvptx/alloca-2-O0_-mfake-ptx-alloca.c b/gcc/testsuite/gcc.target/nvptx/alloca-2-O0_-mfake-ptx-alloca.c index 4cc4d0c93f28..92562cde4008 100644 --- a/gcc/testsuite/gcc.target/nvptx/alloca-2-O0_-mfake-ptx-alloca.c +++ b/gcc/testsuite/gcc.target/nvptx/alloca-2-O0_-mfake-ptx-alloca.c @@ -1,4 +1,4 @@ -/* { dg-do link } */ +/* { dg-do run } */ /* { dg-options {-O0 -mno-soft-stack} } */ /* { dg-additional-options -march=sm_30 } */ /* { dg-additional-options -mfake-ptx-alloca } */ @@ -13,6 +13,7 @@ main(void) /* { dg-final { scan-assembler-times {(?n)^\.extern \.func \(\.param\.u64 %value_out\) __GCC_nvptx__PTX_alloca_not_supported \(\.param\.u64 %in_ar0\);$} 1 } } */ -/* { dg-message __GCC_nvptx__PTX_alloca_not_supported {unresolved symbol} { target *-*-* } 0 } */ +/* { dg-bogus __GCC_nvptx__PTX_alloca_not_supported {unresolved symbol} { target *-*-* } 0 } */ -/* { dg-final output-exists-not } */ +/* { dg-output {GCC/nvptx: sorry, unimplemented: dynamic stack allocation not supported[\r\n]+} } + { dg-shouldfail __GCC_nvptx__PTX_alloca_not_supported } */ diff --git a/libgcc/config/nvptx/alloca.c b/libgcc/config/nvptx/alloca.c new file mode 100644 index 000000000000..09bdeb682859 --- /dev/null +++ b/libgcc/config/nvptx/alloca.c @@ -0,0 +1,38 @@ +/* Fake 'alloca' implementation. + + Copyright (C) 2025 Free Software Foundation, Inc. + + This file is free software; you can redistribute it and/or modify it + under the terms of the GNU General Public License as published by the + Free Software Foundation; either version 3, or (at your option) any + later version. + + This file is distributed in the hope that it will be useful, but + WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + General Public License for more details. + + Under Section 7 of GPL version 3, you are granted additional + permissions described in the GCC Runtime Library Exception, version + 3.1, as published by the Free Software Foundation. + + You should have received a copy of the GNU General Public License and + a copy of the GCC Runtime Library Exception along with this program; + see the files COPYING3 and COPYING.RUNTIME respectively. If not, see + <http://www.gnu.org/licenses/>. */ + +/* For '-mfake-ptx-alloca', in case real PTX 'alloca' is not available. + With this function defined, we don't get a link-time failure + (unresolved symbol '__GCC_nvptx__PTX_alloca_not_supported'), but rather: + successful execution, in case that 'alloca' is not attempted (if only used + in error code paths, for example), and a run-time failure only in case that + 'alloca' is actually attempted. */ + +void * +__GCC_nvptx__PTX_alloca_not_supported (__SIZE_TYPE__ size __attribute__ ((unused))) +{ + __builtin_printf ("GCC/nvptx: sorry, unimplemented:" + " dynamic stack allocation not supported\n"); + __builtin_abort (); + return 0; +} diff --git a/libgcc/config/nvptx/t-nvptx b/libgcc/config/nvptx/t-nvptx index f295898d69fc..5e2e27898fd6 100644 --- a/libgcc/config/nvptx/t-nvptx +++ b/libgcc/config/nvptx/t-nvptx @@ -1,7 +1,8 @@ LIB2ADD=$(srcdir)/config/nvptx/reduction.c \ $(srcdir)/config/nvptx/mgomp.c \ $(srcdir)/config/nvptx/atomic.c \ - $(srcdir)/config/nvptx/unwind-nvptx.c + $(srcdir)/config/nvptx/unwind-nvptx.c \ + $(srcdir)/config/nvptx/alloca.c LIB2ADDEH= LIB2FUNCS_EXCLUDE= -- 2.34.1