https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110472
Bug ID: 110472
Summary: 60% slowdown with fwrapv when using openmp
Product: gcc
Version: 10.2.1
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: ryanpholt at me dot com
Target Milestone: ---
Created attachment 55423
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=55423=edit
Reproduction file
Compiling the attached example with -fwrapv inhibits some optimizations and
results in a massive slowdown. It appears to be related to the use of OpenMP. I
know that fwrapv can result in slowdowns; however, I do not believe that it
needs to in this example.
In the loop nest below, gcc appears to believe the computations with the loop
induction variables (ex. 'i2 = i * 21504 + i1 * 96;') will overflow. However,
the code is looping over fixed size data and so I believe gcc should be able to
determine that overflow is not possible. Perhaps some range analysis stops
working across the openmp runtime boundary? The issue is fixed if I remove the
openmp pragma.
The issue is also fixed if I change the loop induction variables to be declared
as int64_t rather than int.
I also tried clang16 and did not observe the issue with fwrapv.
#pragma omp parallel for \
num_threads(omp_get_max_threads()) \
private(i1,u0,b_u0,i2,i3,i4,i5,i6,i7,i8,i10,i12,i14,i15) \
firstprivate(r2)
for (i = 0; i < 7; i++) {
for (i1 = 0; i1 < 7; i1++) {
u0 = i * -32 + 222;
if (u0 > 32) {
u0 = 32;
}
b_u0 = i1 * -32 + 222;
if (b_u0 > 32) {
b_u0 = 32;
}
i2 = i * 21504 + i1 * 96;
i3 = i * 227328 + (i1 << 10);
for (i4 = 0; i4 < u0; i4++) {
for (i5 = 0; i5 < b_u0; i5++) {
for (i6 = 0; i6 < 3; i6++) {
i7 = i5 + i6;
for (i8 = 0; i8 < 3; i8++) {
for (i10 = 0; i10 < 3; i10++) {
i12 = i4 + i10;
for (i14 = 0; i14 < 32; i14++) {
i15 = ((i3 + 7104 * i4) + (i5 << 5)) + i14;
(r2)[i15] += ((float *)_0[0][0][0][0])[((i2 + 672 * i12) +
3 *
i7) + i8] * ((float *)&__constant_3x3x3x32xf32[0][0][0][0])
[((288 * i10 + 96 * i6) + (i8 << 5)) + i14];
}
}
}
}
}
}
}
}
Repro:
gcc -O3 -fopenmp -lpthread -fwrapv predict.i
./a.out
(Remove the -fwrapv to observe a major speedup)
Using built-in specs.
COLLECT_GCC=gcc
COLLECT_LTO_WRAPPER=/usr/lib/gcc/x86_64-linux-gnu/10/lto-wrapper
OFFLOAD_TARGET_NAMES=nvptx-none:amdgcn-amdhsa:hsa
OFFLOAD_TARGET_DEFAULT=1
Target: x86_64-linux-gnu
Configured with: ../src/configure -v --with-pkgversion='Debian 10.2.1-6'
--with-bugurl=file:///usr/share/doc/gcc-10/README.Bugs
--enable-languages=c,ada,c++,go,brig,d,fortran,objc,obj-c++,m2 --prefix=/usr
--with-gcc-major-version-only --program-suffix=-10
--program-prefix=x86_64-linux-gnu- --enable-shared --enable-linker-build-id
--libexecdir=/usr/lib --without-included-gettext --enable-threads=posix
--libdir=/usr/lib --enable-nls --enable-bootstrap --enable-clocale=gnu
--enable-libstdcxx-debug --enable-libstdcxx-time=yes
--with-default-libstdcxx-abi=new --enable-gnu-unique-object
--disable-vtable-verify --enable-plugin --enable-default-pie --with-system-zlib
--enable-libphobos-checking=release --with-target-system-zlib=auto
--enable-objc-gc=auto --enable-multiarch --disable-werror --with-arch-32=i686
--with-abi=m64 --with-multilib-list=m32,m64,mx32 --enable-multilib
--with-tune=generic
--enable-offload-targets=nvptx-none=/build/gcc-10-Km9U7s/gcc-10-10.2.1/debian/tmp-nvptx/usr,amdgcn-amdhsa=/build/gcc-10-Km9U7s/gcc-10-10.2.1/debian/tmp-gcn/usr,hsa
--without-cuda-driver --enable-checking=release --build=x86_64-linux-gnu
--host=x86_64-linux-gnu --target=x86_64-linux-gnu
--with-build-config=bootstrap-lto-lean --enable-link-mutex
Thread model: posix
Supported LTO compression algorithms: zlib zstd
gcc version 10.2.1 20210110 (Debian 10.2.1-6)
COLLECT_GCC_OPTIONS='-v' '-save-temps' '-O3' '-fopenmp' '-fwrapv'
'-mtune=generic' '-march=x86-64' '-pthread'
/usr/lib/gcc/x86_64-linux-gnu/10/cc1 -E -quiet -v -imultiarch x86_64-linux-gnu
-D_REENTRANT main.c -mtune=generic -march=x86-64 -fopenmp -fwrapv -O3
-fpch-preprocess -fasynchronous-unwind-tables -o main.i
ignoring nonexistent directory "/usr/local/include/x86_64-linux-gnu"
ignoring nonexistent directory "/usr/lib/gcc/x86_64-linux-gnu/10/include-fixed"
ignoring nonexistent directory
"/usr/lib/gcc/x86_64-linux-gnu/10/../../../../x86_64-linux-gnu/include"
#include "..." search starts here:
#include <...> search starts here:
/usr/lib/gcc/x86_64-linux-gnu/10/include
/usr/local/include
/usr/include/x86_64-linux-gnu
/usr/include
End of search list.
COLLECT_GCC_OPTIONS='-v' '-save-temps' '-O3' '-fopenmp'
