[Bug tree-optimization/113622] [11/12/13 Regression] ICE with vectors in named registers
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113622 Richard Biener changed: What|Removed |Added Status|ASSIGNED|NEW Assignee|rguenth at gcc dot gnu.org |unassigned at gcc dot gnu.org
[Bug tree-optimization/113622] [11/12/13 Regression] ICE with vectors in named registers
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113622 --- Comment #26 from Richard Biener --- I've enabled vec_set for hard-regs on the branch and plugged the vectorizer hole for GCC 13. I'm not sure to what extent we need the expansion change.
[Bug tree-optimization/113622] [11/12/13 Regression] ICE with vectors in named registers
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113622 --- Comment #25 from GCC Commits --- The releases/gcc-13 branch has been updated by Richard Biener : https://gcc.gnu.org/g:d4c0800aab864bb95260e12342d18695c6ebbec8 commit r13-8483-gd4c0800aab864bb95260e12342d18695c6ebbec8 Author: Richard Biener Date: Mon Jan 29 09:47:31 2024 +0100 middle-end/113622 - allow .VEC_SET and .VEC_EXTRACT for global hard regs The following expands .VEC_SET and .VEC_EXTRACT instruction selection to global hard registers, not only automatic variables (possibly) promoted to registers. This can avoid some ICEs later and create better code. PR middle-end/113622 * gimple-isel.cc (gimple_expand_vec_set_extract_expr): Also allow DECL_HARD_REGISTER variables. * gcc.target/i386/pr113622-1.c: New testcase. (cherry picked from commit 96bc048d78f804bac0fa7b2ca3b6dd3a04c68217)
[Bug tree-optimization/113622] [11/12/13 Regression] ICE with vectors in named registers
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113622 --- Comment #24 from Xi Ruoyao --- It looks I can rewrite the LoongArch test case (still broken though ICE is stopped) using check-function-bodies. Will try later...
[Bug tree-optimization/113622] [11/12/13 Regression] ICE with vectors in named registers
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113622
--- Comment #23 from rguenther at suse dot de ---
On Tue, 30 Jan 2024, xry111 at gcc dot gnu.org wrote:
> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113622
>
> --- Comment #22 from Xi Ruoyao ---
> On x86_64:
>
> $ cat t.c
> typedef float __attribute__ ((vector_size (16))) vec;
> typedef int __attribute__ ((vector_size (16))) ivec;
> register vec a asm("xmm0"), b asm("xmm1");
> register ivec c asm("xmm2");
>
> void
> test (void)
> {
> for (int i = 0; i < 4; i++)
> c[i] = a[i] < b[i] ? -1 : 1;
> }
> $ gcc/cc1 -msse2 t.c -O2 -fno-vect-cost-model -nostdinc -ffixed-xmm{0,1,2}
> t.c: In function 'test':
> t.c:7:1: internal compiler error: in expand_expr_addr_expr_1, at expr.cc:9139
> 7 | test (void)
> | ^~~~
> 0x10e6d6e expand_expr_addr_expr_1
> ../../gcc/gcc/expr.cc:9139
> 0x10e76e2 expand_expr_addr_expr
> ../../gcc/gcc/expr.cc:9252
> 0x10f73a7 expand_expr_real_1(tree_node*, rtx_def*, machine_mode,
> expand_modifier, rtx_def**, bool)
> ../../gcc/gcc/expr.cc:12585
> 0x10e7dc8 expand_expr_real(tree_node*, rtx_def*, machine_mode,
> expand_modifier,
> rtx_def**, bool)
> ../../gcc/gcc/expr.cc:9440
> 0xef7346 expand_expr(tree_node*, rtx_def*, machine_mode, expand_modifier)
> ../../gcc/gcc/expr.h:316
> 0x10e91fa expand_expr_real_2(separate_ops*, rtx_def*, machine_mode,
> expand_modifier)
> ../../gcc/gcc/expr.cc:9762
> 0x10ef77d expand_expr_real_gassign(gassign*, rtx_def*, machine_mode,
> expand_modifier, rtx_def**, bool)
> ../../gcc/gcc/expr.cc:11096
> 0xf2db31 expand_gimple_stmt_1
> ../../gcc/gcc/cfgexpand.cc:4010
> 0xf2ddd4 expand_gimple_stmt
> ../../gcc/gcc/cfgexpand.cc:4071
> 0xf36844 expand_gimple_basic_block
> ../../gcc/gcc/cfgexpand.cc:6127
> 0xf38ff8 execute
> ../../gcc/gcc/cfgexpand.cc:6866
>
> Should I open a new ticket or add back 14 Regression to the subject?
Please open a new ticked - this seems to be another vectorizer issue.
We end up with the invalid
_28 = (sizetype)
[Bug tree-optimization/113622] [11/12/13 Regression] ICE with vectors in named registers
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113622
--- Comment #22 from Xi Ruoyao ---
On x86_64:
$ cat t.c
typedef float __attribute__ ((vector_size (16))) vec;
typedef int __attribute__ ((vector_size (16))) ivec;
register vec a asm("xmm0"), b asm("xmm1");
register ivec c asm("xmm2");
void
test (void)
{
for (int i = 0; i < 4; i++)
c[i] = a[i] < b[i] ? -1 : 1;
}
$ gcc/cc1 -msse2 t.c -O2 -fno-vect-cost-model -nostdinc -ffixed-xmm{0,1,2}
t.c: In function 'test':
t.c:7:1: internal compiler error: in expand_expr_addr_expr_1, at expr.cc:9139
7 | test (void)
| ^~~~
0x10e6d6e expand_expr_addr_expr_1
../../gcc/gcc/expr.cc:9139
0x10e76e2 expand_expr_addr_expr
../../gcc/gcc/expr.cc:9252
0x10f73a7 expand_expr_real_1(tree_node*, rtx_def*, machine_mode,
expand_modifier, rtx_def**, bool)
../../gcc/gcc/expr.cc:12585
0x10e7dc8 expand_expr_real(tree_node*, rtx_def*, machine_mode, expand_modifier,
rtx_def**, bool)
../../gcc/gcc/expr.cc:9440
0xef7346 expand_expr(tree_node*, rtx_def*, machine_mode, expand_modifier)
../../gcc/gcc/expr.h:316
0x10e91fa expand_expr_real_2(separate_ops*, rtx_def*, machine_mode,
expand_modifier)
../../gcc/gcc/expr.cc:9762
0x10ef77d expand_expr_real_gassign(gassign*, rtx_def*, machine_mode,
expand_modifier, rtx_def**, bool)
../../gcc/gcc/expr.cc:11096
0xf2db31 expand_gimple_stmt_1
../../gcc/gcc/cfgexpand.cc:4010
0xf2ddd4 expand_gimple_stmt
../../gcc/gcc/cfgexpand.cc:4071
0xf36844 expand_gimple_basic_block
../../gcc/gcc/cfgexpand.cc:6127
0xf38ff8 execute
../../gcc/gcc/cfgexpand.cc:6866
Should I open a new ticket or add back 14 Regression to the subject?
[Bug tree-optimization/113622] [11/12/13 Regression] ICE with vectors in named registers
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113622
--- Comment #21 from Xi Ruoyao ---
This still blows up on LoongArch even after r14-8498:
typedef float __attribute__ ((vector_size (16))) vec;
typedef int __attribute__ ((vector_size (16))) ivec;
register vec a asm("f25"), b asm("f26");
register ivec c asm("f27");
void
test (void)
{
for (int i = 0; i < 4; i++)
c[i] = a[i] < b[i] ? -1 : 1;
}
$ gcc/cc1 -msimd=lsx t.c -O2 -fno-vect-cost-model -nostdinc
t.c: In function ‘test’:
t.c:7:1: internal compiler error: in expand_expr_addr_expr_1, at expr.cc:9139
7 | test (void)
| ^~~~
0x102d7cb expand_expr_addr_expr_1
../../gcc/gcc/expr.cc:9139
0x102e13e expand_expr_addr_expr
../../gcc/gcc/expr.cc:9252
0x103df07 expand_expr_real_1(tree_node*, rtx_def*, machine_mode,
expand_modifier, rtx_def**, bool)
../../gcc/gcc/expr.cc:12585
0x102e824 expand_expr_real(tree_node*, rtx_def*, machine_mode, expand_modifier,
rtx_def**, bool)
../../gcc/gcc/expr.cc:9440
0xe45096 expand_expr(tree_node*, rtx_def*, machine_mode, expand_modifier)
../../gcc/gcc/expr.h:316
0x102fc56 expand_expr_real_2(separate_ops*, rtx_def*, machine_mode,
expand_modifier)
../../gcc/gcc/expr.cc:9762
0x10361c9 expand_expr_real_gassign(gassign*, rtx_def*, machine_mode,
expand_modifier, rtx_def**, bool)
../../gcc/gcc/expr.cc:11096
0xe798de expand_gimple_stmt_1
../../gcc/gcc/cfgexpand.cc:4010
0xe79b81 expand_gimple_stmt
../../gcc/gcc/cfgexpand.cc:4071
0xe825e3 expand_gimple_basic_block
../../gcc/gcc/cfgexpand.cc:6127
0xe84b95 execute
../../gcc/gcc/cfgexpand.cc:6866
Interestingly -fno-vect-cost-model is needed to trigger the ICE.
[Bug tree-optimization/113622] [11/12/13 Regression] ICE with vectors in named registers
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113622
--- Comment #20 from GCC Commits ---
The master branch has been updated by Jakub Jelinek :
https://gcc.gnu.org/g:47b81161c98cf2ff5495d4aa6386cc3c87f9d27b
commit r14-8515-g47b81161c98cf2ff5495d4aa6386cc3c87f9d27b
Author: Jakub Jelinek
Date: Tue Jan 30 09:31:22 2024 +0100
testsuite: Fix up pr113622-{2,3}.c for i686-linux [PR113622]
The 2 new tests FAIL for me on i686-linux:
.../gcc/testsuite/gcc.target/i386/pr113622-2.c:5:14: error: data type of
'a' isn't suitable for a register
.../gcc/testsuite/gcc.target/i386/pr113622-2.c:5:29: error: data type of
'b' isn't suitable for a register
.../gcc/testsuite/gcc.target/i386/pr113622-2.c:5:44: error: data type of
'c' isn't suitable for a register
The problem is that the tests use vectors of double, something added
only in SSE2, while the testcases ask for just -msse which only provides
vectors of floats.
So, either it should be using floats instead of doubles, or we need
to add -msse2 to dg-options.
I've done the latter.
2024-01-30 Jakub Jelinek
PR middle-end/113622
* gcc.target/i386/pr113622-2.c: Use -msse2 instead of -msse in
dg-options.
* gcc.target/i386/pr113622-3.c: Likewise.
[Bug tree-optimization/113622] [11/12/13 Regression] ICE with vectors in named registers
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113622 Richard Biener changed: What|Removed |Added Known to work||14.0 Summary|[11/12/13/14 Regression]|[11/12/13 Regression] ICE |ICE with vectors in named |with vectors in named |registers |registers --- Comment #19 from Richard Biener --- Should be fixed on trunk, not sure to what extent backporting is suitable.
