Hi Suwa-san,
On Mon, Jun 17, 2024 at 04:17:15PM +0900, Takayuki 'January June' Suwa wrote:
> The previous constant synthesis logic had some issues that were non-fatal
> but worth considering:
>
> - It didn't work with DFmode literals, because those were cast to SImode
> rather SFmode when splitting into two natural-width words by
> split_double().
>
> - It didn't work with large literals when TARGET_AUTO_LITPOOLS was enabled,
> because those were relaxed MOVI immediates rather references to literal
> pool entries,
>
> - It didn't take into account that when literals with the same RTL
> representation are pooled multiple times within a function, those entries
> are shared (especially important when optimizing for size).
>
> This patch addresses the above issues by making appropriate tweaks to the
> constant synthesis logic.
>
> gcc/ChangeLog:
>
> * config/xtensa/xtensa-protos.h (xtensa_constantsynth):
> Change the second argument from HOST_WIDE_INT to rtx.
> * config/xtensa/xtensa.cc (#include):
> Add "context.h" and "pass_manager.h".
> (machine_function): Add a new hash_map field "litpool_usage".
> (xtensa_constantsynth): Make "src" (the second operand) accept
> RTX literal instead of its value, and treat both bare and pooled
> SI/SFmode literals equally by bit-exact canonicalization into
> CONST_INT RTX internally. And then, make avoid synthesis if
> such multiple identical canonicalized literals are found in same
> function when optimizing for size. Finally, for literals where
> synthesis is not possible or has been avoided, re-emit "move"
> RTXes with canonicalized ones to increase the chances of sharing
> literal pool entries.
> * config/xtensa/xtensa.md (split patterns for constant synthesis):
> Change to simply invoke xtensa_constantsynth() as mentioned above,
> and add new patterns for when TARGET_AUTO_LITPOOLS is enabled.
> ---
> gcc/config/xtensa/xtensa-protos.h | 2 +-
> gcc/config/xtensa/xtensa.cc | 75 ++++++++++++++++++++++++++++---
> gcc/config/xtensa/xtensa.md | 56 ++++++++++++++---------
> 3 files changed, 103 insertions(+), 30 deletions(-)
This series introduced a few ICE regressions:
+FAIL: gcc.dg/atomic/c11-atomic-exec-2.c -Os (internal compiler error:
Segmentation fault)
+FAIL: gcc.dg/atomic/c11-atomic-exec-3.c -Os (internal compiler error:
Segmentation fault)
+FAIL: gcc.dg/atomic/c11-atomic-exec-4.c -Os (internal compiler error:
Segmentation fault)
+FAIL: gcc.dg/torture/vec-cvt-1.c -Os (internal compiler error: Segmentation
fault)
+FAIL: c-c++-common/torture/complex-sign-mixed-add.c -Os (internal compiler
error: Segmentation fault)
+FAIL: c-c++-common/torture/complex-sign-mixed-div.c -Os (internal compiler
error: Segmentation fault)
+FAIL: c-c++-common/torture/complex-sign-mixed-sub.c -Os (internal compiler
error: Segmentation fault)
+FAIL: gfortran.dg/bind-c-contiguous-1.f90 -Os (internal compiler error:
Segmentation fault)
+FAIL: gfortran.dg/bind-c-contiguous-4.f90 -Os (internal compiler error:
Segmentation fault)
+FAIL: gfortran.dg/minlocval_4.f90 -Os (internal compiler error:
Segmentation fault)
they all have a backtrace like this:
/home/jcmvbkbc/ws/tensilica/gcc/gcc/gcc/testsuite/gcc.dg/atomic/c11-atomic-exec-4.c:
In function 'test_main_long_double_postinc':
/home/jcmvbkbc/ws/tensilica/gcc/gcc/gcc/testsuite/gcc.dg/atomic/c11-atomic-exec-4.c:73:1:
internal compiler error: Segmentation fault
/home/jcmvbkbc/ws/tensilica/gcc/gcc/gcc/testsuite/gcc.dg/atomic/c11-atomic-exec-4.c:97:1:
note: in expansion of macro 'TEST_FUNCS'
0xf0493f crash_signal
/home/jcmvbkbc/ws/tensilica/gcc/gcc/gcc/toplev.cc:319
0x7fcc65b98d5f ???
./signal/../sysdeps/unix/sysv/linux/x86_64/sigaction.c:0
0x98cd63 lookup_page_table_entry
/home/jcmvbkbc/ws/tensilica/gcc/gcc/gcc/ggc-page.cc:630
0x98cd63 ggc_set_mark(void const*)
/home/jcmvbkbc/ws/tensilica/gcc/gcc/gcc/ggc-page.cc:1553
0x12b31bd gt_ggc_mx_hash_map_rtx_int_(void*)
./gt-xtensa.h:39
0xc19207 gt_ggc_mx_function(void*)
/home/jcmvbkbc/ws/tensilica/gcc/builds/gcc-15-1382-g448482d3d5c2-xtensa-call0-le/gcc/gtype-desc.cc:1696
0xc19207 gt_ggc_mx_function(void*)
/home/jcmvbkbc/ws/tensilica/gcc/builds/gcc-15-1382-g448482d3d5c2-xtensa-call0-le/gcc/gtype-desc.cc:1680
0x85ffe7 gt_ggc_mx_lang_tree_node(void*)
./gt-c-c-decl.h:289
0x86019c gt_ggc_mx_lang_tree_node(void*)
./gt-c-c-decl.h:381
0x860134 gt_ggc_mx_lang_tree_node(void*)
./gt-c-c-decl.h:365
0x85fa04 gt_ggc_mx_lang_tree_node(void*)
./gt-c-c-decl.h:259
0x86019c gt_ggc_mx_lang_tree_node(void*)
./gt-c-c-decl.h:381
0x86063e gt_ggc_mx_lang_tree_node(void*)
./gt-c-c-decl.h:204
0x8601be gt_ggc_mx_lang_tree_node(void*)
./gt-c-c-decl.h:383
0x85fa04 gt_ggc_mx_lang_tree_node(void*)
./gt-c-c-decl.h:259
0x86019c gt_ggc_mx_lang_tree_node(void*)
./gt-c-c-decl.h:381
0x8600ee gt_ggc_mx_lang_tree_node(void*)
./gt-c-c-decl.h:360
0xc99b58 gt_ggc_mx_ipa_return_value_summary(void*)
./gt-ipa-prop.h:44
0xc99b58 gt_ggc_mx_ipa_return_value_summary(void*)
./gt-ipa-prop.h:39
0xc99b58 gt_ggc_mx(ipa_return_value_summary*&)
./gt-ipa-prop.h:62
--
Thanks.
-- Max
