| Message ID | 54D21CD9.7040201@st.com |
|---|---|
| State | New |
| Headers | show |
On Wed, Feb 4, 2015 at 2:21 PM, Christian Bruel <christian.bruel@st.com> wrote: > While trying to reduce the PR64835 case for ARM and x86, I noticed that the > alignment flags are cleared for x86 when attribute optimized is used. > > With the attached testcases, the visible effects are twofold : > > 1) Functions compiled in with attribute optimize (-O2) are not aligned as if > they were with the -O2 flag. > > 2) can_inline_edge_p fails because opts_for_fn (caller->decl) != opts_for_fn > (callee->decl)) even-though they are compiled with the same optimization > level. 2015-02-06 Christian Bruel <christian.bruel@st.com> PR target/64835 * config/i386/i386.c (ix86_default_align): New function. (ix86_override_options_after_change): Call ix86_default_align. (TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE): New hook. (ix86_override_options_after_change): New function. 2015-02-06 Christian Bruel <christian.bruel@st.com> PR target/64835 * gcc.dg/ipa/iinline-attr.c: New test. * gcc.target/i386/iinline-attr-2.c: New test. OK for mainline. Thanks, Uros
On Mon, Apr 27, 2015 at 8:01 PM, Uros Bizjak <ubizjak@gmail.com> wrote: > On Wed, Feb 4, 2015 at 2:21 PM, Christian Bruel <christian.bruel@st.com> wrote: >> While trying to reduce the PR64835 case for ARM and x86, I noticed that the >> alignment flags are cleared for x86 when attribute optimized is used. >> >> With the attached testcases, the visible effects are twofold : >> >> 1) Functions compiled in with attribute optimize (-O2) are not aligned as if >> they were with the -O2 flag. >> >> 2) can_inline_edge_p fails because opts_for_fn (caller->decl) != opts_for_fn >> (callee->decl)) even-though they are compiled with the same optimization >> level. > > 2015-02-06 Christian Bruel <christian.bruel@st.com> > > PR target/64835 > * config/i386/i386.c (ix86_default_align): New function. > (ix86_override_options_after_change): Call ix86_default_align. > (TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE): New hook. > (ix86_override_options_after_change): New function. > > 2015-02-06 Christian Bruel <christian.bruel@st.com> > > PR target/64835 > * gcc.dg/ipa/iinline-attr.c: New test. > * gcc.target/i386/iinline-attr-2.c: New test. > > OK for mainline. Hi Christian, I noticed case gcc.dg/ipa/iinline-attr.c failed on aarch64. The original patch is x86 specific, while the case is added as general one. Could you please have a look at this? FAIL: gcc.dg/ipa/iinline-attr.c scan-ipa-dump inline "hooray[^\\n]*inline copy in test" Thanks, bin > > Thanks, > Uros
OK I've have a look, thanks Christian On 04/30/2015 10:27 AM, Bin.Cheng wrote: > On Mon, Apr 27, 2015 at 8:01 PM, Uros Bizjak <ubizjak@gmail.com> wrote: >> On Wed, Feb 4, 2015 at 2:21 PM, Christian Bruel <christian.bruel@st.com> wrote: >>> While trying to reduce the PR64835 case for ARM and x86, I noticed that the >>> alignment flags are cleared for x86 when attribute optimized is used. >>> >>> With the attached testcases, the visible effects are twofold : >>> >>> 1) Functions compiled in with attribute optimize (-O2) are not aligned as if >>> they were with the -O2 flag. >>> >>> 2) can_inline_edge_p fails because opts_for_fn (caller->decl) != opts_for_fn >>> (callee->decl)) even-though they are compiled with the same optimization >>> level. >> >> 2015-02-06 Christian Bruel <christian.bruel@st.com> >> >> PR target/64835 >> * config/i386/i386.c (ix86_default_align): New function. >> (ix86_override_options_after_change): Call ix86_default_align. >> (TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE): New hook. >> (ix86_override_options_after_change): New function. >> >> 2015-02-06 Christian Bruel <christian.bruel@st.com> >> >> PR target/64835 >> * gcc.dg/ipa/iinline-attr.c: New test. >> * gcc.target/i386/iinline-attr-2.c: New test. >> >> OK for mainline. > > Hi Christian, > I noticed case gcc.dg/ipa/iinline-attr.c failed on aarch64. The > original patch is x86 specific, while the case is added as general > one. Could you please have a look at this? > > FAIL: gcc.dg/ipa/iinline-attr.c scan-ipa-dump inline > "hooray[^\\n]*inline copy in test" > > Thanks, > bin >> >> Thanks, >> Uros
2015-02-06 Christian Bruel <christian.bruel@st.com> PR target/64835 * config/i386/i386.c (ix86_default_align): New function. (ix86_override_options_after_change): Call ix86_default_align. (TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE): New hook. (ix86_override_options_after_change): New function. 2015-02-06 Christian Bruel <christian.bruel@st.com> PR target/64835 * gcc.dg/ipa/iinline-attr.c: New test. * gcc.target/i386/iinline-attr-2.c: New test. Index: gcc/config/i386/i386.c =================================================================== --- gcc/config/i386/i386.c (revision 220394) +++ gcc/config/i386/i386.c (working copy) @@ -3105,6 +3105,35 @@ } +/* Default align_* from the processor table. */ + +static void +ix86_default_align (struct gcc_options *opts) +{ + if (opts->x_align_loops == 0) + { + opts->x_align_loops = processor_target_table[ix86_tune].align_loop; + align_loops_max_skip = processor_target_table[ix86_tune].align_loop_max_skip; + } + if (opts->x_align_jumps == 0) + { + opts->x_align_jumps = processor_target_table[ix86_tune].align_jump; + align_jumps_max_skip = processor_target_table[ix86_tune].align_jump_max_skip; + } + if (opts->x_align_functions == 0) + { + opts->x_align_functions = processor_target_table[ix86_tune].align_func; + } +} + +/* Implement TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE hook. */ + +static void +ix86_override_options_after_change (void) +{ + ix86_default_align (&global_options); +} + /* Override various settings based on options. If MAIN_ARGS_P, the options are from the command line, otherwise they are from attributes. */ @@ -3902,20 +3931,7 @@ opts->x_ix86_regparm = REGPARM_MAX; /* Default align_* from the processor table. */ - if (opts->x_align_loops == 0) - { - opts->x_align_loops = processor_target_table[ix86_tune].align_loop; - align_loops_max_skip = processor_target_table[ix86_tune].align_loop_max_skip; - } - if (opts->x_align_jumps == 0) - { - opts->x_align_jumps = processor_target_table[ix86_tune].align_jump; - align_jumps_max_skip = processor_target_table[ix86_tune].align_jump_max_skip; - } - if (opts->x_align_functions == 0) - { - opts->x_align_functions = processor_target_table[ix86_tune].align_func; - } + ix86_default_align (opts); /* Provide default for -mbranch-cost= value. */ if (!opts_set->x_ix86_branch_cost) @@ -51928,6 +51944,9 @@ #undef TARGET_PROMOTE_FUNCTION_MODE #define TARGET_PROMOTE_FUNCTION_MODE ix86_promote_function_mode +#undef TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE +#define TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE ix86_override_options_after_change + #undef TARGET_MEMBER_TYPE_FORCES_BLK #define TARGET_MEMBER_TYPE_FORCES_BLK ix86_member_type_forces_blk Index: gcc/testsuite/gcc.dg/ipa/iinline-attr.c =================================================================== --- gcc/testsuite/gcc.dg/ipa/iinline-attr.c (revision 0) +++ gcc/testsuite/gcc.dg/ipa/iinline-attr.c (working copy) @@ -0,0 +1,27 @@ +/* Verify that simple indirect calls are inlined even when + attribute __optimize is used. */ +/* { dg-do compile } */ +/* { dg-options "-O2 -fdump-ipa-inline" } */ + +extern void non_existent(int); + +static void hooray () +{ + non_existent (1); +} + +__attribute__ ((__optimize__ ("O2"))) +static void hiphip (void (*f)()) +{ + non_existent (2); + f (); +} + +int test (void) +{ + hiphip (hooray); + return 0; +} + +/* { dg-final { scan-ipa-dump "hooray\[^\\n\]*inline copy in test" "inline" } } */ +/* { dg-final { cleanup-ipa-dump "inline" } } */ Index: gcc/testsuite/gcc.target/i386/iinline-attr-2.c =================================================================== --- gcc/testsuite/gcc.target/i386/iinline-attr-2.c (revision 0) +++ gcc/testsuite/gcc.target/i386/iinline-attr-2.c (working copy) @@ -0,0 +1,29 @@ +/* Verify that alignment flags are set when attribute __optimize is used. */ +/* { dg-do compile } */ +/* { dg-options "-O2" } */ + +extern void non_existent(int); + +__attribute__ ((__optimize__ ("O2"))) +static void hooray () +{ + non_existent (1); +} + +__attribute__ ((__optimize__ ("O2"))) +static void hiphip (void (*f)()) +{ + non_existent (2); + f (); +} + +__attribute__ ((__optimize__ ("O2"))) +int test (void) +{ + hiphip (hooray); + return 0; +} + +/* { dg-final { scan-assembler "p2align" } } */ + +
While trying to reduce the PR64835 case for ARM and x86, I noticed that the alignment flags are cleared for x86 when attribute optimized is used. With the attached testcases, the visible effects are twofold : 1) Functions compiled in with attribute optimize (-O2) are not aligned as if they were with the -O2 flag. 2) can_inline_edge_p fails because opts_for_fn (caller->decl) != opts_for_fn (callee->decl)) even-though they are compiled with the same optimization level. indeed we have: <optimization_node 0x7ffff7522000 align_functions (0x10) align_jumps (0x10) align_loops (0x10) flag_sched_stalled_insns_dep (0x1) flag_tree_parallelize_loops (0x1) flag_fp_contract_mode (0x2) flag_ira_algorithm (0) flag_ira_region (0x2) flag_simd_cost_model (0) flag_vect_cost_model (0x1) optimize (0x2) flag_aggressive_loop_optimizations (0x1) ... <optimization_node 0x7ffff7522188 flag_sched_stalled_insns_dep (0x1) flag_tree_parallelize_loops (0x1) flag_fp_contract_mode (0x2) flag_ira_algorithm (0) flag_ira_region (0x2) flag_simd_cost_model (0) flag_vect_cost_model (0x1) optimize (0x2) flag_aggressive_loop_optimizations (0x1) ... The problem is that the alignment flags are not recomputed when setting the attribute flags in DECL_FUNCTION_SPECIFIC_OPTIMIZATION. Implementing the TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE hook to set them fixes the problem. NB: TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE is also used to hold options from attribute target, so.this patch is a prerequisite to fix PR/64835 on ARM and x86 bootstrapped, regtested with no new failures for x86_64-unknown-linux-gnu Comments ? I'd like to candidate this for trunk when stage1 opens again. Many Thanks Christian