This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH, rs6000] generate loop code for memcmp inline expansion
- From: Szabolcs Nagy <szabolcs dot nagy at arm dot com>
- To: Aaron Sawdey <acsawdey at linux dot vnet dot ibm dot com>, Segher Boessenkool <segher at kernel dot crashing dot org>
- Cc: nd at arm dot com, gcc-patches at gcc dot gnu dot org, David Edelsohn <dje dot gcc at gmail dot com>, Jeff Law <law at redhat dot com>
- Date: Wed, 10 Jan 2018 14:25:36 +0000
- Subject: Re: [PATCH, rs6000] generate loop code for memcmp inline expansion
- Authentication-results: sourceware.org; auth=none
- Authentication-results: spf=none (sender IP is ) smtp.mailfrom=Szabolcs dot Nagy at arm dot com;
- Nodisclaimer: True
- References: <1513007385.30918.2.camel@linux.vnet.ibm.com> <20171212161328.GS10515@gate.crashing.org> <1515440268.10978.33.camel@linux.vnet.ibm.com>
- Spamdiagnosticmetadata: NSPM
- Spamdiagnosticoutput: 1:99
On 08/01/18 19:37, Aaron Sawdey wrote:
> On Tue, 2017-12-12 at 10:13 -0600, Segher Boessenkool wrote:
>> > Please fix those trivialities, and it's okay for trunk (after the
>> > rtlanal patch is approved too). Thanks!
> Here's the final version of this, which is committed as 256351.
>
>
> 2018-01-08 Aaron Sawdey <acsawdey@linux.vnet.ibm.com>
>
> * config/rs6000/rs6000-string.c (do_load_for_compare_from_addr): New
> function.
> (do_ifelse): New function.
> (do_isel): New function.
> (do_sub3): New function.
> (do_add3): New function.
> (do_load_mask_compare): New function.
> (do_overlap_load_compare): New function.
> (expand_compare_loop): New function.
> (expand_block_compare): Call expand_compare_loop() when appropriate.
> * config/rs6000/rs6000.opt (-mblock-compare-inline-limit): Change
> option description.
> (-mblock-compare-inline-loop-limit): New option.
>
...
> Index: gcc/testsuite/gcc.dg/memcmp-1.c
> ===================================================================
> --- gcc/testsuite/gcc.dg/memcmp-1.c (revision 256350)
> +++ gcc/testsuite/gcc.dg/memcmp-1.c (working copy)
> @@ -14,11 +14,80 @@
> #ifndef NRAND
> #define NRAND 10000
> #endif
> -#define MAX_SZ 200
> +#define MAX_SZ 600
>
i see timeouts when running aarch64-none-elf tests in some
emulator environments:
WARNING: program timed out.
FAIL: gcc.dg/memcmp-1.c execution test
if there is a way to reduce the iteration count or the
tested variants that would help slow targets.
> +#define DEF_RS(ALIGN) \
> +static void test_memcmp_runtime_size_ ## ALIGN (const char *str1, \
> + const char *str2, \
> + size_t sz, int expect) \
> +{ \
> + char three[8192] __attribute__ ((aligned (4096))); \
> + char four[8192] __attribute__ ((aligned (4096))); \
> + char *a, *b; \
> + int i,j,a1,a2,r; \
> + for (j = 0; j < 2; j++) \
> + { \
> + for (i = 0; i < 2; i++) \
> + { \
> + a = three+i*ALIGN+j*(4096-2*i*ALIGN); \
> + b = four+i*ALIGN+j*(4096-2*i*ALIGN); \
> + memcpy(a,str1,sz); \
> + memcpy(b,str2,sz); \
> + asm(" "); \
> + r = memcmp(a,b,sz); \
> + asm(" "); \
> + if ( r < 0 && !(expect < 0) ) abort(); \
> + if ( r > 0 && !(expect > 0) ) abort(); \
> + if ( r == 0 && !(expect == 0) ) abort(); \
> + } \
> + } \
> +}
> +
> +DEF_RS(1)
> +DEF_RS(2)
> +DEF_RS(4)
> +DEF_RS(8)
> +DEF_RS(16)
> +
> +static void test_memcmp_runtime_size (const char *str1, const char *str2,
> + size_t sz, int expect)
> +{
> + char three[8192] __attribute__ ((aligned (4096)));
> + char four[8192] __attribute__ ((aligned (4096)));
> + char *a, *b;
> + int i,j,a1,a2,r;
> + test_memcmp_runtime_size_1 (str1,str2,sz,expect);
> + test_memcmp_runtime_size_2 (str1,str2,sz,expect);
> + test_memcmp_runtime_size_4 (str1,str2,sz,expect);
> + test_memcmp_runtime_size_8 (str1,str2,sz,expect);
> + test_memcmp_runtime_size_16 (str1,str2,sz,expect);
> + for (j = 0; j < 2; j++)
> + {
> + for (i = 0; i < 2; i++)
> + {
> + for (a1=0; a1 < 2*sizeof(void *); a1++)
> + {
> + for (a2=0; a2 < 2*sizeof(void *); a2++)
> + {
> + a = three+i*a1+j*(4096-2*i*a1);
> + b = four+i*a2+j*(4096-2*i*a2);
> + memcpy(a,str1,sz);
> + memcpy(b,str2,sz);
> + asm(" ");
> + r = memcmp(a,b,sz);
> + asm(" ");
> + if ( r < 0 && !(expect < 0) ) abort();
> + if ( r > 0 && !(expect > 0) ) abort();
> + if ( r == 0 && !(expect == 0) ) abort();
> + }
> + }
> + }
> + }
> +}
> +
> static void test_driver_memcmp (void (test_memcmp)(const char *, const char *, int),
> void (test_strncmp)(const char *, const char *, int),
> - size_t sz, int align)
> + size_t sz, int align)
> {
> char buf1[MAX_SZ*2+10],buf2[MAX_SZ*2+10];
> size_t test_sz = (sz<MAX_SZ)?sz:MAX_SZ;
> @@ -35,11 +104,12 @@
> buf1[j] = rand() & 0xff;
> buf2[j] = rand() & 0xff;
> }
> + e = lib_memcmp(buf1,buf2,sz);
> + (*test_memcmp)(buf1,buf2,e);
> + test_memcmp_runtime_size (buf1, buf2, sz, e);
> + e = lib_strncmp(buf1,buf2,sz);
> + (*test_strncmp)(buf1,buf2,e);
> }
> - e = lib_memcmp(buf1,buf2,sz);
> - (*test_memcmp)(buf1,buf2,e);
> - e = lib_strncmp(buf1,buf2,sz);
> - (*test_strncmp)(buf1,buf2,e);
> }
> for(diff_pos = ((test_sz>10)?(test_sz-10):0); diff_pos < test_sz+10; diff_pos++)
> for(zero_pos = ((test_sz>10)?(test_sz-10):0); zero_pos < test_sz+10; zero_pos++)
> @@ -53,6 +123,9 @@
> (*test_memcmp)(buf1,buf2,e);
> (*test_memcmp)(buf2,buf1,-e);
> (*test_memcmp)(buf2,buf2,0);
> + test_memcmp_runtime_size (buf1, buf2, sz, e);
> + test_memcmp_runtime_size (buf2, buf1, sz, -e);
> + test_memcmp_runtime_size (buf2, buf2, sz, 0);
> e = lib_strncmp(buf1,buf2,sz);
> (*test_strncmp)(buf1,buf2,e);
> (*test_strncmp)(buf2,buf1,-e);
> @@ -61,6 +134,7 @@
> buf2[diff_pos] = 0;
> e = lib_memcmp(buf1,buf2,sz);
> (*test_memcmp)(buf1,buf2,e);
> + test_memcmp_runtime_size (buf1, buf2, sz, e);
> e = lib_strncmp(buf1,buf2,sz);
> (*test_strncmp)(buf1,buf2,e);
> memset(buf2+diff_pos,'B',sizeof(buf2)-diff_pos);
> @@ -68,6 +142,8 @@
> e = lib_memcmp(buf1,buf2,sz);
> (*test_memcmp)(buf1,buf2,e);
> (*test_memcmp)(buf2,buf1,-e);
> + test_memcmp_runtime_size (buf1, buf2, sz, e);
> + test_memcmp_runtime_size (buf2, buf1, sz, -e);
> e = lib_strncmp(buf1,buf2,sz);
> (*test_strncmp)(buf1,buf2,e);
> (*test_strncmp)(buf2,buf1,-e);
> @@ -371,7 +447,14 @@
> DEF_TEST(100,4)
> DEF_TEST(100,8)
> DEF_TEST(100,16)
> +DEF_TEST(191,1)
> +DEF_TEST(192,1)
> +DEF_TEST(193,1)
> +DEF_TEST(200,1)
> +DEF_TEST(400,1)
> #else
> +DEF_TEST(1,1)
> +DEF_TEST(2,1)
> DEF_TEST(3,1)
> DEF_TEST(4,1)
> DEF_TEST(5,1)
> @@ -389,6 +472,8 @@
> DEF_TEST(32,1)
> DEF_TEST(100,1)
> DEF_TEST(100,8)
> +DEF_TEST(180,1)
> +DEF_TEST(180,8)
> #endif
>
> int
> @@ -395,7 +480,7 @@
> main(int argc, char **argv)
> {
> #ifdef TEST_ALL
> - RUN_TEST(1,1)
> + RUN_TEST(1,1)
> RUN_TEST(1,2)
> RUN_TEST(1,4)
> RUN_TEST(1,8)
> @@ -645,7 +730,14 @@
> RUN_TEST(100,4)
> RUN_TEST(100,8)
> RUN_TEST(100,16)
> + RUN_TEST(191,1)
> + RUN_TEST(192,1)
> + RUN_TEST(193,1)
> + RUN_TEST(200,1)
> + RUN_TEST(400,1)
> #else
> + RUN_TEST(1,1)
> + RUN_TEST(2,1)
> RUN_TEST(3,1)
> RUN_TEST(4,1)
> RUN_TEST(5,1)
> @@ -663,5 +755,7 @@
> RUN_TEST(32,1)
> RUN_TEST(100,1)
> RUN_TEST(100,8)
> + RUN_TEST(180,1)
> + RUN_TEST(180,8)
> #endif
> }