| Message ID | 1468517497-7859-1-git-send-email-damien.lespiau@intel.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Thu, Jul 14, 2016 at 06:31:37PM +0100, Damien Lespiau wrote: > Depending how the gcc was compiled it may be necessary to enable SSE4 > instructions explicitly. Otherwise: > > CC gem_gtt_speed.o > In file included from gem_gtt_speed.c:54:0: > /usr/lib/gcc/x86_64-linux-gnu/4.8/include/smmintrin.h:31:3: error: #error "SSE4.1 instruction set not enabled" > # error "SSE4.1 instruction set not enabled" So the challenge is getting the function attribute applied to the include. Ah, can you try #pragma GCC target ("sse4.1") in those blocks instead? > @@ -76,6 +76,7 @@ gem_ctx_basic_CFLAGS = $(AM_CFLAGS) $(THREAD_CFLAGS) > gem_ctx_basic_LDADD = $(LDADD) -lpthread > gem_ctx_thrash_CFLAGS = $(AM_CFLAGS) $(THREAD_CFLAGS) > gem_ctx_thrash_LDADD = $(LDADD) -lpthread > +gem_exec_flush_CFLAGS = $(AM_CFLAGS) -msse4 Note, it should be -msse4.1 -Chris
On Thu, Jul 14, 2016 at 06:44:59PM +0100, Chris Wilson wrote: > On Thu, Jul 14, 2016 at 06:31:37PM +0100, Damien Lespiau wrote: > > Depending how the gcc was compiled it may be necessary to enable SSE4 > > instructions explicitly. Otherwise: > > > > CC gem_gtt_speed.o > > In file included from gem_gtt_speed.c:54:0: > > /usr/lib/gcc/x86_64-linux-gnu/4.8/include/smmintrin.h:31:3: error: #error "SSE4.1 instruction set not enabled" > > # error "SSE4.1 instruction set not enabled" > > So the challenge is getting the function attribute applied to the > include. > > Ah, can you try > #pragma GCC target ("sse4.1") > in those blocks instead? Oh, and we probably need an #if GCC > 4.y to be fully backwards compatible. -Chris
On Thu, Jul 14, 2016 at 06:48:26PM +0100, Chris Wilson wrote: > On Thu, Jul 14, 2016 at 06:44:59PM +0100, Chris Wilson wrote: > > On Thu, Jul 14, 2016 at 06:31:37PM +0100, Damien Lespiau wrote: > > > Depending how the gcc was compiled it may be necessary to enable SSE4 > > > instructions explicitly. Otherwise: > > > > > > CC gem_gtt_speed.o > > > In file included from gem_gtt_speed.c:54:0: > > > /usr/lib/gcc/x86_64-linux-gnu/4.8/include/smmintrin.h:31:3: error: #error "SSE4.1 instruction set not enabled" > > > # error "SSE4.1 instruction set not enabled" > > > > So the challenge is getting the function attribute applied to the > > include. > > > > Ah, can you try > > #pragma GCC target ("sse4.1") > > in those blocks instead? Yup, that also seems to enable sse4.1 for the rest of the compilation unit. > Oh, and we probably need an #if GCC > 4.y to be fully backwards > compatible. Couldn't find in less than 5 mins this information, gave up. Someone exhibiting the problem will have to fix it :)
diff --git a/tests/Makefile.am b/tests/Makefile.am index 102b8a6..8c6b0a3 100644 --- a/tests/Makefile.am +++ b/tests/Makefile.am @@ -76,6 +76,7 @@ gem_ctx_basic_CFLAGS = $(AM_CFLAGS) $(THREAD_CFLAGS) gem_ctx_basic_LDADD = $(LDADD) -lpthread gem_ctx_thrash_CFLAGS = $(AM_CFLAGS) $(THREAD_CFLAGS) gem_ctx_thrash_LDADD = $(LDADD) -lpthread +gem_exec_flush_CFLAGS = $(AM_CFLAGS) -msse4 gem_exec_parallel_CFLAGS = $(AM_CFLAGS) $(THREAD_CFLAGS) gem_exec_parallel_LDADD = $(LDADD) -lpthread gem_fence_thrash_CFLAGS = $(AM_CFLAGS) $(THREAD_CFLAGS) @@ -84,6 +85,7 @@ gem_fence_upload_CFLAGS = $(AM_CFLAGS) $(THREAD_CFLAGS) gem_fence_upload_LDADD = $(LDADD) -lpthread gem_flink_race_CFLAGS = $(AM_CFLAGS) $(THREAD_CFLAGS) gem_flink_race_LDADD = $(LDADD) -lpthread +gem_gtt_speed_CFLAGS = $(AM_CFLAGS) -msse4 gem_mmap_gtt_CFLAGS = $(AM_CFLAGS) $(THREAD_CFLAGS) gem_mmap_gtt_LDADD = $(LDADD) -lpthread gem_mmap_wc_CFLAGS = $(AM_CFLAGS) $(THREAD_CFLAGS)
Depending how the gcc was compiled it may be necessary to enable SSE4 instructions explicitly. Otherwise: CC gem_gtt_speed.o In file included from gem_gtt_speed.c:54:0: /usr/lib/gcc/x86_64-linux-gnu/4.8/include/smmintrin.h:31:3: error: #error "SSE4.1 instruction set not enabled" # error "SSE4.1 instruction set not enabled" ^ gem_gtt_speed.c: In function ‘streaming_load’: gem_gtt_speed.c:59:2: error: unknown type name ‘__m128i’ __m128i tmp, *s = src; ^ gem_gtt_speed.c:65:3: error: implicit declaration of function ‘_mm_stream_load_si128’ [-Werror=implicit-function-declaration] tmp += _mm_stream_load_si128(s++); ^ gem_gtt_speed.c:65:3: warning: nested extern declaration of ‘_mm_stream_load_si128’ [-Wnested-externs] gem_gtt_speed.c:70:2: error: unknown type name ‘__m128i’ *(volatile __m128i *)src = tmp; ^ CC: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by: Damien Lespiau <damien.lespiau@intel.com> --- tests/Makefile.am | 2 ++ 1 file changed, 2 insertions(+)