[Mesa-dev] [PATCH V4] mesa: add SSE optimisation for glDrawElements

Ian Romanick idr at freedesktop.org
Mon Oct 27 10:36:26 PDT 2014


On 10/27/2014 08:04 AM, Bruno Jimenez wrote:
> On Mon, 2014-10-27 at 08:37 +1100, Timothy Arceri wrote:
>> Makes use of SSE to speed up compute of min and max elements
>>
>> Callgrind cpu usage results from pts benchmarks:
>>
>> Openarena 0.8.8: 3.67% -> 1.03%
>> UrbanTerror: 2.36% -> 0.81%
>>
>> Signed-off-by: Timothy Arceri <t_arceri at yahoo.com.au>
>> ---
>>  src/mesa/Android.libmesa_dricore.mk |  3 +-
>>  src/mesa/Makefile.am                |  3 +-
>>  src/mesa/main/sse_minmax.c          | 93 +++++++++++++++++++++++++++++++++++++
>>  src/mesa/main/sse_minmax.h          | 30 ++++++++++++
>>  src/mesa/vbo/vbo_exec_array.c       | 14 ++++--
>>  5 files changed, 138 insertions(+), 5 deletions(-)
>>  create mode 100644 src/mesa/main/sse_minmax.c
>>  create mode 100644 src/mesa/main/sse_minmax.h
>>
>> V4:
>> - fixed bugs with incrementing pointer and updating counters
>>
>> V3:
>>
>> - Removed sse_minmax.c from Makefile.sources
>> - handle the first few values without SSE until the pointer is aligned
>>  and use _mm_load_si128 rather than _mm_loadu_si128
>> - guard the call to the SSE code better at build time

All of the V# information should be captured in the actual commit message.

>> diff --git a/src/mesa/Android.libmesa_dricore.mk b/src/mesa/Android.libmesa_dricore.mk
>> index 1e6d948..52d626f 100644
>> --- a/src/mesa/Android.libmesa_dricore.mk
>> +++ b/src/mesa/Android.libmesa_dricore.mk
>> @@ -51,7 +51,8 @@ endif # MESA_ENABLE_ASM
>>  
>>  ifeq ($(ARCH_X86_HAVE_SSE4_1),true)
>>  LOCAL_SRC_FILES += \
>> -	$(SRCDIR)main/streaming-load-memcpy.c
>> +	$(SRCDIR)main/streaming-load-memcpy.c \
>> +	$(SRCDIR)main/sse_minmax.c
>>  LOCAL_CFLAGS := -msse4.1
>>  endif
>>  
>> diff --git a/src/mesa/Makefile.am b/src/mesa/Makefile.am
>> index e71bccb..932db4f 100644
>> --- a/src/mesa/Makefile.am
>> +++ b/src/mesa/Makefile.am
>> @@ -151,7 +151,8 @@ libmesagallium_la_LIBADD = \
>>  	$(ARCH_LIBS)
>>  
>>  libmesa_sse41_la_SOURCES = \
>> -	main/streaming-load-memcpy.c
>> +	main/streaming-load-memcpy.c \
>> +	main/sse_minmax.c
>>  libmesa_sse41_la_CFLAGS = $(AM_CFLAGS) -msse4.1
>>  
>>  pkgconfigdir = $(libdir)/pkgconfig
>> diff --git a/src/mesa/main/sse_minmax.c b/src/mesa/main/sse_minmax.c
>> new file mode 100644
>> index 0000000..6bf5fe3
>> --- /dev/null
>> +++ b/src/mesa/main/sse_minmax.c
>> @@ -0,0 +1,93 @@
>> +/*
>> + * Copyright © 2014 Timothy Arceri
>> + *
>> + * Permission is hereby granted, free of charge, to any person obtaining a
>> + * copy of this software and associated documentation files (the "Software"),
>> + * to deal in the Software without restriction, including without limitation
>> + * the rights to use, copy, modify, merge, publish, distribute, sublicense,
>> + * and/or sell copies of the Software, and to permit persons to whom the
>> + * Software is furnished to do so, subject to the following conditions:
>> + *
>> + * The above copyright notice and this permission notice (including the next
>> + * paragraph) shall be included in all copies or substantial portions of the
>> + * Software.
>> + *
>> + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
>> + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
>> + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
>> + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
>> + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
>> + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
>> + * IN THE SOFTWARE.
>> + *
>> + * Author:
>> + *    Timothy Arceri <t_arceri at yahoo.com.au>
>> + *
>> + */
>> +
>> +#ifdef __SSE4_1__
>> +#include "main/sse_minmax.h"
>> +#include <smmintrin.h>
>> +#include <stdint.h>
>> +
>> +void
>> +_mesa_uint_array_min_max(const unsigned *ui_indices, unsigned *min_index,
>> +                         unsigned *max_index, const unsigned count)
>> +{
>> +   unsigned max_ui = 0;
>> +   unsigned min_ui = ~0U;
>> +   unsigned i = 0;
>> +   unsigned aligned_count = count;
>> +
>> +   /* handle the first few values without SSE until the pointer is aligned */
>> +   while (((uintptr_t)ui_indices & 15) && aligned_count) {
>> +      if (*ui_indices > max_ui)
>> +         max_ui = *ui_indices;
>> +      if (*ui_indices < min_ui)
>> +         min_ui = *ui_indices;
>> +
>> +      aligned_count--;
>> +      ui_indices++;
>> +   }
>> +
>> +   if (aligned_count >= 4) {
>                           ^^
> 
> Hi,
> 
> I have been thinking and I think that you can change that 4 for an 8. In
> the case aligned_count == 4 there's no gain in using SSE, as you will
> have to do a final reduction from 4 elements to 1.

Either that or the initial check for whether or not to call the SSE
version could be predicated on size.  Something like

        if (cpu_has_sse4_1 && count >= 8) {

The actual threshold may be higher than 8 (and is certainly higher than
4, as you point out).  It would take some careful microbenchmarks and
measurement to find the actual tipping point.  Maybe just use 8 and add
a FINISHME comment for now?

> - Bruno
> 
>> +      unsigned max_arr[4] __attribute__ ((aligned (16)));
>> +      unsigned min_arr[4] __attribute__ ((aligned (16)));
>> +      unsigned vec_count;
>> +      __m128i max_ui4 = _mm_setzero_si128();
>> +      __m128i min_ui4 = _mm_set1_epi32(~0U);
>> +      __m128i ui_indices4;
>> +      __m128i *ui_indices_ptr;
>> +
>> +      vec_count = aligned_count & ~0x3;
>> +      ui_indices_ptr = (__m128i*)ui_indices;
>> +      for (i = 0; i < vec_count / 4; i++) {
>> +         ui_indices4 = _mm_load_si128(&ui_indices_ptr[i]);
>> +         max_ui4 = _mm_max_epu32(ui_indices4, max_ui4);
>> +         min_ui4 = _mm_min_epu32(ui_indices4, min_ui4);
>> +      }
>> +
>> +      _mm_store_si128((__m128i*)max_arr, max_ui4);
>> +      _mm_store_si128((__m128i*)min_arr, min_ui4);
>> +
>> +      for (i = 0; i < 4; i++) {
>> +         if (max_arr[i] > max_ui)
>> +            max_ui = max_arr[i];
>> +         if (min_arr[i] < min_ui)
>> +            min_ui = min_arr[i];
>> +      }
>> +      i = vec_count;
>> +   }
>> +
>> +   for (; i < aligned_count; i++) {
>> +      if (ui_indices[i] > max_ui)
>> +         max_ui = ui_indices[i];
>> +      if (ui_indices[i] < min_ui)
>> +         min_ui = ui_indices[i];
>> +   }
>> +
>> +   *min_index = min_ui;
>> +   *max_index = max_ui;
>> +}
>> +
>> +#endif
>> diff --git a/src/mesa/main/sse_minmax.h b/src/mesa/main/sse_minmax.h
>> new file mode 100644
>> index 0000000..953c4e9
>> --- /dev/null
>> +++ b/src/mesa/main/sse_minmax.h
>> @@ -0,0 +1,30 @@
>> +/*
>> + * Copyright © 2014 Timothy Arceri
>> + *
>> + * Permission is hereby granted, free of charge, to any person obtaining a
>> + * copy of this software and associated documentation files (the "Software"),
>> + * to deal in the Software without restriction, including without limitation
>> + * the rights to use, copy, modify, merge, publish, distribute, sublicense,
>> + * and/or sell copies of the Software, and to permit persons to whom the
>> + * Software is furnished to do so, subject to the following conditions:
>> + *
>> + * The above copyright notice and this permission notice (including the next
>> + * paragraph) shall be included in all copies or substantial portions of the
>> + * Software.
>> + *
>> + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
>> + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
>> + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
>> + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
>> + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
>> + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
>> + * IN THE SOFTWARE.
>> + *
>> + * Author:
>> + *    Timothy Arceri <t_arceri at yahoo.com.au>
>> + *
>> + */
>> +
>> +void
>> +_mesa_uint_array_min_max(const unsigned *ui_indices, unsigned *min_index,
>> +                         unsigned *max_index, const unsigned count);
>> diff --git a/src/mesa/vbo/vbo_exec_array.c b/src/mesa/vbo/vbo_exec_array.c
>> index 045dbb5..e623b36 100644
>> --- a/src/mesa/vbo/vbo_exec_array.c
>> +++ b/src/mesa/vbo/vbo_exec_array.c
>> @@ -36,6 +36,8 @@
>>  #include "main/enums.h"
>>  #include "main/macros.h"
>>  #include "main/transformfeedback.h"
>> +#include "main/sse_minmax.h"
>> +#include "x86/common_x86_asm.h"
>>  
>>  #include "vbo_context.h"
>>  
>> @@ -119,10 +121,16 @@ vbo_get_minmax_index(struct gl_context *ctx,
>>           }
>>        }
>>        else {
>> -         for (i = 0; i < count; i++) {
>> -            if (ui_indices[i] > max_ui) max_ui = ui_indices[i];
>> -            if (ui_indices[i] < min_ui) min_ui = ui_indices[i];
>> +#if defined(USE_SSE41)
>> +         if (cpu_has_sse4_1) {
>> +            _mesa_uint_array_min_max(ui_indices, &min_ui, &max_ui, count);
>>           }
>> +         else
>> +#endif
>> +            for (i = 0; i < count; i++) {
>> +               if (ui_indices[i] > max_ui) max_ui = ui_indices[i];
>> +               if (ui_indices[i] < min_ui) min_ui = ui_indices[i];
>> +            }
>>        }
>>        *min_index = min_ui;
>>        *max_index = max_ui;
> 
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/mesa-dev



More information about the mesa-dev mailing list