[Mesa-dev] [PATCH 2/5] i965/vec4: adding vec4_cmod_propagation optimization

Matt Turner mattst88 at gmail.com
Mon Oct 12 18:10:08 PDT 2015


On Mon, Oct 12, 2015 at 4:25 PM, Matt Turner <mattst88 at gmail.com> wrote:
> On Sat, Oct 10, 2015 at 4:24 AM, Alejandro Piñeiro <apinheiro at igalia.com> wrote:
>> vec4 port of fs_cmod_propagation.
>>
>> Shader-db results:
>> total instructions in shared programs: 6241226 -> 6224469 (-0.27%)
>> instructions in affected programs:     498213 -> 481456 (-3.36%)
>> helped:                                3082
>> HURT:                                  0
>> ---
>>
>> The final outcome is really similar to fs_brw_cmod_propagation. In
>> fact the only difference is that on fs we have this:
>>          if (scan_inst->overwrites_reg(inst->src[0])) {
>>             if (scan_inst->is_partial_write() ||
>>                 scan_inst->dst.reg_offset != inst->src[0].reg_offset)
>>                break;
>>
>> And on vec4 (this commit) we have this:
>>          if (inst->src[0].in_range(scan_inst->dst,
>>                                    scan_inst->regs_written)) {
>>
>>             if ((scan_inst->predicate && scan_inst->opcode != BRW_OPCODE_SEL) ||
>>                 scan_inst->dst.reg_offset != inst->src[0].reg_offset ||
>>                 (scan_inst->dst.writemask != WRITEMASK_X && scan_inst->dst.writemask != WRITEMASK_XYZW))
>>                break;
>>
>>             if (scan_inst->dst.writemask == WRITEMASK_XYZW &&
>>                 inst->src[0].swizzle != BRW_SWIZZLE_XYZW) {
>>                break;
>>             }
>>
>> So at some point I thought about refactoring it and having one common,
>> like with opt_predicated_break, but that one was possible with just
>> backend_instructions, while here we would need to deal with
>> vec4_instructions and fs_inst, that could be somewhat messy, so
>> I'm leaving this as it is.
>>
>>  src/mesa/drivers/dri/i965/Makefile.sources         |   1 +
>>  src/mesa/drivers/dri/i965/brw_vec4.cpp             |   1 +
>>  src/mesa/drivers/dri/i965/brw_vec4.h               |   1 +
>>  .../drivers/dri/i965/brw_vec4_cmod_propagation.cpp | 163 +++++++++++++++++++++
>>  4 files changed, 166 insertions(+)
>>  create mode 100644 src/mesa/drivers/dri/i965/brw_vec4_cmod_propagation.cpp
>>
>> diff --git a/src/mesa/drivers/dri/i965/Makefile.sources b/src/mesa/drivers/dri/i965/Makefile.sources
>> index 81ef628..c1836d6 100644
>> --- a/src/mesa/drivers/dri/i965/Makefile.sources
>> +++ b/src/mesa/drivers/dri/i965/Makefile.sources
>> @@ -56,6 +56,7 @@ i965_compiler_FILES = \
>>         brw_util.c \
>>         brw_util.h \
>>         brw_vec4_builder.h \
>> +       brw_vec4_cmod_propagation.cpp \
>>         brw_vec4_copy_propagation.cpp \
>>         brw_vec4.cpp \
>>         brw_vec4_cse.cpp \
>> diff --git a/src/mesa/drivers/dri/i965/brw_vec4.cpp b/src/mesa/drivers/dri/i965/brw_vec4.cpp
>> index e966b96..55e381b 100644
>> --- a/src/mesa/drivers/dri/i965/brw_vec4.cpp
>> +++ b/src/mesa/drivers/dri/i965/brw_vec4.cpp
>> @@ -1867,6 +1867,7 @@ vec4_visitor::run()
>>        OPT(dead_code_eliminate);
>>        OPT(dead_control_flow_eliminate, this);
>>        OPT(opt_copy_propagation);
>> +      OPT(opt_cmod_propagation);
>>        OPT(opt_cse);
>>        OPT(opt_algebraic);
>>        OPT(opt_register_coalesce);
>> diff --git a/src/mesa/drivers/dri/i965/brw_vec4.h b/src/mesa/drivers/dri/i965/brw_vec4.h
>> index 5e3500c..3c1711d 100644
>> --- a/src/mesa/drivers/dri/i965/brw_vec4.h
>> +++ b/src/mesa/drivers/dri/i965/brw_vec4.h
>> @@ -149,6 +149,7 @@ public:
>>     int var_range_start(unsigned v, unsigned n) const;
>>     int var_range_end(unsigned v, unsigned n) const;
>>     bool virtual_grf_interferes(int a, int b);
>> +   bool opt_cmod_propagation();
>>     bool opt_copy_propagation(bool do_constant_prop = true);
>>     bool opt_cse_local(bblock_t *block);
>>     bool opt_cse();
>> diff --git a/src/mesa/drivers/dri/i965/brw_vec4_cmod_propagation.cpp b/src/mesa/drivers/dri/i965/brw_vec4_cmod_propagation.cpp
>> new file mode 100644
>> index 0000000..7e39d2b
>> --- /dev/null
>> +++ b/src/mesa/drivers/dri/i965/brw_vec4_cmod_propagation.cpp
>> @@ -0,0 +1,163 @@
>> +/*
>> + * Copyright © 2015 Intel Corporation
>> + *
>> + * 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.
>> + *
>> + * Authors:
>> + *    Alejandro Piñeiro Iglesias <apinheiro at igalia.com>
>> + *
>> + * Based on brw_fs_cmod_propagation.cpp
>> + */
>> +
>> +/** @file brw_vec4_cmod_propagation.cpp
>> + *
>> + * Really similar to brw_fs_cmod_propagation but adapted to vec4 needs. Check
>> + * brw_fs_cmod_propagation for further details on the rationale behind this
>> + * optimization.
>> + */
>> +
>> +#include "brw_vec4.h"
>> +#include "brw_cfg.h"
>> +
>> +namespace brw {
>> +
>> +static bool
>> +opt_cmod_propagation_local(bblock_t *block)
>> +{
>> +   bool progress = false;
>> +   int ip = block->end_ip + 1;
>> +
>> +   foreach_inst_in_block_reverse_safe(vec4_instruction, inst, block) {
>> +      ip--;
>> +
>> +      if ((inst->opcode != BRW_OPCODE_AND &&
>> +           inst->opcode != BRW_OPCODE_CMP &&
>> +           inst->opcode != BRW_OPCODE_MOV) ||
>> +          inst->predicate != BRW_PREDICATE_NONE ||
>> +          !inst->dst.is_null() ||
>> +          inst->src[0].file != GRF ||
>> +          inst->src[0].abs)
>> +         continue;
>> +
>> +      if (inst->opcode == BRW_OPCODE_AND &&
>> +          !(inst->src[1].is_one() &&
>> +            inst->conditional_mod == BRW_CONDITIONAL_NZ &&
>> +            !inst->src[0].negate))
>> +         continue;
>> +
>> +      if (inst->opcode == BRW_OPCODE_CMP && !inst->src[1].is_zero())
>> +         continue;
>> +
>> +      if (inst->opcode == BRW_OPCODE_MOV &&
>> +          inst->conditional_mod != BRW_CONDITIONAL_NZ)
>> +         continue;
>> +
>> +      bool read_flag = false;
>> +      foreach_inst_in_block_reverse_starting_from(vec4_instruction, scan_inst, inst,
>> +                                                  block) {
>> +         if (inst->src[0].in_range(scan_inst->dst,
>> +                                   scan_inst->regs_written)) {
>> +
>
> Extra new line.
>
>> +            if ((scan_inst->predicate && scan_inst->opcode != BRW_OPCODE_SEL) ||
>> +                scan_inst->dst.reg_offset != inst->src[0].reg_offset ||
>> +                (scan_inst->dst.writemask != WRITEMASK_X && scan_inst->dst.writemask != WRITEMASK_XYZW))
>
> Line wrap this.
>
>> +               break;
>> +
>> +            if (scan_inst->dst.writemask == WRITEMASK_XYZW &&
>> +                inst->src[0].swizzle != BRW_SWIZZLE_XYZW) {
>> +               break;
>> +            }
>
> I'd just combine both of these blocks:
>
>    if ((scan_inst->predicate && scan_inst->opcode != BRW_OPCODE_SEL) ||
>        scan_inst->dst.reg_offset != inst->src[0].reg_offset ||
>        (scan_inst->dst.writemask != WRITEMASK_X &&
>         scan_inst->dst.writemask != WRITEMASK_XYZW) ||
>        (scan_inst->dst.writemask == WRITEMASK_XYZW &&
>         inst->src[0].swizzle != BRW_SWIZZLE_XYZW))
>       break;
>
> But I believe it's correct (if maybe conservative).
>
> Reviewed-by: Matt Turner <mattst88 at gmail.com>

Looks like this is causing an intermittent failure on HSW in our
Jenkins system (but I'm not able to reproduce locally) and a
consistent failure on G45. I'll investigate that.


More information about the mesa-dev mailing list