[Mesa-dev] [PATCH 2/5] i965/vec4: adding vec4_cmod_propagation optimization
Alejandro Piñeiro
apinheiro at igalia.com
Sat Oct 10 13:57:37 PDT 2015
On 10/10/15 16:54, Jason Ekstrand 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
> Would you mind cherry-picking this back onto
> 4e0a8e0a50c9ac91cb7a70b92b8d9c6fcc02b7aa (the commit right before we
> made NIR non-optional) and get some GLSL IR vs. NIR vec4-only numbers
> with this patch? I'd like to know what it does to that delta as well.
FWIW, the previous shader-db numbers were done without grepping for
vec4. Matt mentioned that he preferred that way. As asked, the numbers
for this email will be vec4-only numbers (so grepping for vec4).
So, the shader-db numbers of IR vs NIR at that reference commit are the
following:
total instructions in shared programs: 1848027 -> 1648216 (-10.81%)
instructions in affected programs: 1660279 -> 1460468 (-12.03%)
helped: 14668
HURT: 1369
And IR vs NIR numbers cherry-picking the optimization are the following:
total instructions in shared programs: 1845902 -> 1631459 (-11.62%)
instructions in affected programs: 1663398 -> 1448955 (-12.89%)
helped: 14863
HURT: 1203
FWIW, we need to take into account that this commit is also helping IR.
The shader-db numbers of IR reference vs IR cherry picking are the
following:
total instructions in shared programs: 1848027 -> 1845902 (-0.11%)
instructions in affected programs: 195042 -> 192917 (-1.09%)
helped: 1033
HURT: 0
And for that reason, probably it is worth to compare IR at the reference
versus NIR results cherry-picking, that are the following:
total instructions in shared programs: 1848027 -> 1631459 (-11.72%)
instructions in affected programs: 1672237 -> 1455669 (-12.95%)
helped: 14955
HURT: 1194
>
> Thanks!
> --Jason
You are welcome. Thanks for the patch reviewing.
Best regards
>
>> ---
>>
>> 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)) {
>> +
>> + 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;
>> + }
>> +
>> + /* CMP's result is the same regardless of dest type. */
>> + if (inst->conditional_mod == BRW_CONDITIONAL_NZ &&
>> + scan_inst->opcode == BRW_OPCODE_CMP &&
>> + (inst->dst.type == BRW_REGISTER_TYPE_D ||
>> + inst->dst.type == BRW_REGISTER_TYPE_UD)) {
>> + inst->remove(block);
>> + progress = true;
>> + break;
>> + }
>> +
>> + /* If the AND wasn't handled by the previous case, it isn't safe
>> + * to remove it.
>> + */
>> + if (inst->opcode == BRW_OPCODE_AND)
>> + break;
>> +
>> + /* Comparisons operate differently for ints and floats */
>> + if (scan_inst->dst.type != inst->dst.type &&
>> + (scan_inst->dst.type == BRW_REGISTER_TYPE_F ||
>> + inst->dst.type == BRW_REGISTER_TYPE_F))
>> + break;
>> +
>> + /* If the instruction generating inst's source also wrote the
>> + * flag, and inst is doing a simple .nz comparison, then inst
>> + * is redundant - the appropriate value is already in the flag
>> + * register. Delete inst.
>> + */
>> + if (inst->conditional_mod == BRW_CONDITIONAL_NZ &&
>> + !inst->src[0].negate &&
>> + scan_inst->writes_flag()) {
>> + inst->remove(block);
>> + progress = true;
>> + break;
>> + }
>> +
>> + /* Otherwise, try propagating the conditional. */
>> + enum brw_conditional_mod cond =
>> + inst->src[0].negate ? brw_swap_cmod(inst->conditional_mod)
>> + : inst->conditional_mod;
>> +
>> + if (scan_inst->can_do_cmod() &&
>> + ((!read_flag && scan_inst->conditional_mod == BRW_CONDITIONAL_NONE) ||
>> + scan_inst->conditional_mod == cond)) {
>> + scan_inst->conditional_mod = cond;
>> + inst->remove(block);
>> + progress = true;
>> + }
>> + break;
>> + }
>> +
>> + if (scan_inst->writes_flag())
>> + break;
>> +
>> + read_flag = read_flag || scan_inst->reads_flag();
>> + }
>> + }
>> +
>> + return progress;
>> +}
>> +
>> +bool
>> +vec4_visitor::opt_cmod_propagation()
>> +{
>> + bool progress = false;
>> +
>> + foreach_block_reverse(block, cfg) {
>> + progress = opt_cmod_propagation_local(block) || progress;
>> + }
>> +
>> + if (progress)
>> + invalidate_live_intervals();
>> +
>> + return progress;
>> +}
>> +
>> +} /* namespace brw */
>> --
>> 2.1.4
>>
>> _______________________________________________
>> mesa-dev mailing list
>> mesa-dev at lists.freedesktop.org
>> http://lists.freedesktop.org/mailman/listinfo/mesa-dev
--
Alejandro Piñeiro (apinheiro at igalia.com)
More information about the mesa-dev
mailing list