On 5 November 2013 14:58, Paul Berry <stereotype...@gmail.com> wrote:
> On 1 November 2013 17:31, Matt Turner <matts...@gmail.com> wrote: > >> total instructions in shared programs: 1649485 -> 1649157 (-0.02%) >> instructions in affected programs: 7823 -> 7495 (-4.19%) >> --- >> src/mesa/drivers/dri/i965/Makefile.sources | 1 + >> src/mesa/drivers/dri/i965/brw_fs.cpp | 1 + >> src/mesa/drivers/dri/i965/brw_fs.h | 1 + >> .../dri/i965/brw_fs_peephole_predicated_break.cpp | 96 >> ++++++++++++++++++++++ >> 4 files changed, 99 insertions(+) >> create mode 100644 >> src/mesa/drivers/dri/i965/brw_fs_peephole_predicated_break.cpp >> >> diff --git a/src/mesa/drivers/dri/i965/Makefile.sources >> b/src/mesa/drivers/dri/i965/Makefile.sources >> index 14b2f61..fcc6fdb 100644 >> --- a/src/mesa/drivers/dri/i965/Makefile.sources >> +++ b/src/mesa/drivers/dri/i965/Makefile.sources >> @@ -59,6 +59,7 @@ i965_FILES = \ >> brw_fs_fp.cpp \ >> brw_fs_generator.cpp \ >> brw_fs_live_variables.cpp \ >> + brw_fs_peephole_predicated_break.cpp \ >> brw_fs_reg_allocate.cpp \ >> brw_fs_vector_splitting.cpp \ >> brw_fs_visitor.cpp \ >> diff --git a/src/mesa/drivers/dri/i965/brw_fs.cpp >> b/src/mesa/drivers/dri/i965/brw_fs.cpp >> index 7064910..5c8443f 100644 >> --- a/src/mesa/drivers/dri/i965/brw_fs.cpp >> +++ b/src/mesa/drivers/dri/i965/brw_fs.cpp >> @@ -3149,6 +3149,7 @@ fs_visitor::run() >> progress = opt_algebraic() || progress; >> progress = opt_cse() || progress; >> progress = opt_copy_propagate() || progress; >> + progress = opt_peephole_predicated_break() || progress; >> progress = dead_code_eliminate() || progress; >> progress = dead_code_eliminate_local() || progress; >> progress = register_coalesce() || progress; >> diff --git a/src/mesa/drivers/dri/i965/brw_fs.h >> b/src/mesa/drivers/dri/i965/brw_fs.h >> index 43e4761..09e5174 100644 >> --- a/src/mesa/drivers/dri/i965/brw_fs.h >> +++ b/src/mesa/drivers/dri/i965/brw_fs.h >> @@ -362,6 +362,7 @@ public: >> bool try_emit_saturate(ir_expression *ir); >> bool try_emit_mad(ir_expression *ir, int mul_arg); >> void try_replace_with_sel(); >> + bool opt_peephole_predicated_break(); >> void emit_bool_to_cond_code(ir_rvalue *condition); >> void emit_if_gen6(ir_if *ir); >> void emit_unspill(fs_inst *inst, fs_reg reg, uint32_t spill_offset, >> diff --git >> a/src/mesa/drivers/dri/i965/brw_fs_peephole_predicated_break.cpp >> b/src/mesa/drivers/dri/i965/brw_fs_peephole_predicated_break.cpp >> new file mode 100644 >> index 0000000..7845113 >> --- /dev/null >> +++ b/src/mesa/drivers/dri/i965/brw_fs_peephole_predicated_break.cpp >> @@ -0,0 +1,96 @@ >> +/* >> + * Copyright © 2013 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. >> + */ >> + >> +#include "brw_fs.h" >> +#include "brw_cfg.h" >> + >> +/** @file brw_fs_peephole_predicated_break.cpp >> + * >> + * Loops are often structured as >> + * >> + * loop: >> + * CMP.f0 >> + * (+f0) IF >> + * BREAK >> + * ENDIF >> + * ... >> + * WHILE loop >> + * >> + * This peephole pass removes the IF and ENDIF instructions and >> predicates the >> + * BREAK, dropping two instructions from the loop body. >> > > Instead of this special-purpose peephole optimization, I'm wondering if it > would be more useful to create a more general optimization pass that just > converts the entire IF block to predicated instructions, assuming certain > conditions hold. Those conditions would be something like: > > - If any instruction but the last writes to the flag register, then don't > do the optimization. > - If there's nested control flow, then don't do the optimization. > - If any instruction inside the IF block is incompatible with predication > (are there any such instructions? I don't know) then don't do the > optimization. > - If the block is larger than a certain heuristically-determined size, > then don't do the optimization. > > Such an optimization could in principle even handle IF/ELSE blocks, by > inverting the predicate in the ELSE section. > Not to pressure you or anything, but I happened to be looking at the geometry shader for orbital-explorer this evening, and there are a bunch of examples of code like this: 0x00000140: cmp.l.f0(8) null g51<4,4,1>.wF g1<0,4,1>.xF { align16 WE_normal 1Q switch }; 0x00000150: (+f0) if(8) 0 0 null 0x00000000UD { align16 WE_normal 1Q switch }; 0x00000160: add(8) g45<1>.xD g45<4,4,1>.xD 1D { align16 WE_normal 1Q }; 0x00000170: endif(8) 2 null 0x00000002UD { align16 WE_normal 1Q switch }; which would be a perfect candidate for the more general-purpose optimization I was proposing. (granted it's a geometry shader, so it's less interesting to optimize since most programs are fragment limited). > > > Having said all that, I think this optimization pass does what you intend, > so with the exception of one comment I've noted below, the patch is: > > Reviewed-by: Paul Berry <stereotype...@gmail.com> > > >> + */ >> + >> +bool >> +fs_visitor::opt_peephole_predicated_break() >> +{ >> + bool progress = false; >> + >> + cfg_t cfg(this); >> + >> + for (int b = 0; b < cfg.num_blocks; b++) { >> + bblock_t *block = cfg.blocks[b]; >> + >> + /* IF instructions, by definition, can only be found at the ends of >> + * basic blocks. >> + */ >> + fs_inst *if_inst = (fs_inst *) block->end; >> + if (if_inst->opcode != BRW_OPCODE_IF) >> + continue; >> + >> + fs_inst *inst = (fs_inst *) if_inst->next; >> + if (inst->opcode != BRW_OPCODE_BREAK && inst->opcode != >> BRW_OPCODE_CONTINUE) >> + continue; >> + >> + fs_inst *endif_inst = (fs_inst *) inst->next; >> + if (endif_inst->opcode != BRW_OPCODE_ENDIF) >> + continue; >> + >> + /* For Sandybridge with IF with embedded comparison we need to >> emit an >> + * instruction to set the flag register. >> + */ >> + if (brw->gen == 6 && if_inst->conditional_mod) { >> + fs_inst *cmp_inst = CMP(reg_null_d, if_inst->src[0], >> if_inst->src[1], >> + if_inst->conditional_mod); >> + if_inst->insert_before(cmp_inst); >> + inst->predicate = BRW_PREDICATE_NORMAL; >> + } else { >> + inst->predicate = if_inst->predicate; >> + inst->predicate_inverse = if_inst->predicate_inverse; >> + } >> + >> + if_inst->remove(); >> + endif_inst->remove(); >> + >> + /* By removing the ENDIF instruction we removed a basic block. >> Skip over >> + * it for the next iteration. >> + */ >> + b++; >> > > The correctness of this code depends on basic blocks being generated in > order (something which could conceivably change in the future). Just for > the sake of safety, can we have an assertion or two here just to verify > that the basic block being skipped is the same basic block that was removed? > > >> + >> + progress = true; >> + } >> + >> + if (progress) >> + invalidate_live_intervals(); >> + >> + return progress; >> +} >> -- >> 1.8.3.2 >> >> _______________________________________________ >> mesa-dev mailing list >> mesa-dev@lists.freedesktop.org >> http://lists.freedesktop.org/mailman/listinfo/mesa-dev >> > >
_______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev