On Wednesday, February 04, 2015 08:21:25 PM Matt Turner wrote: > And then the opt_combine_constants() pass will pull them out into > registers. This will allow us to do some algebraic optimizations on MAD > and LRP. > > total instructions in shared programs: 5946656 -> 5931320 (-0.26%) > instructions in affected programs: 778247 -> 762911 (-1.97%) > helped: 3780 > HURT: 6 > GAINED: 12 > LOST: 12 > --- > .../drivers/dri/i965/brw_fs_combine_constants.cpp | 22 > +++++++++++++++++++--- > .../drivers/dri/i965/brw_fs_copy_propagation.cpp | 6 ++++++ > 2 files changed, 25 insertions(+), 3 deletions(-) > > diff --git a/src/mesa/drivers/dri/i965/brw_fs_combine_constants.cpp > b/src/mesa/drivers/dri/i965/brw_fs_combine_constants.cpp > index ad4a965..e9341b7 100644 > --- a/src/mesa/drivers/dri/i965/brw_fs_combine_constants.cpp > +++ b/src/mesa/drivers/dri/i965/brw_fs_combine_constants.cpp > @@ -60,6 +60,18 @@ could_coissue(const fs_inst *inst) > } > } >
/**
* Returns true for instructions that don't support immediate sources.
*/
> +static bool
> +must_promote_imm(const fs_inst *inst)
> +{
> + switch (inst->opcode) {
> + case BRW_OPCODE_MAD:
> + case BRW_OPCODE_LRP:
> + return true;
Looking at fs_visitor::fix_math_operand, there are a few more cases we
could add in a follow up patch:
case SHADER_OPCODE_RCP:
case SHADER_OPCODE_RSQ:
case SHADER_OPCODE_SQRT:
case SHADER_OPCODE_EXP2:
case SHADER_OPCODE_LOG2:
case SHADER_OPCODE_POW:
case SHADER_OPCODE_INT_QUOTIENT:
case SHADER_OPCODE_INT_REMAINDER:
case SHADER_OPCODE_SIN:
case SHADER_OPCODE_COS:
return brw->gen < 8;
and perhaps delete the copy propagation prohibitions and some of
fix_math_operand(). Just an idea :)
> + default:
> + return false;
> + }
> +}
> +
> struct reg_link {
> DECLARE_RALLOC_CXX_OPERATORS(reg_link)
>
> @@ -86,6 +98,7 @@ struct imm {
> uint16_t reg;
>
> uint16_t uses_by_coissue;
/**
* Whether this constant is used by an instruction that can't handle an
* immediate source (and already has to be promoted to a GRF).
*/
> + bool must_promote;
>
> bool inserted;
>
> @@ -153,12 +166,13 @@ fs_visitor::opt_combine_constants()
> unsigned ip = -1;
>
> /* Make a pass through all instructions and count the number of times each
> - * constant is used by coissueable instructions.
> + * constant is used by coissueable instructions or instructions that
> cannot
> + * take immediate arguments.
> */
> foreach_block_and_inst(block, fs_inst, inst, cfg) {
> ip++;
>
> - if (!could_coissue(inst))
> + if (!could_coissue(inst) && !must_promote_imm(inst))
> continue;
>
> for (int i = 0; i < inst->sources; i++) {
> @@ -175,6 +189,7 @@ fs_visitor::opt_combine_constants()
> imm->block = intersection;
> imm->uses->push_tail(link(const_ctx, &inst->src[i]));
> imm->uses_by_coissue += could_coissue(inst);
> + imm->must_promote = imm->must_promote || must_promote_imm(inst);
> imm->last_use_ip = ip;
> } else {
> imm = new_imm(&table, const_ctx);
> @@ -184,6 +199,7 @@ fs_visitor::opt_combine_constants()
> imm->uses->push_tail(link(const_ctx, &inst->src[i]));
> imm->val = val;
> imm->uses_by_coissue = could_coissue(inst);
> + imm->must_promote = must_promote_imm(inst);
> imm->inserted = false;
> imm->first_use_ip = ip;
> imm->last_use_ip = ip;
> @@ -197,7 +213,7 @@ fs_visitor::opt_combine_constants()
> for (int i = 0; i < table.len;) {
> struct imm *imm = &table.imm[i];
>
> - if (imm->uses_by_coissue < 4) {
> + if (!imm->must_promote && imm->uses_by_coissue < 4) {
> table.imm[i] = table.imm[table.len - 1];
> table.len--;
> continue;
> diff --git a/src/mesa/drivers/dri/i965/brw_fs_copy_propagation.cpp
> b/src/mesa/drivers/dri/i965/brw_fs_copy_propagation.cpp
> index 68a266c..1b47851 100644
> --- a/src/mesa/drivers/dri/i965/brw_fs_copy_propagation.cpp
> +++ b/src/mesa/drivers/dri/i965/brw_fs_copy_propagation.cpp
> @@ -574,6 +574,12 @@ fs_visitor::try_constant_propagate(fs_inst *inst,
> acp_entry *entry)
> progress = true;
> break;
>
> + case BRW_OPCODE_MAD:
> + case BRW_OPCODE_LRP:
> + inst->src[i] = val;
> + progress = true;
> + break;
> +
> default:
> break;
> }
>
LGTM.
Reviewed-by: Kenneth Graunke <kenn...@whitecape.org>
signature.asc
Description: This is a digitally signed message part.
_______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev
