OpenHarmony/third_party_mesa3d
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

glsl: remove GLSL IR inverse comparison optimisations

Browse Source 
As per 7d85dc4f35 GLSL IR is not smart enough to handle this
correctly for NANs.

Shader-db radeonsi (RX 6800):

Totals from affected shaders:
SGPRS: 26848 -> 26848 (0.00 %)
VGPRS: 13552 -> 13552 (0.00 %)
Spilled SGPRs: 134 -> 134 (0.00 %)
Spilled VGPRs: 0 -> 0 (0.00 %)
Private memory VGPRs: 0 -> 0 (0.00 %)
Scratch size: 0 -> 0 (0.00 %) dwords per thread
Code Size: 635000 -> 630988 (-0.63 %) bytes
Max Waves: 5474 -> 5474 (0.00 %)

Shader-db iris (BDW):

total instructions in shared programs: 17538859 -> 17539018 (<.01%)
instructions in affected programs: 29369 -> 29528 (0.54%)
helped: 3
HURT: 126
helped stats (abs) min: 1 max: 1 x̄: 1.00 x̃: 1
helped stats (rel) min: 0.49% max: 0.49% x̄: 0.49% x̃: 0.49%
HURT stats (abs)   min: 1 max: 2 x̄: 1.29 x̃: 1
HURT stats (rel)   min: 0.27% max: 1.32% x̄: 0.61% x̃: 0.54%
95% mean confidence interval for instructions value: 1.13 1.33
95% mean confidence interval for instructions %-change: 0.54% 0.63%
Instructions are HURT.

total loops in shared programs: 4866 -> 4866 (0.00%)
loops in affected programs: 0 -> 0
helped: 0
HURT: 0

total cycles in shared programs: 858548230 -> 858548915 (<.01%)
cycles in affected programs: 1331737 -> 1332422 (0.05%)
helped: 0
HURT: 92
HURT stats (abs)   min: 2 max: 49 x̄: 7.45 x̃: 6
HURT stats (rel)   min: 0.01% max: 1.90% x̄: 0.12% x̃: 0.05%
95% mean confidence interval for cycles value: 5.72 9.17
95% mean confidence interval for cycles %-change: 0.05% 0.19%
Cycles are HURT.

Note: With the addition of "nir/comparison_pre: See through an inot to
apply the optimization", idr's shader-db results are:

All Broadwell and newer Intel platforms had similar results. (Ice Lake shown)
total instructions in shared programs: 19940805 -> 19940802 (<.01%)
instructions in affected programs: 582 -> 579 (-0.52%)
helped: 3 / HURT: 0

total cycles in shared programs: 858431633 -> 858431747 (<.01%)
cycles in affected programs: 4938 -> 5052 (2.31%)
helped: 0 / HURT: 3

All older Intel platforms had similar results. (Haswell shown)
total instructions in shared programs: 16715626 -> 16715670 (<.01%)
instructions in affected programs: 9496 -> 9540 (0.46%)
helped: 0 / HURT: 44

total cycles in shared programs: 881224396 -> 881232314 (<.01%)
cycles in affected programs: 600610 -> 608528 (1.32%)
helped: 6 / HURT: 44

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18006>
This commit is contained in:
Timothy Arceri
2022-08-10 12:14:07 +10:00
committed by Marge Bot
parent 5473536798
commit f182b1952a
1 changed files with 0 additions and 30 deletions
Download Patch File Download Diff File Expand all files Collapse all files

30
src/compiler/glsl/opt_algebraic.cpp
View File

@@ -431,36 +431,6 @@ ir_algebraic_visitor::handle_expression(ir_expression *ir)
}
break;
case ir_unop_logic_not: {
enum ir_expression_operation new_op = ir_unop_logic_not;
if (op_expr[0] == NULL)
break;
switch (op_expr[0]->operation) {
case ir_binop_less: new_op = ir_binop_gequal; break;
case ir_binop_gequal: new_op = ir_binop_less; break;
case ir_binop_equal: new_op = ir_binop_nequal; break;
case ir_binop_nequal: new_op = ir_binop_equal; break;
case ir_binop_all_equal: new_op = ir_binop_any_nequal; break;
case ir_binop_any_nequal: new_op = ir_binop_all_equal; break;
default:
/* The default case handler is here to silence a warning from GCC.
*/
break;
}
if (new_op != ir_unop_logic_not) {
return new(mem_ctx) ir_expression(new_op,
ir->type,
op_expr[0]->operands[0],
op_expr[0]->operands[1]);
}
break;
}
case ir_unop_saturate:
if (op_expr[0] && op_expr[0]->operation == ir_binop_add) {
ir_expression *b2f_0 = op_expr[0]->operands[0]->as_expression();