freedreno/ir3: remove Sethi-Ullman numbering pass
We haven't used this for a while. Signed-off-by: Rob Clark <robdclark@chromium.org> Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5048>
This commit is contained in:
@@ -53,8 +53,7 @@ ir3_SOURCES := \
|
||||
ir3/ir3_ra_regset.c \
|
||||
ir3/ir3_sched.c \
|
||||
ir3/ir3_shader.c \
|
||||
ir3/ir3_shader.h \
|
||||
ir3/ir3_sun.c
|
||||
ir3/ir3_shader.h
|
||||
|
||||
ir3_GENERATED_FILES := \
|
||||
ir3/ir3_nir_trig.c \
|
||||
|
||||
@@ -324,7 +324,6 @@ struct ir3_instruction {
|
||||
*/
|
||||
struct set *uses;
|
||||
|
||||
int sun; /* Sethi–Ullman number, used by sched */
|
||||
int use_count; /* currently just updated/used by cp */
|
||||
|
||||
/* Used during CP and RA stages. For collect and shader inputs/
|
||||
@@ -482,8 +481,6 @@ struct ir3 {
|
||||
/* List of ir3_array's: */
|
||||
struct list_head array_list;
|
||||
|
||||
unsigned max_sun; /* max Sethi–Ullman number */
|
||||
|
||||
#ifdef DEBUG
|
||||
unsigned block_count, instr_count;
|
||||
#endif
|
||||
@@ -1206,9 +1203,6 @@ void ir3_cp(struct ir3 *ir, struct ir3_shader_variant *so);
|
||||
/* group neighbors and insert mov's to resolve conflicts: */
|
||||
void ir3_group(struct ir3 *ir);
|
||||
|
||||
/* Sethi–Ullman numbering: */
|
||||
void ir3_sun(struct ir3 *ir);
|
||||
|
||||
/* scheduling: */
|
||||
void ir3_sched_add_deps(struct ir3 *ir);
|
||||
int ir3_sched(struct ir3 *ir);
|
||||
|
||||
@@ -3583,9 +3583,6 @@ ir3_compile_shader_nir(struct ir3_compiler *compiler,
|
||||
|
||||
ir3_debug_print(ir, "AFTER DCE");
|
||||
|
||||
/* do Sethi–Ullman numbering before scheduling: */
|
||||
ir3_sun(ir);
|
||||
|
||||
ret = ir3_sched(ir);
|
||||
if (ret) {
|
||||
DBG("SCHED failed!");
|
||||
@@ -3748,8 +3745,6 @@ ir3_compile_shader_nir(struct ir3_compiler *compiler,
|
||||
if (so->type == MESA_SHADER_FRAGMENT)
|
||||
so->total_in = max_bary + 1;
|
||||
|
||||
so->max_sun = ir->max_sun;
|
||||
|
||||
/* Collect sampling instructions eligible for pre-dispatch. */
|
||||
collect_tex_prefetches(ctx, ir);
|
||||
|
||||
|
||||
@@ -1,111 +0,0 @@
|
||||
/*
|
||||
* Copyright (C) 2018 Rob Clark <robclark@freedesktop.org>
|
||||
*
|
||||
* 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:
|
||||
* Rob Clark <robclark@freedesktop.org>
|
||||
*/
|
||||
|
||||
|
||||
#include "util/u_math.h"
|
||||
|
||||
#include "ir3.h"
|
||||
|
||||
/*
|
||||
* A simple pass to do Sethi–Ullman numbering, as described in "Generalizations
|
||||
* of the Sethi-Ullman algorithm for register allocation"[1]. This is used by
|
||||
* the scheduler pass.
|
||||
*
|
||||
* TODO this could probably be more clever about flow control, ie. if a src
|
||||
* is computed in multiple paths into a block, I think we should only have to
|
||||
* consider the worst-case.
|
||||
*
|
||||
* [1] https://www.cs.princeton.edu/~appel/papers/sun.pdf
|
||||
*/
|
||||
|
||||
static unsigned
|
||||
number_instr(struct ir3_instruction *instr)
|
||||
{
|
||||
if (ir3_instr_check_mark(instr))
|
||||
return instr->sun;
|
||||
|
||||
struct ir3_instruction *src;
|
||||
const unsigned n = __ssa_src_cnt(instr);
|
||||
unsigned a[n];
|
||||
unsigned b[n];
|
||||
unsigned i = 0;
|
||||
|
||||
/* TODO I think including false-deps in the calculation is the right
|
||||
* thing to do:
|
||||
*/
|
||||
foreach_ssa_src_n (src, n, instr) {
|
||||
if (__is_false_dep(instr, n))
|
||||
continue;
|
||||
if (src->block != instr->block) {
|
||||
a[i] = 1;
|
||||
} else {
|
||||
a[i] = number_instr(src);
|
||||
}
|
||||
b[i] = dest_regs(src);
|
||||
i++;
|
||||
}
|
||||
|
||||
/*
|
||||
* Rπ = max(aπ(1), bπ(1) + max(aπ(2), bπ(2) + max(..., bπ(k−1) + max(aπ(k), bπ(k)))...):
|
||||
*/
|
||||
unsigned last_r = 0;
|
||||
|
||||
for (int k = i - 1; k >= 0; k--) {
|
||||
unsigned r = MAX2(a[k], b[k] + last_r);
|
||||
|
||||
if (k > 0)
|
||||
r += b[k-1];
|
||||
|
||||
last_r = r;
|
||||
}
|
||||
|
||||
last_r = MAX2(last_r, dest_regs(instr));
|
||||
|
||||
instr->sun = last_r;
|
||||
|
||||
return instr->sun;
|
||||
}
|
||||
|
||||
void
|
||||
ir3_sun(struct ir3 *ir)
|
||||
{
|
||||
unsigned max = 0;
|
||||
|
||||
ir3_clear_mark(ir);
|
||||
|
||||
struct ir3_instruction *out;
|
||||
foreach_output (out, ir)
|
||||
max = MAX2(max, number_instr(out));
|
||||
|
||||
foreach_block (block, &ir->block_list) {
|
||||
for (unsigned i = 0; i < block->keeps_count; i++)
|
||||
max = MAX2(max, number_instr(block->keeps[i]));
|
||||
if (block->condition)
|
||||
max = MAX2(max, number_instr(block->condition));
|
||||
}
|
||||
|
||||
ir->max_sun = max;
|
||||
}
|
||||
@@ -98,7 +98,6 @@ libfreedreno_ir3_files = files(
|
||||
'ir3_sched.c',
|
||||
'ir3_shader.c',
|
||||
'ir3_shader.h',
|
||||
'ir3_sun.c',
|
||||
)
|
||||
|
||||
libfreedreno_ir3 = static_library(
|
||||
|
||||
Reference in New Issue
Block a user