fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-05-08 09:08:10 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

i915/corm: deferred constant allocation with per-channel UBO mixing

Browse source 
When opts.deferred_const is set, defer scalar load_const allocation
until the consuming ALU instruction. coalesce_constants resolves
deferred constants with a preferred register hint so co-occurring
constants pack into the same CONST register, avoiding dual-constant
MOV penalties.

Also fix per-channel UBO constant flags: mark only the actually
loaded channels with I915_CONSTFLAG_USER_CH(comp+i) instead of
setting all user bits, leaving free channels for immediates.

shader-db (I915_FS=nir): 210/403 compiled, 3202 alu
shader-db (I915_FS=both): nir won 210 (26 identical, 16 tied, 165 better, 3 only),
  77 TGSI, 116 neither

Assisted-by: Claude
This commit is contained in:
Adam Jackson 2026-05-06 12:59:01 -04:00
parent 28400d7c6c
commit 75ef9f6d65
1 changed files with 64 additions and 4 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

68
src/gallium/drivers/i915/i915_fpc_nir.c
View file

@ -22,6 +22,7 @@ struct nir_to_i915 {
uint32_t *ureg_map;
uint32_t **def_csr;
float *deferred_const;
unsigned ureg_map_size;
int *last_use;
@ -75,10 +76,28 @@ set_ureg(struct nir_to_i915 *c, nir_def *def, uint32_t ureg)
c->ureg_map[def->index] = ureg;
}
static bool
is_deferred(struct nir_to_i915 *c, unsigned ssa_index)
{
return c->ureg_map[ssa_index] == UREG_BAD;
}
static uint32_t
resolve_const(struct nir_to_i915 *c, unsigned ssa_index, int preferred_reg)
{
uint32_t ureg = i915_emit_const1f_prefer(c->p,
c->deferred_const[ssa_index],
preferred_reg);
c->ureg_map[ssa_index] = ureg;
return ureg;
}
static uint32_t
src_ureg(struct nir_to_i915 *c, nir_src *src)
{
assert(src->ssa->index < c->ureg_map_size);
if (c->ureg_map[src->ssa->index] == UREG_BAD)
resolve_const(c, src->ssa->index, -1);
return c->ureg_map[src->ssa->index];
}
@ -179,9 +198,17 @@ emit_load_const(struct nir_to_i915 *c, nir_load_const_instr *load)
struct i915_fp_compile *p = c->p;
switch (load->def.num_components) {
case 1:
set_ureg(c, &load->def, i915_emit_const1f(p, load->value[0].f32));
case 1: {
float val = load->value[0].f32;
if (c->opts.deferred_const &&
val != 0.0f && val != 1.0f && val != -1.0f) {
c->deferred_const[load->def.index] = val;
set_ureg(c, &load->def, UREG_BAD);
} else {
set_ureg(c, &load->def, i915_emit_const1f(p, val));
}
break;
}
case 2:
set_ureg(c, &load->def,
i915_emit_const2f(p, load->value[0].f32,
@ -205,6 +232,35 @@ emit_load_const(struct nir_to_i915 *c, nir_load_const_instr *load)
}
}
static void
coalesce_constants(struct nir_to_i915 *c, nir_alu_instr *alu)
{
unsigned n = nir_op_infos[alu->op].num_inputs;
unsigned deferred[3];
unsigned nr_deferred = 0;
int preferred = -1;
for (unsigned i = 0; i < n; i++) {
unsigned idx = alu->src[i].src.ssa->index;
if (is_deferred(c, idx)) {
deferred[nr_deferred++] = idx;
} else {
uint32_t ureg = c->ureg_map[idx];
if (GET_UREG_TYPE(ureg) == REG_TYPE_CONST && preferred < 0)
preferred = GET_UREG_NR(ureg);
}
}
if (nr_deferred == 0)
return;
for (unsigned i = 0; i < nr_deferred; i++) {
uint32_t ureg = resolve_const(c, deferred[i], preferred);
if (preferred < 0 && GET_UREG_TYPE(ureg) == REG_TYPE_CONST)
preferred = GET_UREG_NR(ureg);
}
}
static void
emit_alu(struct nir_to_i915 *c, nir_alu_instr *alu)
{
@ -214,6 +270,8 @@ emit_alu(struct nir_to_i915 *c, nir_alu_instr *alu)
uint32_t dest = UREG(REG_TYPE_R, i915_get_temp(p));
set_ureg(c, def, dest);
coalesce_constants(c, alu);
uint32_t src0 = 0, src1 = 0, src2 = 0;
if (nir_op_infos[alu->op].num_inputs >= 1)
src0 = alu_src_ureg(c, &alu->src[0]);
@ -756,7 +814,7 @@ emit_intrinsic(struct nir_to_i915 *c, nir_intrinsic_instr *intr)
}
for (unsigned i = 0; i < intr->def.num_components; i++)
ifs->constant_flags[slot] |= I915_CONSTFLAG_USER;
ifs->constant_flags[slot] |= I915_CONSTFLAG_USER_CH(comp + i);
ifs->num_constants = MAX2(ifs->num_constants, slot + 1);
uint32_t reg = UREG(REG_TYPE_CONST, slot);
@ -791,7 +849,7 @@ emit_intrinsic(struct nir_to_i915 *c, nir_intrinsic_instr *intr)
}
for (unsigned i = 0; i < intr->def.num_components; i++)
ifs->constant_flags[slot] |= I915_CONSTFLAG_USER;
ifs->constant_flags[slot] |= I915_CONSTFLAG_USER_CH(comp + i);
ifs->num_constants = MAX2(ifs->num_constants, slot + 1);
uint32_t reg = UREG(REG_TYPE_CONST, slot);
@ -937,6 +995,7 @@ i915_translate_fragment_program_nir(struct i915_context *i915,
.ureg_map_size = impl->ssa_alloc,
.ureg_map = CALLOC(impl->ssa_alloc, sizeof(uint32_t)),
.def_csr = CALLOC(impl->ssa_alloc, sizeof(uint32_t *)),
.deferred_const = CALLOC(impl->ssa_alloc, sizeof(float)),
.last_use = CALLOC(impl->ssa_alloc, sizeof(int)),
};
@ -1017,6 +1076,7 @@ cleanup:
ralloc_free(p->error);
FREE(c.last_use);
FREE(c.deferred_const);
FREE(c.def_csr);
FREE(c.ureg_map);
FREE(p);