mirror of
https://gitlab.freedesktop.org/mesa/mesa.git
synced 2026-01-19 17:40:32 +01:00
etnaviv: prep for UBOs
Allow UBO relocs and only emitting uniforms that are actually used. GC7000Lite has no address register, so upload uniforms to a UBO object to LOAD from. I removed the code to check for changes to individual uniforms and just reupload to entire uniform state when the state is dirty. I think there was very limited benefit to it and it isn't compatible with relocs. Signed-off-by: Jonathan Marek <jonathan@marek.ca> Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de> Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
This commit is contained in:
parent
ca58c1120e
commit
ee1ed59458
8 changed files with 70 additions and 78 deletions
|
|
@ -2265,13 +2265,20 @@ etna_compile_check_limits(struct etna_compile *c)
|
|||
static void
|
||||
copy_uniform_state_to_shader(struct etna_compile *c, struct etna_shader_variant *sobj)
|
||||
{
|
||||
uint32_t count = c->imm_size;
|
||||
uint32_t count = c->imm_base + c->imm_size;
|
||||
struct etna_shader_uniform_info *uinfo = &sobj->uniforms;
|
||||
|
||||
uinfo->const_count = c->imm_base;
|
||||
uinfo->imm_count = count;
|
||||
uinfo->imm_data = mem_dup(c->imm_data, count * sizeof(*c->imm_data));
|
||||
uinfo->imm_contents = mem_dup(c->imm_contents, count * sizeof(*c->imm_contents));
|
||||
|
||||
uinfo->imm_data = malloc(count * sizeof(*c->imm_data));
|
||||
for (unsigned i = 0; i < c->imm_base; i++)
|
||||
uinfo->imm_data[i] = i;
|
||||
memcpy(&uinfo->imm_data[c->imm_base], c->imm_data, c->imm_size * sizeof(*c->imm_data));
|
||||
|
||||
uinfo->imm_contents = malloc(count * sizeof(*c->imm_contents));
|
||||
for (unsigned i = 0; i < c->imm_base; i++)
|
||||
uinfo->imm_contents[i] = ETNA_IMMEDIATE_UNIFORM;
|
||||
memcpy(&uinfo->imm_contents[c->imm_base], c->imm_contents, c->imm_size * sizeof(*c->imm_contents));
|
||||
|
||||
etna_set_shader_uniforms_dirty_flags(sobj);
|
||||
}
|
||||
|
|
@ -2486,14 +2493,14 @@ etna_dump_shader(const struct etna_shader_variant *shader)
|
|||
|
||||
printf("num loops: %i\n", shader->num_loops);
|
||||
printf("num temps: %i\n", shader->num_temps);
|
||||
printf("num const: %i\n", shader->uniforms.const_count);
|
||||
printf("immediates:\n");
|
||||
for (int idx = 0; idx < shader->uniforms.imm_count; ++idx) {
|
||||
printf(" [%i].%s = %f (0x%08x)\n",
|
||||
(idx + shader->uniforms.const_count) / 4,
|
||||
printf(" [%i].%s = %f (0x%08x) (%d)\n",
|
||||
idx / 4,
|
||||
tgsi_swizzle_names[idx % 4],
|
||||
*((float *)&shader->uniforms.imm_data[idx]),
|
||||
shader->uniforms.imm_data[idx]);
|
||||
shader->uniforms.imm_data[idx],
|
||||
shader->uniforms.imm_contents[idx]);
|
||||
}
|
||||
printf("inputs:\n");
|
||||
for (int idx = 0; idx < shader->infile.num_reg; ++idx) {
|
||||
|
|
|
|||
|
|
@ -88,15 +88,17 @@ struct etna_shader_state {
|
|||
enum etna_immediate_contents {
|
||||
ETNA_IMMEDIATE_UNUSED = 0,
|
||||
ETNA_IMMEDIATE_CONSTANT,
|
||||
ETNA_IMMEDIATE_UNIFORM,
|
||||
ETNA_IMMEDIATE_TEXRECT_SCALE_X,
|
||||
ETNA_IMMEDIATE_TEXRECT_SCALE_Y,
|
||||
ETNA_IMMEDIATE_UBO0_ADDR,
|
||||
ETNA_IMMEDIATE_UBOMAX_ADDR = ETNA_IMMEDIATE_UBO0_ADDR + 255,
|
||||
};
|
||||
|
||||
struct etna_shader_uniform_info {
|
||||
enum etna_immediate_contents *imm_contents;
|
||||
uint32_t *imm_data;
|
||||
uint32_t imm_count;
|
||||
uint32_t const_count;
|
||||
};
|
||||
|
||||
struct etna_context {
|
||||
|
|
|
|||
|
|
@ -105,8 +105,8 @@ required_stream_size(struct etna_context *ctx)
|
|||
size += ctx->vertex_elements->num_elements + 1;
|
||||
|
||||
/* uniforms - worst case (2 words per uniform load) */
|
||||
size += ctx->shader.vs->uniforms.const_count * 2;
|
||||
size += ctx->shader.fs->uniforms.const_count * 2;
|
||||
size += ctx->shader.vs->uniforms.imm_count * 2;
|
||||
size += ctx->shader.fs->uniforms.imm_count * 2;
|
||||
|
||||
/* shader */
|
||||
size += ctx->shader_state.vs_inst_mem_size + 1;
|
||||
|
|
@ -583,16 +583,6 @@ etna_emit_state(struct etna_context *ctx)
|
|||
static const uint32_t uniform_dirty_bits =
|
||||
ETNA_DIRTY_SHADER | ETNA_DIRTY_CONSTBUF;
|
||||
|
||||
if (dirty & (uniform_dirty_bits | ctx->shader.vs->uniforms_dirty_bits))
|
||||
etna_uniforms_write(
|
||||
ctx, ctx->shader.vs, &ctx->constant_buffer[PIPE_SHADER_VERTEX],
|
||||
ctx->shader_state.VS_UNIFORMS, &ctx->shader_state.vs_uniforms_size);
|
||||
|
||||
if (dirty & (uniform_dirty_bits | ctx->shader.fs->uniforms_dirty_bits))
|
||||
etna_uniforms_write(
|
||||
ctx, ctx->shader.fs, &ctx->constant_buffer[PIPE_SHADER_FRAGMENT],
|
||||
ctx->shader_state.PS_UNIFORMS, &ctx->shader_state.ps_uniforms_size);
|
||||
|
||||
/**** Large dynamically-sized state ****/
|
||||
bool do_uniform_flush = ctx->specs.halti < 5;
|
||||
if (dirty & (ETNA_DIRTY_SHADER)) {
|
||||
|
|
@ -672,22 +662,13 @@ etna_emit_state(struct etna_context *ctx)
|
|||
|
||||
if (do_uniform_flush)
|
||||
etna_set_state(stream, VIVS_VS_UNIFORM_CACHE, VIVS_VS_UNIFORM_CACHE_FLUSH);
|
||||
etna_set_state_multi(stream, ctx->specs.vs_uniforms_offset,
|
||||
ctx->shader_state.vs_uniforms_size,
|
||||
ctx->shader_state.VS_UNIFORMS);
|
||||
|
||||
etna_uniforms_write(ctx, ctx->shader.vs, &ctx->constant_buffer[PIPE_SHADER_VERTEX]);
|
||||
|
||||
if (do_uniform_flush)
|
||||
etna_set_state(stream, VIVS_VS_UNIFORM_CACHE, VIVS_VS_UNIFORM_CACHE_FLUSH | VIVS_VS_UNIFORM_CACHE_PS);
|
||||
etna_set_state_multi(stream, ctx->specs.ps_uniforms_offset,
|
||||
ctx->shader_state.ps_uniforms_size,
|
||||
ctx->shader_state.PS_UNIFORMS);
|
||||
|
||||
/* Copy uniforms to gpu3d, so that incremental updates to uniforms are
|
||||
* possible as long as the
|
||||
* same shader remains bound */
|
||||
memcpy(ctx->gpu3d.VS_UNIFORMS, ctx->shader_state.VS_UNIFORMS,
|
||||
ctx->shader_state.vs_uniforms_size * 4);
|
||||
memcpy(ctx->gpu3d.PS_UNIFORMS, ctx->shader_state.PS_UNIFORMS,
|
||||
ctx->shader_state.ps_uniforms_size * 4);
|
||||
etna_uniforms_write(ctx, ctx->shader.fs, &ctx->constant_buffer[PIPE_SHADER_FRAGMENT]);
|
||||
|
||||
if (ctx->specs.halti >= 5) {
|
||||
/* HALTI5 needs to be prompted to pre-fetch shaders */
|
||||
|
|
@ -699,26 +680,16 @@ etna_emit_state(struct etna_context *ctx)
|
|||
/* ideally this cache would only be flushed if there are VS uniform changes */
|
||||
if (do_uniform_flush)
|
||||
etna_set_state(stream, VIVS_VS_UNIFORM_CACHE, VIVS_VS_UNIFORM_CACHE_FLUSH);
|
||||
etna_coalesce_start(stream, &coalesce);
|
||||
for (int x = 0; x < ctx->shader.vs->uniforms.const_count; ++x) {
|
||||
if (ctx->gpu3d.VS_UNIFORMS[x] != ctx->shader_state.VS_UNIFORMS[x]) {
|
||||
etna_coalsence_emit(stream, &coalesce, ctx->specs.vs_uniforms_offset + x*4, ctx->shader_state.VS_UNIFORMS[x]);
|
||||
ctx->gpu3d.VS_UNIFORMS[x] = ctx->shader_state.VS_UNIFORMS[x];
|
||||
}
|
||||
}
|
||||
etna_coalesce_end(stream, &coalesce);
|
||||
|
||||
if (dirty & (uniform_dirty_bits | ctx->shader.vs->uniforms_dirty_bits))
|
||||
etna_uniforms_write(ctx, ctx->shader.vs, &ctx->constant_buffer[PIPE_SHADER_VERTEX]);
|
||||
|
||||
/* ideally this cache would only be flushed if there are PS uniform changes */
|
||||
if (do_uniform_flush)
|
||||
etna_set_state(stream, VIVS_VS_UNIFORM_CACHE, VIVS_VS_UNIFORM_CACHE_FLUSH | VIVS_VS_UNIFORM_CACHE_PS);
|
||||
etna_coalesce_start(stream, &coalesce);
|
||||
for (int x = 0; x < ctx->shader.fs->uniforms.const_count; ++x) {
|
||||
if (ctx->gpu3d.PS_UNIFORMS[x] != ctx->shader_state.PS_UNIFORMS[x]) {
|
||||
etna_coalsence_emit(stream, &coalesce, ctx->specs.ps_uniforms_offset + x*4, ctx->shader_state.PS_UNIFORMS[x]);
|
||||
ctx->gpu3d.PS_UNIFORMS[x] = ctx->shader_state.PS_UNIFORMS[x];
|
||||
}
|
||||
}
|
||||
etna_coalesce_end(stream, &coalesce);
|
||||
|
||||
if (dirty & (uniform_dirty_bits | ctx->shader.fs->uniforms_dirty_bits))
|
||||
etna_uniforms_write(ctx, ctx->shader.fs, &ctx->constant_buffer[PIPE_SHADER_FRAGMENT]);
|
||||
}
|
||||
/**** End of state update ****/
|
||||
#undef EMIT_STATE
|
||||
|
|
|
|||
|
|
@ -262,13 +262,9 @@ struct compiled_shader_state {
|
|||
uint32_t GL_VARYING_COMPONENT_USE[2];
|
||||
uint32_t GL_HALTI5_SH_SPECIALS;
|
||||
unsigned vs_inst_mem_size;
|
||||
unsigned vs_uniforms_size;
|
||||
unsigned ps_inst_mem_size;
|
||||
unsigned ps_uniforms_size;
|
||||
uint32_t *VS_INST_MEM;
|
||||
uint32_t VS_UNIFORMS[ETNA_MAX_UNIFORMS * 4];
|
||||
uint32_t *PS_INST_MEM;
|
||||
uint32_t PS_UNIFORMS[ETNA_MAX_UNIFORMS * 4];
|
||||
struct etna_reloc PS_INST_ADDR;
|
||||
struct etna_reloc VS_INST_ADDR;
|
||||
};
|
||||
|
|
|
|||
|
|
@ -293,8 +293,7 @@ dump_shader_info(struct etna_shader_variant *v, struct pipe_debug_callback *debu
|
|||
|
||||
pipe_debug_message(debug, SHADER_INFO, "\n"
|
||||
"SHADER-DB: %s prog %d/%d: %u instructions %u temps\n"
|
||||
"SHADER-DB: %s prog %d/%d: %u immediates %u consts\n"
|
||||
"SHADER-DB: %s prog %d/%d: %u loops\n",
|
||||
"SHADER-DB: %s prog %d/%d: %u immediates %u loops\n",
|
||||
etna_shader_stage(v),
|
||||
v->shader->id, v->id,
|
||||
v->code_size,
|
||||
|
|
@ -302,9 +301,6 @@ dump_shader_info(struct etna_shader_variant *v, struct pipe_debug_callback *debu
|
|||
etna_shader_stage(v),
|
||||
v->shader->id, v->id,
|
||||
v->uniforms.imm_count,
|
||||
v->uniforms.const_count,
|
||||
etna_shader_stage(v),
|
||||
v->shader->id, v->id,
|
||||
v->num_loops);
|
||||
}
|
||||
|
||||
|
|
|
|||
|
|
@ -42,6 +42,7 @@
|
|||
#include "util/u_inlines.h"
|
||||
#include "util/u_math.h"
|
||||
#include "util/u_memory.h"
|
||||
#include "util/u_upload_mgr.h"
|
||||
|
||||
static void
|
||||
etna_set_stencil_ref(struct pipe_context *pctx, const struct pipe_stencil_ref *sr)
|
||||
|
|
@ -96,6 +97,11 @@ etna_set_constant_buffer(struct pipe_context *pctx,
|
|||
/* there is no support for ARB_uniform_buffer_object */
|
||||
assert(cb->buffer == NULL && cb->user_buffer != NULL);
|
||||
|
||||
if (!cb->buffer) {
|
||||
struct pipe_constant_buffer *cb = &ctx->constant_buffer[shader];
|
||||
u_upload_data(pctx->const_uploader, 0, cb->buffer_size, 16, cb->user_buffer, &cb->buffer_offset, &cb->buffer);
|
||||
}
|
||||
|
||||
ctx->dirty |= ETNA_DIRTY_CONSTBUF;
|
||||
}
|
||||
|
||||
|
|
|
|||
|
|
@ -29,6 +29,7 @@
|
|||
#include "etnaviv_compiler.h"
|
||||
#include "etnaviv_context.h"
|
||||
#include "etnaviv_util.h"
|
||||
#include "etnaviv_emit.h"
|
||||
#include "pipe/p_defines.h"
|
||||
#include "util/u_math.h"
|
||||
|
||||
|
|
@ -60,40 +61,55 @@ get_texrect_scale(const struct etna_context *ctx, bool frag,
|
|||
void
|
||||
etna_uniforms_write(const struct etna_context *ctx,
|
||||
const struct etna_shader_variant *sobj,
|
||||
struct pipe_constant_buffer *cb, uint32_t *uniforms,
|
||||
unsigned *size)
|
||||
struct pipe_constant_buffer *cb)
|
||||
{
|
||||
struct etna_cmd_stream *stream = ctx->stream;
|
||||
const struct etna_shader_uniform_info *uinfo = &sobj->uniforms;
|
||||
bool frag = false;
|
||||
bool frag = (sobj == ctx->shader.fs);
|
||||
uint32_t base = frag ? ctx->specs.ps_uniforms_offset : ctx->specs.vs_uniforms_offset;
|
||||
|
||||
if (cb->user_buffer) {
|
||||
unsigned size = MIN2(cb->buffer_size, uinfo->const_count * 4);
|
||||
if (!uinfo->imm_count)
|
||||
return;
|
||||
|
||||
memcpy(uniforms, cb->user_buffer, size);
|
||||
}
|
||||
|
||||
if (sobj == ctx->shader.fs)
|
||||
frag = true;
|
||||
etna_cmd_stream_reserve(stream, align(uinfo->imm_count + 1, 2));
|
||||
etna_emit_load_state(stream, base >> 2, uinfo->imm_count, 0);
|
||||
|
||||
for (uint32_t i = 0; i < uinfo->imm_count; i++) {
|
||||
uint32_t val = uinfo->imm_data[i];
|
||||
|
||||
switch (uinfo->imm_contents[i]) {
|
||||
case ETNA_IMMEDIATE_CONSTANT:
|
||||
uniforms[i + uinfo->const_count] = uinfo->imm_data[i];
|
||||
etna_cmd_stream_emit(stream, val);
|
||||
break;
|
||||
|
||||
case ETNA_IMMEDIATE_UNIFORM:
|
||||
assert(cb->user_buffer && val * 4 < cb->buffer_size);
|
||||
etna_cmd_stream_emit(stream, ((uint32_t*) cb->user_buffer)[val]);
|
||||
break;
|
||||
|
||||
case ETNA_IMMEDIATE_TEXRECT_SCALE_X:
|
||||
case ETNA_IMMEDIATE_TEXRECT_SCALE_Y:
|
||||
uniforms[i + uinfo->const_count] =
|
||||
get_texrect_scale(ctx, frag, uinfo->imm_contents[i], uinfo->imm_data[i]);
|
||||
etna_cmd_stream_emit(stream,
|
||||
get_texrect_scale(ctx, frag, uinfo->imm_contents[i], val));
|
||||
break;
|
||||
|
||||
case ETNA_IMMEDIATE_UBO0_ADDR ... ETNA_IMMEDIATE_UBOMAX_ADDR:
|
||||
assert(uinfo->imm_contents[i] == ETNA_IMMEDIATE_UBO0_ADDR);
|
||||
etna_cmd_stream_reloc(stream, &(struct etna_reloc) {
|
||||
.bo = etna_resource(cb->buffer)->bo,
|
||||
.flags = ETNA_RELOC_READ,
|
||||
.offset = cb->buffer_offset + val,
|
||||
});
|
||||
break;
|
||||
|
||||
case ETNA_IMMEDIATE_UNUSED:
|
||||
/* nothing to do */
|
||||
etna_cmd_stream_emit(stream, 0);
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
*size = uinfo->const_count + uinfo->imm_count;
|
||||
if ((uinfo->imm_count % 2) == 0)
|
||||
etna_cmd_stream_emit(stream, 0);
|
||||
}
|
||||
|
||||
void
|
||||
|
|
@ -103,8 +119,7 @@ etna_set_shader_uniforms_dirty_flags(struct etna_shader_variant *sobj)
|
|||
|
||||
for (uint32_t i = 0; i < sobj->uniforms.imm_count; i++) {
|
||||
switch (sobj->uniforms.imm_contents[i]) {
|
||||
case ETNA_IMMEDIATE_UNUSED:
|
||||
case ETNA_IMMEDIATE_CONSTANT:
|
||||
default:
|
||||
break;
|
||||
|
||||
case ETNA_IMMEDIATE_TEXRECT_SCALE_X:
|
||||
|
|
|
|||
|
|
@ -36,8 +36,7 @@ struct pipe_constant_buffer;
|
|||
void
|
||||
etna_uniforms_write(const struct etna_context *ctx,
|
||||
const struct etna_shader_variant *sobj,
|
||||
struct pipe_constant_buffer *cb, uint32_t *uniforms,
|
||||
unsigned *size);
|
||||
struct pipe_constant_buffer *cb);
|
||||
|
||||
void
|
||||
etna_set_shader_uniforms_dirty_flags(struct etna_shader_variant *sobj);
|
||||
|
|
|
|||
Loading…
Add table
Reference in a new issue