r600g: don't snoop context state while building shaders

Let's use the shader key describing the state.

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
This commit is contained in:
Marek Olšák 2012-09-17 23:22:00 +02:00
parent eb1d87fb94
commit b652180107
3 changed files with 43 additions and 28 deletions

View file

@ -254,6 +254,13 @@ struct r600_pipe_shader_selector {
unsigned nr_ps_max_color_exports;
};
struct r600_shader_key {
unsigned color_two_side:1;
unsigned alpha_to_one:1;
unsigned dual_src_blend:1;
unsigned nr_cbufs:4;
};
struct r600_pipe_shader {
struct r600_pipe_shader_selector *selector;
struct r600_pipe_shader *next_variant;
@ -266,7 +273,7 @@ struct r600_pipe_shader {
unsigned flatshade;
unsigned pa_cl_vs_out_cntl;
unsigned nr_ps_color_outputs;
unsigned key;
struct r600_shader_key key;
unsigned db_shader_control;
unsigned ps_depth_export;
};
@ -567,7 +574,9 @@ void r600_resume_timer_queries(struct r600_context *ctx);
void r600_init_context_resource_functions(struct r600_context *r600);
/* r600_shader.c */
int r600_pipe_shader_create(struct pipe_context *ctx, struct r600_pipe_shader *shader);
int r600_pipe_shader_create(struct pipe_context *ctx,
struct r600_pipe_shader *shader,
struct r600_shader_key key);
#ifdef HAVE_OPENCL
int r600_compute_shader_create(struct pipe_context * ctx,
LLVMModuleRef mod, struct r600_bytecode * bytecode);

View file

@ -103,9 +103,13 @@ static int r600_pipe_shader(struct pipe_context *ctx, struct r600_pipe_shader *s
return 0;
}
static int r600_shader_from_tgsi(struct r600_context * rctx, struct r600_pipe_shader *pipeshader);
static int r600_shader_from_tgsi(struct r600_screen *rscreen,
struct r600_pipe_shader *pipeshader,
struct r600_shader_key key);
int r600_pipe_shader_create(struct pipe_context *ctx, struct r600_pipe_shader *shader)
int r600_pipe_shader_create(struct pipe_context *ctx,
struct r600_pipe_shader *shader,
struct r600_shader_key key)
{
static int dump_shaders = -1;
struct r600_context *rctx = (struct r600_context *)ctx;
@ -136,7 +140,7 @@ int r600_pipe_shader_create(struct pipe_context *ctx, struct r600_pipe_shader *s
}
}
}
r = r600_shader_from_tgsi(rctx, shader);
r = r600_shader_from_tgsi(rctx->screen, shader, key);
if (r) {
R600_ERR("translation from TGSI failed !\n");
return r;
@ -1165,7 +1169,9 @@ static int process_twoside_color_inputs(struct r600_shader_ctx *ctx)
return 0;
}
static int r600_shader_from_tgsi(struct r600_context * rctx, struct r600_pipe_shader *pipeshader)
static int r600_shader_from_tgsi(struct r600_screen *rscreen,
struct r600_pipe_shader *pipeshader,
struct r600_shader_key key)
{
struct r600_shader *shader = &pipeshader->shader;
struct tgsi_token *tokens = pipeshader->selector->tokens;
@ -1190,7 +1196,7 @@ static int r600_shader_from_tgsi(struct r600_context * rctx, struct r600_pipe_sh
ctx.shader = shader;
ctx.native_integers = true;
r600_bytecode_init(ctx.bc, rctx->chip_class, rctx->family);
r600_bytecode_init(ctx.bc, rscreen->chip_class, rscreen->family);
ctx.tokens = tokens;
tgsi_scan_shader(tokens, &ctx.info);
tgsi_parse_init(&ctx.parse, tokens);
@ -1206,7 +1212,7 @@ static int r600_shader_from_tgsi(struct r600_context * rctx, struct r600_pipe_sh
shader->nr_ps_color_exports = 0;
shader->nr_ps_max_color_exports = 0;
shader->two_side = (ctx.type == TGSI_PROCESSOR_FRAGMENT) && rctx->two_side;
shader->two_side = key.color_two_side;
/* register allocations */
/* Values [0,127] correspond to GPR[0..127].
@ -1340,7 +1346,7 @@ static int r600_shader_from_tgsi(struct r600_context * rctx, struct r600_pipe_sh
}
}
if (shader->fs_write_all && rctx->chip_class >= EVERGREEN)
if (shader->fs_write_all && rscreen->chip_class >= EVERGREEN)
shader->nr_ps_max_color_exports = 8;
if (ctx.fragcoord_input >= 0) {
@ -1583,17 +1589,17 @@ static int r600_shader_from_tgsi(struct r600_context * rctx, struct r600_pipe_sh
case TGSI_PROCESSOR_FRAGMENT:
if (shader->output[i].name == TGSI_SEMANTIC_COLOR) {
/* never export more colors than the number of CBs */
if (next_pixel_base && next_pixel_base >= (rctx->nr_cbufs + rctx->dual_src_blend * 1)) {
if (next_pixel_base && next_pixel_base >= key.nr_cbufs + key.dual_src_blend) {
/* skip export */
j--;
continue;
}
output[j].swizzle_w = rctx->alpha_to_one && rctx->multisample_enable && !rctx->cb0_is_integer ? 5 : 3;
output[j].swizzle_w = key.alpha_to_one ? 5 : 3;
output[j].array_base = next_pixel_base++;
output[j].type = V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PIXEL;
shader->nr_ps_color_exports++;
if (shader->fs_write_all && (rctx->chip_class >= EVERGREEN)) {
for (k = 1; k < rctx->nr_cbufs; k++) {
if (shader->fs_write_all && (rscreen->chip_class >= EVERGREEN)) {
for (k = 1; k < key.nr_cbufs; k++) {
j++;
memset(&output[j], 0, sizeof(struct r600_bytecode_output));
output[j].gpr = shader->output[i].gpr;
@ -1601,7 +1607,7 @@ static int r600_shader_from_tgsi(struct r600_context * rctx, struct r600_pipe_sh
output[j].swizzle_x = 0;
output[j].swizzle_y = 1;
output[j].swizzle_z = 2;
output[j].swizzle_w = rctx->alpha_to_one && rctx->multisample_enable && !rctx->cb0_is_integer ? 5 : 3;
output[j].swizzle_w = key.alpha_to_one ? 5 : 3;
output[j].burst_count = 1;
output[j].barrier = 1;
output[j].array_base = next_pixel_base++;

View file

@ -728,19 +728,19 @@ static void *r600_create_vertex_elements(struct pipe_context *ctx, unsigned coun
}
/* Compute the key for the hw shader variant */
static INLINE unsigned r600_shader_selector_key(struct pipe_context * ctx,
static INLINE struct r600_shader_key r600_shader_selector_key(struct pipe_context * ctx,
struct r600_pipe_shader_selector * sel)
{
struct r600_context *rctx = (struct r600_context *)ctx;
unsigned key;
struct r600_shader_key key;
memset(&key, 0, sizeof(key));
if (sel->type == PIPE_SHADER_FRAGMENT) {
key = rctx->two_side |
((rctx->alpha_to_one && rctx->multisample_enable && !rctx->cb0_is_integer) << 1) |
(MIN2(sel->nr_ps_max_color_exports, rctx->nr_cbufs + rctx->dual_src_blend) << 2);
} else
key = 0;
key.color_two_side = rctx->two_side;
key.alpha_to_one = rctx->alpha_to_one && rctx->multisample_enable && !rctx->cb0_is_integer;
key.dual_src_blend = rctx->dual_src_blend;
key.nr_cbufs = rctx->nr_cbufs;
}
return key;
}
@ -750,7 +750,7 @@ static int r600_shader_select(struct pipe_context *ctx,
struct r600_pipe_shader_selector* sel,
unsigned *dirty)
{
unsigned key;
struct r600_shader_key key;
struct r600_context *rctx = (struct r600_context *)ctx;
struct r600_pipe_shader * shader = NULL;
int r;
@ -761,7 +761,7 @@ static int r600_shader_select(struct pipe_context *ctx,
* This path is also used for most shaders that don't need multiple
* variants, it will cost just a computation of the key and this
* test. */
if (likely(sel->current && sel->current->key == key)) {
if (likely(sel->current && memcmp(&sel->current->key, &key, sizeof(key)) == 0)) {
return 0;
}
@ -769,7 +769,7 @@ static int r600_shader_select(struct pipe_context *ctx,
if (sel->num_shaders > 1) {
struct r600_pipe_shader *p = sel->current, *c = p->next_variant;
while (c && c->key != key) {
while (c && memcmp(&c->key, &key, sizeof(key)) != 0) {
p = c;
c = c->next_variant;
}
@ -784,10 +784,10 @@ static int r600_shader_select(struct pipe_context *ctx,
shader = CALLOC(1, sizeof(struct r600_pipe_shader));
shader->selector = sel;
r = r600_pipe_shader_create(ctx, shader);
r = r600_pipe_shader_create(ctx, shader, key);
if (unlikely(r)) {
R600_ERR("Failed to build shader variant (type=%u, key=%u) %d\n",
sel->type, key, r);
R600_ERR("Failed to build shader variant (type=%u) %d\n",
sel->type, r);
sel->current = NULL;
return r;
}