i965: move compiled_once flag to brw_program

This allows us to delete brw_shader and removes the last use of
gl_linked_shader in the codegen paths.

Reviewed-by: Eric Anholt <eric@anholt.net>
This commit is contained in:
Timothy Arceri 2016-11-04 13:37:21 +11:00
parent 8417bf528e
commit 718a0cf49f
8 changed files with 23 additions and 48 deletions

View file

@ -333,6 +333,8 @@ struct brw_state_flags {
struct brw_program {
struct gl_program program;
GLuint id;
bool compiled_once;
};
@ -350,12 +352,6 @@ struct gen4_fragment_program {
};
struct brw_shader {
struct gl_linked_shader base;
bool compiled_once;
};
/**
* Bitmask indicating which fragment shader inputs represent varyings (and
* hence have to be delivered to the fragment shader by the SF/SBE stage).

View file

@ -66,10 +66,6 @@ brw_codegen_cs_prog(struct brw_context *brw,
bool start_busy = false;
double start_time = 0;
struct brw_shader *cs =
(struct brw_shader *) prog->_LinkedShaders[MESA_SHADER_COMPUTE];
assert (cs);
memset(&prog_data, 0, sizeof(prog_data));
if (prog->Comp.SharedSize > 64 * 1024) {
@ -134,11 +130,11 @@ brw_codegen_cs_prog(struct brw_context *brw,
return false;
}
if (unlikely(brw->perf_debug) && cs) {
if (cs->compiled_once) {
if (unlikely(brw->perf_debug)) {
if (cp->compiled_once) {
_mesa_problem(&brw->ctx, "CS programs shouldn't need recompiles");
}
cs->compiled_once = true;
cp->compiled_once = true;
if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
perf_debug("CS compile took %.03f ms and stalled the GPU\n",

View file

@ -116,8 +116,6 @@ brw_codegen_gs_prog(struct brw_context *brw,
* padding around uniform values below vec4 size, so the worst case is that
* every uniform is a float which gets padded to the size of a vec4.
*/
struct gl_linked_shader *gs = prog->_LinkedShaders[MESA_SHADER_GEOMETRY];
struct brw_shader *bgs = (struct brw_shader *) gs;
int param_count = gp->program.nir->num_uniforms / 4;
prog_data.base.base.param =
@ -154,7 +152,7 @@ brw_codegen_gs_prog(struct brw_context *brw,
char *error_str;
const unsigned *program =
brw_compile_gs(brw->screen->compiler, brw, mem_ctx, key,
&prog_data, gs->Program->nir, prog,
&prog_data, gp->program.nir, prog,
st_index, &program_size, &error_str);
if (program == NULL) {
ralloc_strcat(&prog->data->InfoLog, error_str);
@ -165,14 +163,14 @@ brw_codegen_gs_prog(struct brw_context *brw,
}
if (unlikely(brw->perf_debug)) {
if (bgs->compiled_once) {
if (gp->compiled_once) {
brw_gs_debug_recompile(brw, prog, key);
}
if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
perf_debug("GS compile took %.03f ms and stalled the GPU\n",
(get_time() - start_time) * 1000);
}
bgs->compiled_once = true;
gp->compiled_once = true;
}
/* Scratch space is used for register spilling */

View file

@ -177,14 +177,12 @@ process_glsl_ir(struct brw_context *brw,
extern "C" struct gl_linked_shader *
brw_new_shader(gl_shader_stage stage)
{
struct brw_shader *shader;
shader = rzalloc(NULL, struct brw_shader);
struct gl_linked_shader *shader = rzalloc(NULL, struct gl_linked_shader);
if (shader) {
shader->base.Stage = stage;
shader->Stage = stage;
}
return &shader->base;
return shader;
}
static void

View file

@ -277,15 +277,13 @@ brw_codegen_tcs_prog(struct brw_context *brw,
}
if (unlikely(brw->perf_debug)) {
struct gl_linked_shader *tcs = shader_prog ?
shader_prog->_LinkedShaders[MESA_SHADER_TESS_CTRL] : NULL;
struct brw_shader *btcs = (struct brw_shader *) tcs;
if (btcs) {
if (btcs->compiled_once) {
if (tcp) {
if (tcp->compiled_once) {
brw_tcs_debug_recompile(brw, shader_prog, key);
}
btcs->compiled_once = true;
tcp->compiled_once = true;
}
if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
perf_debug("TCS compile took %.03f ms and stalled the GPU\n",
(get_time() - start_time) * 1000);

View file

@ -195,17 +195,14 @@ brw_codegen_tes_prog(struct brw_context *brw,
}
if (unlikely(brw->perf_debug)) {
struct gl_linked_shader *tes =
shader_prog->_LinkedShaders[MESA_SHADER_TESS_EVAL];
struct brw_shader *btes = (struct brw_shader *) tes;
if (btes->compiled_once) {
if (tep->compiled_once) {
brw_tes_debug_recompile(brw, shader_prog, key);
}
if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
perf_debug("TES compile took %.03f ms and stalled the GPU\n",
(get_time() - start_time) * 1000);
}
btes->compiled_once = true;
tep->compiled_once = true;
}
/* Scratch space is used for register spilling */

View file

@ -97,13 +97,9 @@ brw_codegen_vs_prog(struct brw_context *brw,
struct brw_vs_prog_data prog_data;
struct brw_stage_prog_data *stage_prog_data = &prog_data.base.base;
void *mem_ctx;
struct brw_shader *vs = NULL;
bool start_busy = false;
double start_time = 0;
if (prog)
vs = (struct brw_shader *) prog->_LinkedShaders[MESA_SHADER_VERTEX];
memset(&prog_data, 0, sizeof(prog_data));
/* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
@ -202,15 +198,15 @@ brw_codegen_vs_prog(struct brw_context *brw,
return false;
}
if (unlikely(brw->perf_debug) && vs) {
if (vs->compiled_once) {
if (unlikely(brw->perf_debug)) {
if (vp->compiled_once) {
brw_vs_debug_recompile(brw, prog, key);
}
if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
perf_debug("VS compile took %.03f ms and stalled the GPU\n",
(get_time() - start_time) * 1000);
}
vs->compiled_once = true;
vp->compiled_once = true;
}
/* Scratch space is used for register spilling */

View file

@ -85,14 +85,10 @@ brw_codegen_wm_prog(struct brw_context *brw,
void *mem_ctx = ralloc_context(NULL);
struct brw_wm_prog_data prog_data;
const GLuint *program;
struct brw_shader *fs = NULL;
GLuint program_size;
bool start_busy = false;
double start_time = 0;
if (prog)
fs = (struct brw_shader *)prog->_LinkedShaders[MESA_SHADER_FRAGMENT];
memset(&prog_data, 0, sizeof(prog_data));
/* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
@ -161,10 +157,10 @@ brw_codegen_wm_prog(struct brw_context *brw,
return false;
}
if (unlikely(brw->perf_debug) && fs) {
if (fs->compiled_once)
if (unlikely(brw->perf_debug)) {
if (fp->compiled_once)
brw_wm_debug_recompile(brw, prog, key);
fs->compiled_once = true;
fp->compiled_once = true;
if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
perf_debug("FS compile took %.03f ms and stalled the GPU\n",