Mesa (master): i965: Make the param pointer arrays for the WM dynamically sized.
Eric Anholt
anholt at kemper.freedesktop.org
Fri Sep 7 15:56:50 UTC 2012
Module: Mesa
Branch: master
Commit: f144b78dfbb97a70121be6f20d10bad8111267e3
URL: http://cgit.freedesktop.org/mesa/mesa/commit/?id=f144b78dfbb97a70121be6f20d10bad8111267e3
Author: Eric Anholt <eric at anholt.net>
Date: Sun Aug 26 21:19:05 2012 -0700
i965: Make the param pointer arrays for the WM dynamically sized.
Saves 26.5MB of wasted memory allocation in the l4d2 demo.
v2: Rebase on compare func change, fix comments.
Reviewed-by: Ian Romanick <ian.d.romanick at intel.com> (v1)
Reviewed-by: Kenneth Graunke <kenneth at whitecape.org>
---
src/mesa/drivers/dri/i965/brw_context.h | 7 ++++-
src/mesa/drivers/dri/i965/brw_fs.cpp | 2 -
src/mesa/drivers/dri/i965/brw_state_cache.c | 5 ++++
src/mesa/drivers/dri/i965/brw_wm.c | 35 +++++++++++++++++++++++++++
src/mesa/drivers/dri/i965/brw_wm.h | 1 +
5 files changed, 46 insertions(+), 4 deletions(-)
diff --git a/src/mesa/drivers/dri/i965/brw_context.h b/src/mesa/drivers/dri/i965/brw_context.h
index 12b28bd..0fe640c 100644
--- a/src/mesa/drivers/dri/i965/brw_context.h
+++ b/src/mesa/drivers/dri/i965/brw_context.h
@@ -322,8 +322,8 @@ struct brw_wm_prog_data {
* These must be the last fields of the struct (see
* brw_wm_prog_data_compare()).
*/
- const float *param[MAX_UNIFORMS * 4]; /* should be: BRW_MAX_CURBE */
- const float *pull_param[MAX_UNIFORMS * 4];
+ const float **param;
+ const float **pull_param;
};
/**
@@ -631,6 +631,7 @@ struct brw_cache_item {
typedef bool (*cache_aux_compare_func)(const void *a, const void *b,
int aux_size, const void *key);
+typedef void (*cache_aux_free_func)(const void *aux);
struct brw_cache {
struct brw_context *brw;
@@ -648,6 +649,8 @@ struct brw_cache {
* outside of the prog_data). If NULL, a plain memcmp is done.
*/
cache_aux_compare_func aux_compare[BRW_MAX_CACHE];
+ /** Optional functions for freeing other pointers attached to a prog_data. */
+ cache_aux_free_func aux_free[BRW_MAX_CACHE];
};
diff --git a/src/mesa/drivers/dri/i965/brw_fs.cpp b/src/mesa/drivers/dri/i965/brw_fs.cpp
index 167ea08..cbeefe5 100644
--- a/src/mesa/drivers/dri/i965/brw_fs.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs.cpp
@@ -528,8 +528,6 @@ fs_visitor::setup_uniform_values(int loc, const glsl_type *type)
for (unsigned int i = 0; i < type->vector_elements; i++) {
unsigned int param = c->prog_data.nr_params++;
- assert(param < ARRAY_SIZE(c->prog_data.param));
-
this->param_index[param] = loc;
this->param_offset[param] = i;
}
diff --git a/src/mesa/drivers/dri/i965/brw_state_cache.c b/src/mesa/drivers/dri/i965/brw_state_cache.c
index 92b81c7..b8a2fc3 100644
--- a/src/mesa/drivers/dri/i965/brw_state_cache.c
+++ b/src/mesa/drivers/dri/i965/brw_state_cache.c
@@ -343,6 +343,7 @@ brw_init_caches(struct brw_context *brw)
cache->aux_compare[BRW_VS_PROG] = brw_vs_prog_data_compare;
cache->aux_compare[BRW_WM_PROG] = brw_wm_prog_data_compare;
+ cache->aux_free[BRW_WM_PROG] = brw_wm_prog_data_free;
}
static void
@@ -357,6 +358,10 @@ brw_clear_cache(struct brw_context *brw, struct brw_cache *cache)
for (i = 0; i < cache->size; i++) {
for (c = cache->items[i]; c; c = next) {
next = c->next;
+ if (cache->aux_free[c->cache_id]) {
+ const void *item_aux = c->key + c->key_size;
+ cache->aux_free[c->cache_id](item_aux);
+ }
free((void *)c->key);
free(c);
}
diff --git a/src/mesa/drivers/dri/i965/brw_wm.c b/src/mesa/drivers/dri/i965/brw_wm.c
index 6ef8cc2..995e8f3 100644
--- a/src/mesa/drivers/dri/i965/brw_wm.c
+++ b/src/mesa/drivers/dri/i965/brw_wm.c
@@ -273,6 +273,15 @@ brw_wm_prog_data_compare(const void *in_a, const void *in_b,
return true;
}
+void
+brw_wm_prog_data_free(const void *in_prog_data)
+{
+ const struct brw_wm_prog_data *prog_data = in_prog_data;
+
+ ralloc_free((void *)prog_data->param);
+ ralloc_free((void *)prog_data->pull_param);
+}
+
/**
* All Mesa program -> GPU code generation goes through this function.
* Depending on the instructions used (i.e. flow control instructions)
@@ -286,8 +295,12 @@ bool do_wm_prog(struct brw_context *brw,
struct intel_context *intel = &brw->intel;
struct brw_wm_compile *c;
const GLuint *program;
+ struct gl_shader *fs = NULL;
GLuint program_size;
+ if (prog)
+ fs = prog->_LinkedShaders[MESA_SHADER_FRAGMENT];
+
c = brw->wm.compile_data;
if (c == NULL) {
brw->wm.compile_data = rzalloc(NULL, struct brw_wm_compile);
@@ -310,6 +323,28 @@ bool do_wm_prog(struct brw_context *brw,
c->vreg = vreg;
c->refs = refs;
}
+
+ /* Allocate the references to the uniforms that will end up in the
+ * prog_data associated with the compiled program, and which will be freed
+ * by the state cache.
+ */
+ if (fs) {
+ int param_count = fs->num_uniform_components;
+ /* The backend also sometimes adds params for texture size. */
+ param_count += 2 * BRW_MAX_TEX_UNIT;
+
+ c->prog_data.param = rzalloc_array(c, const float *, param_count);
+ c->prog_data.pull_param = rzalloc_array(c, const float *, param_count);
+ } else {
+ /* brw_wm_pass0.c will also add references to 0.0 and 1.0 which are
+ * uploaded as push parameters.
+ */
+ int param_count = (fp->program.Base.Parameters->NumParameters + 2) * 4;
+ c->prog_data.param = rzalloc_array(c, const float *, param_count);
+ /* The old backend never does pull constants. */
+ c->prog_data.pull_param = NULL;
+ }
+
memcpy(&c->key, key, sizeof(*key));
c->fp = fp;
diff --git a/src/mesa/drivers/dri/i965/brw_wm.h b/src/mesa/drivers/dri/i965/brw_wm.h
index d2936c2..a4b2046 100644
--- a/src/mesa/drivers/dri/i965/brw_wm.h
+++ b/src/mesa/drivers/dri/i965/brw_wm.h
@@ -486,5 +486,6 @@ void brw_wm_debug_recompile(struct brw_context *brw,
const struct brw_wm_prog_key *key);
bool brw_wm_prog_data_compare(const void *a, const void *b,
int aux_size, const void *key);
+void brw_wm_prog_data_free(const void *in_prog_data);
#endif
More information about the mesa-commit
mailing list