Mesa (master): nvc0: optimize blend cso by checking which by-RT data actually differs
Christoph Bumiller
chrisbmr at kemper.freedesktop.org
Tue May 29 15:02:08 UTC 2012
Module: Mesa
Branch: master
Commit: 1a21e36b6818991ced32d0e3d1acc1fea0d936ce
URL: http://cgit.freedesktop.org/mesa/mesa/commit/?id=1a21e36b6818991ced32d0e3d1acc1fea0d936ce
Author: Christoph Bumiller <e0425955 at student.tuwien.ac.at>
Date: Thu May 24 21:18:22 2012 +0200
nvc0: optimize blend cso by checking which by-RT data actually differs
Can save about 200 bytes of command buffer space.
---
src/gallium/drivers/nvc0/nvc0_state.c | 163 +++++++++++++++++++--------------
1 files changed, 96 insertions(+), 67 deletions(-)
diff --git a/src/gallium/drivers/nvc0/nvc0_state.c b/src/gallium/drivers/nvc0/nvc0_state.c
index 01bd5ce..cce2a48 100644
--- a/src/gallium/drivers/nvc0/nvc0_state.c
+++ b/src/gallium/drivers/nvc0/nvc0_state.c
@@ -86,77 +86,106 @@ static void *
nvc0_blend_state_create(struct pipe_context *pipe,
const struct pipe_blend_state *cso)
{
- struct nvc0_blend_stateobj *so = CALLOC_STRUCT(nvc0_blend_stateobj);
- int i;
- uint32_t ms;
+ struct nvc0_blend_stateobj *so = CALLOC_STRUCT(nvc0_blend_stateobj);
+ int i;
+ int r; /* reference */
+ uint32_t ms;
+ uint8_t blend_en = 0;
+ boolean indep_masks = FALSE;
+ boolean indep_funcs = FALSE;
- so->pipe = *cso;
+ so->pipe = *cso;
- SB_IMMED_3D(so, BLEND_INDEPENDENT, cso->independent_blend_enable);
-
- if (!cso->logicop_enable)
- SB_IMMED_3D(so, LOGIC_OP_ENABLE, 0);
-
- if (cso->logicop_enable) {
- SB_BEGIN_3D(so, LOGIC_OP_ENABLE, 2);
- SB_DATA (so, 1);
- SB_DATA (so, nvgl_logicop_func(cso->logicop_func));
-
- SB_IMMED_3D(so, MACRO_BLEND_ENABLES, 0);
- } else
- if (!cso->independent_blend_enable) {
- SB_IMMED_3D(so,
- MACRO_BLEND_ENABLES, cso->rt[0].blend_enable ? 0xff : 0);
-
- if (cso->rt[0].blend_enable) {
- SB_BEGIN_3D(so, BLEND_EQUATION_RGB, 5);
- SB_DATA (so, nvgl_blend_eqn(cso->rt[0].rgb_func));
- SB_DATA (so, nvc0_blend_fac(cso->rt[0].rgb_src_factor));
- SB_DATA (so, nvc0_blend_fac(cso->rt[0].rgb_dst_factor));
- SB_DATA (so, nvgl_blend_eqn(cso->rt[0].alpha_func));
- SB_DATA (so, nvc0_blend_fac(cso->rt[0].alpha_src_factor));
- SB_BEGIN_3D(so, BLEND_FUNC_DST_ALPHA, 1);
- SB_DATA (so, nvc0_blend_fac(cso->rt[0].alpha_dst_factor));
- }
-
- SB_IMMED_3D(so, COLOR_MASK_COMMON, 1);
- SB_BEGIN_3D(so, COLOR_MASK(0), 1);
- SB_DATA (so, nvc0_colormask(cso->rt[0].colormask));
- } else {
- uint8_t en = 0;
-
- for (i = 0; i < 8; ++i) {
- if (!cso->rt[i].blend_enable)
- continue;
- en |= 1 << i;
-
- SB_BEGIN_3D(so, IBLEND_EQUATION_RGB(i), 6);
- SB_DATA (so, nvgl_blend_eqn(cso->rt[i].rgb_func));
- SB_DATA (so, nvc0_blend_fac(cso->rt[i].rgb_src_factor));
- SB_DATA (so, nvc0_blend_fac(cso->rt[i].rgb_dst_factor));
- SB_DATA (so, nvgl_blend_eqn(cso->rt[i].alpha_func));
- SB_DATA (so, nvc0_blend_fac(cso->rt[i].alpha_src_factor));
- SB_DATA (so, nvc0_blend_fac(cso->rt[i].alpha_dst_factor));
- }
- SB_IMMED_3D(so, MACRO_BLEND_ENABLES, en);
-
- SB_IMMED_3D(so, COLOR_MASK_COMMON, 0);
- SB_BEGIN_3D(so, COLOR_MASK(0), 8);
- for (i = 0; i < 8; ++i)
+ /* check which states actually have differing values */
+ if (cso->independent_blend_enable) {
+ for (r = 0; r < 8 && !cso->rt[r].blend_enable; ++r);
+ blend_en |= 1 << r;
+ for (i = r + 1; i < 8; ++i) {
+ if (!cso->rt[i].blend_enable)
+ continue;
+ blend_en |= 1 << i;
+ if (cso->rt[i].rgb_func != cso->rt[r].rgb_func ||
+ cso->rt[i].rgb_src_factor != cso->rt[r].rgb_src_factor ||
+ cso->rt[i].rgb_dst_factor != cso->rt[r].rgb_dst_factor ||
+ cso->rt[i].alpha_func != cso->rt[r].alpha_func ||
+ cso->rt[i].alpha_src_factor != cso->rt[r].alpha_src_factor ||
+ cso->rt[i].alpha_dst_factor != cso->rt[r].alpha_dst_factor) {
+ indep_funcs = TRUE;
+ break;
+ }
+ }
+ for (; i < 8; ++i)
+ blend_en |= (cso->rt[i].blend_enable ? 1 : 0) << i;
+
+ for (i = 1; i < 8; ++i) {
+ if (cso->rt[i].colormask != cso->rt[0].colormask) {
+ indep_masks = TRUE;
+ break;
+ }
+ }
+ } else {
+ r = 0;
+ if (cso->rt[0].blend_enable)
+ blend_en = 0xff;
+ }
+
+ if (cso->logicop_enable) {
+ SB_BEGIN_3D(so, LOGIC_OP_ENABLE, 2);
+ SB_DATA (so, 1);
+ SB_DATA (so, nvgl_logicop_func(cso->logicop_func));
+
+ SB_IMMED_3D(so, MACRO_BLEND_ENABLES, 0);
+ } else {
+ SB_IMMED_3D(so, LOGIC_OP_ENABLE, 0);
+
+ SB_IMMED_3D(so, BLEND_INDEPENDENT, indep_funcs);
+ SB_IMMED_3D(so, MACRO_BLEND_ENABLES, blend_en);
+ if (indep_funcs) {
+ for (i = 0; i < 8; ++i) {
+ if (cso->rt[i].blend_enable) {
+ SB_BEGIN_3D(so, IBLEND_EQUATION_RGB(i), 6);
+ SB_DATA (so, nvgl_blend_eqn(cso->rt[i].rgb_func));
+ SB_DATA (so, nvc0_blend_fac(cso->rt[i].rgb_src_factor));
+ SB_DATA (so, nvc0_blend_fac(cso->rt[i].rgb_dst_factor));
+ SB_DATA (so, nvgl_blend_eqn(cso->rt[i].alpha_func));
+ SB_DATA (so, nvc0_blend_fac(cso->rt[i].alpha_src_factor));
+ SB_DATA (so, nvc0_blend_fac(cso->rt[i].alpha_dst_factor));
+ }
+ }
+ } else
+ if (blend_en) {
+ SB_BEGIN_3D(so, BLEND_EQUATION_RGB, 5);
+ SB_DATA (so, nvgl_blend_eqn(cso->rt[r].rgb_func));
+ SB_DATA (so, nvc0_blend_fac(cso->rt[r].rgb_src_factor));
+ SB_DATA (so, nvc0_blend_fac(cso->rt[r].rgb_dst_factor));
+ SB_DATA (so, nvgl_blend_eqn(cso->rt[r].alpha_func));
+ SB_DATA (so, nvc0_blend_fac(cso->rt[r].alpha_src_factor));
+ SB_BEGIN_3D(so, BLEND_FUNC_DST_ALPHA, 1);
+ SB_DATA (so, nvc0_blend_fac(cso->rt[r].alpha_dst_factor));
+ }
+
+ SB_IMMED_3D(so, COLOR_MASK_COMMON, !indep_masks);
+ if (indep_masks) {
+ SB_BEGIN_3D(so, COLOR_MASK(0), 8);
+ for (i = 0; i < 8; ++i)
SB_DATA(so, nvc0_colormask(cso->rt[i].colormask));
- }
+ } else {
+ SB_BEGIN_3D(so, COLOR_MASK(0), 1);
+ SB_DATA (so, nvc0_colormask(cso->rt[0].colormask));
+ }
+ }
- ms = 0;
- if (cso->alpha_to_coverage)
- ms |= NVC0_3D_MULTISAMPLE_CTRL_ALPHA_TO_COVERAGE;
- if (cso->alpha_to_one)
- ms |= NVC0_3D_MULTISAMPLE_CTRL_ALPHA_TO_ONE;
+ ms = 0;
+ if (cso->alpha_to_coverage)
+ ms |= NVC0_3D_MULTISAMPLE_CTRL_ALPHA_TO_COVERAGE;
+ if (cso->alpha_to_one)
+ ms |= NVC0_3D_MULTISAMPLE_CTRL_ALPHA_TO_ONE;
- SB_BEGIN_3D(so, MULTISAMPLE_CTRL, 1);
- SB_DATA (so, ms);
+ SB_BEGIN_3D(so, MULTISAMPLE_CTRL, 1);
+ SB_DATA (so, ms);
- assert(so->size <= (sizeof(so->state) / sizeof(so->state[0])));
- return so;
+ assert(so->size <= (sizeof(so->state) / sizeof(so->state[0])));
+ return so;
}
static void
More information about the mesa-commit
mailing list