i965: Upload separate per-stage sampler state tables.

Also upload separate sampler default/texture border color entries.

At the moment, this is completely idiotic: both tables contain exactly
the same contents, so we're simply wasting batch space and CPU time.

However, soon we'll only upload data for textures actually /used/ in
a particular stage, which will usually make the VS table empty and
very likely eliminate all redundancy.  This is just a stepping stone.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
This commit is contained in:
Kenneth Graunke
2012-08-21 23:54:19 -07:00
parent 9525bcf5f7
commit decc708c7c
8 changed files with 72 additions and 38 deletions
+12 -9
View File
@@ -1064,11 +1064,6 @@ struct brw_context
GLuint last_bufsz;
} curbe;
/** SAMPLER_STATE count and offset */
struct {
uint32_t offset;
} sampler;
/**
* Layout of vertex data exiting the geometry portion of the pipleine.
* This comes from the geometry shader if one exists, otherwise from the
@@ -1110,7 +1105,13 @@ struct brw_context
uint32_t bind_bo_offset;
uint32_t surf_offset[BRW_MAX_VS_SURFACES];
/** SAMPLER_STATE count and table offset */
uint32_t sampler_count;
uint32_t sampler_offset;
/** Offsets in the batch to sampler default colors (texture border color)
*/
uint32_t sdc_offset[BRW_MAX_TEX_UNIT];
} vs;
struct {
@@ -1153,10 +1154,6 @@ struct brw_context
struct {
struct brw_wm_prog_data *prog_data;
/** offsets in the batch to sampler default colors (texture border color)
*/
uint32_t sdc_offset[BRW_MAX_TEX_UNIT];
GLuint render_surf;
drm_intel_bo *scratch_bo;
@@ -1184,7 +1181,13 @@ struct brw_context
uint32_t bind_bo_offset;
uint32_t surf_offset[BRW_MAX_WM_SURFACES];
/** SAMPLER_STATE count and table offset */
uint32_t sampler_count;
uint32_t sampler_offset;
/** Offsets in the batch to sampler default colors (texture border color)
*/
uint32_t sdc_offset[BRW_MAX_TEX_UNIT];
struct {
struct ra_regs *regs;
+2 -2
View File
@@ -157,12 +157,12 @@ brw_upload_vs_unit(struct brw_context *brw)
*/
if (brw->vs.sampler_count) {
vs->vs5.sampler_state_pointer =
(brw->batch.bo->offset + brw->sampler.offset) >> 5;
(brw->batch.bo->offset + brw->vs.sampler_offset) >> 5;
drm_intel_bo_emit_reloc(brw->batch.bo,
brw->vs.state_offset +
offsetof(struct brw_vs_unit_state, vs5),
brw->batch.bo,
brw->sampler.offset | vs->vs5.sampler_count,
brw->vs.sampler_offset | vs->vs5.sampler_count,
I915_GEM_DOMAIN_INSTRUCTION, 0);
}
@@ -195,6 +195,7 @@ static void brw_update_sampler_state(struct brw_context *brw,
int unit,
int ss_index,
struct brw_sampler_state *sampler,
uint32_t sampler_state_table_offset,
uint32_t *sdc_offset)
{
struct gl_context *ctx = &brw->ctx;
@@ -347,7 +348,7 @@ static void brw_update_sampler_state(struct brw_context *brw,
*sdc_offset) >> 5;
drm_intel_bo_emit_reloc(brw->batch.bo,
brw->sampler.offset +
sampler_state_table_offset +
ss_index * sizeof(struct brw_sampler_state) +
offsetof(struct brw_sampler_state, ss2),
brw->batch.bo, *sdc_offset,
@@ -366,7 +367,10 @@ static void brw_update_sampler_state(struct brw_context *brw,
static void
brw_upload_samplers(struct brw_context *brw)
brw_upload_sampler_state_table(struct brw_context *brw,
uint32_t *sampler_count,
uint32_t *sst_offset,
uint32_t *sdc_offset)
{
struct gl_context *ctx = &brw->ctx;
struct brw_sampler_state *samplers;
@@ -380,17 +384,15 @@ brw_upload_samplers(struct brw_context *brw)
/* ARB programs use the texture unit number as the sampler index, so we
* need to find the highest unit used. A bit-count will not work.
*/
brw->wm.sampler_count = _mesa_fls(SamplersUsed);
/* Currently we only use one sampler state table. Mirror the count. */
brw->vs.sampler_count = brw->wm.sampler_count;
*sampler_count = _mesa_fls(SamplersUsed);
if (brw->wm.sampler_count == 0)
if (*sampler_count == 0)
return;
samplers = brw_state_batch(brw, AUB_TRACE_SAMPLER_STATE,
brw->wm.sampler_count * sizeof(*samplers),
32, &brw->sampler.offset);
memset(samplers, 0, brw->wm.sampler_count * sizeof(*samplers));
*sampler_count * sizeof(*samplers),
32, sst_offset);
memset(samplers, 0, *sampler_count * sizeof(*samplers));
for (unsigned s = 0; s < brw->wm.sampler_count; s++) {
if (SamplersUsed & (1 << s)) {
@@ -398,13 +400,27 @@ brw_upload_samplers(struct brw_context *brw)
fs->SamplerUnits[s] : vs->SamplerUnits[s];
if (ctx->Texture.Unit[unit]._ReallyEnabled)
brw_update_sampler_state(brw, unit, s, &samplers[s],
&brw->wm.sdc_offset[s]);
*sst_offset, &sdc_offset[s]);
}
}
brw->state.dirty.cache |= CACHE_NEW_SAMPLER;
}
static void
brw_upload_samplers(struct brw_context *brw)
{
brw_upload_sampler_state_table(brw,
&brw->wm.sampler_count,
&brw->wm.sampler_offset,
brw->wm.sdc_offset);
brw_upload_sampler_state_table(brw,
&brw->vs.sampler_count,
&brw->vs.sampler_offset,
brw->vs.sdc_offset);
}
const struct brw_tracked_state brw_samplers = {
.dirty = {
.mesa = _NEW_TEXTURE,
+4 -4
View File
@@ -150,7 +150,7 @@ brw_upload_wm_unit(struct brw_context *brw)
if (brw->wm.sampler_count) {
/* reloc */
wm->wm4.sampler_state_pointer = (brw->batch.bo->offset +
brw->sampler.offset) >> 5;
brw->wm.sampler_offset) >> 5;
} else {
wm->wm4.sampler_state_pointer = 0;
}
@@ -229,9 +229,9 @@ brw_upload_wm_unit(struct brw_context *brw)
drm_intel_bo_emit_reloc(brw->batch.bo,
brw->wm.state_offset +
offsetof(struct brw_wm_unit_state, wm4),
brw->batch.bo, (brw->sampler.offset |
wm->wm4.stats_enable |
(wm->wm4.sampler_count << 2)),
brw->batch.bo, (brw->wm.sampler_offset |
wm->wm4.stats_enable |
(wm->wm4.sampler_count << 2)),
I915_GEM_DOMAIN_INSTRUCTION, 0);
}
@@ -39,9 +39,9 @@ upload_sampler_state_pointers(struct brw_context *brw)
GS_SAMPLER_STATE_CHANGE |
PS_SAMPLER_STATE_CHANGE |
(4 - 2));
OUT_BATCH(brw->sampler.offset); /* VS */
OUT_BATCH(brw->vs.sampler_offset); /* VS */
OUT_BATCH(0); /* GS */
OUT_BATCH(brw->sampler.offset);
OUT_BATCH(brw->wm.sampler_offset);
ADVANCE_BATCH();
}
+24 -9
View File
@@ -185,7 +185,10 @@ gen7_update_sampler_state(struct brw_context *brw, int unit, int ss_index,
static void
gen7_upload_samplers(struct brw_context *brw)
gen7_upload_sampler_state_table(struct brw_context *brw,
uint32_t *sampler_count,
uint32_t *sst_offset,
uint32_t *sdc_offset)
{
struct gl_context *ctx = &brw->ctx;
struct gen7_sampler_state *samplers;
@@ -196,17 +199,15 @@ gen7_upload_samplers(struct brw_context *brw)
GLbitfield SamplersUsed = vs->SamplersUsed | fs->SamplersUsed;
brw->wm.sampler_count = _mesa_fls(SamplersUsed);
/* Currently we only use one sampler state table. Mirror the count. */
brw->vs.sampler_count = brw->wm.sampler_count;
*sampler_count = _mesa_fls(SamplersUsed);
if (brw->wm.sampler_count == 0)
if (*sampler_count == 0)
return;
samplers = brw_state_batch(brw, AUB_TRACE_SAMPLER_STATE,
brw->wm.sampler_count * sizeof(*samplers),
32, &brw->sampler.offset);
memset(samplers, 0, brw->wm.sampler_count * sizeof(*samplers));
*sampler_count * sizeof(*samplers),
32, sst_offset);
memset(samplers, 0, *sampler_count * sizeof(*samplers));
for (unsigned s = 0; s < brw->wm.sampler_count; s++) {
if (SamplersUsed & (1 << s)) {
@@ -214,13 +215,27 @@ gen7_upload_samplers(struct brw_context *brw)
fs->SamplerUnits[s] : vs->SamplerUnits[s];
if (ctx->Texture.Unit[unit]._ReallyEnabled)
gen7_update_sampler_state(brw, unit, s, &samplers[s],
&brw->wm.sdc_offset[s]);
&sdc_offset[s]);
}
}
brw->state.dirty.cache |= CACHE_NEW_SAMPLER;
}
static void
gen7_upload_samplers(struct brw_context *brw)
{
gen7_upload_sampler_state_table(brw,
&brw->wm.sampler_count,
&brw->wm.sampler_offset,
brw->wm.sdc_offset);
gen7_upload_sampler_state_table(brw,
&brw->vs.sampler_count,
&brw->vs.sampler_offset,
brw->vs.sdc_offset);
}
const struct brw_tracked_state gen7_samplers = {
.dirty = {
.mesa = _NEW_TEXTURE,
+1 -1
View File
@@ -48,7 +48,7 @@ upload_vs_state(struct brw_context *brw)
/* CACHE_NEW_SAMPLER */
BEGIN_BATCH(2);
OUT_BATCH(_3DSTATE_SAMPLER_STATE_POINTERS_VS << 16 | (2 - 2));
OUT_BATCH(brw->sampler.offset);
OUT_BATCH(brw->vs.sampler_offset);
ADVANCE_BATCH();
if (brw->vs.push_const_size == 0) {
+1 -1
View File
@@ -125,7 +125,7 @@ upload_ps_state(struct brw_context *brw)
/* CACHE_NEW_SAMPLER */
BEGIN_BATCH(2);
OUT_BATCH(_3DSTATE_SAMPLER_STATE_POINTERS_PS << 16 | (2 - 2));
OUT_BATCH(brw->sampler.offset);
OUT_BATCH(brw->wm.sampler_offset);
ADVANCE_BATCH();
/* CACHE_NEW_WM_PROG */