Emitting the dirty blocks sometimes accounted for about 30% percent of the time
spent in r600_dri.so, so I made a seperate array with only the dirty blocks.
---
src/gallium/drivers/r600/r600.h | 2 ++
src/gallium/winsys/r600/drm/evergreen_hw_context.c | 18 +++++++++++++++---
src/gallium/winsys/r600/drm/r600_hw_context.c | 20 +++++++++++++++++---
src/gallium/winsys/r600/drm/r600_priv.h | 1 +
4 files changed, 35 insertions(+), 6 deletions(-)
diff --git a/src/gallium/drivers/r600/r600.h b/src/gallium/drivers/r600/r600.h
index 59a255d..8acb8c6 100644
--- a/src/gallium/drivers/r600/r600.h
+++ b/src/gallium/drivers/r600/r600.h
@@ -221,6 +221,8 @@ struct r600_context {
struct r600_range range[256];
unsigned nblocks;
struct r600_block **blocks;
+ unsigned ndirty;
+ struct r600_block **dirty;
unsigned pm4_ndwords;
unsigned pm4_cdwords;
unsigned pm4_dirty_cdwords;
diff --git a/src/gallium/winsys/r600/drm/evergreen_hw_context.c
b/src/gallium/winsys/r600/drm/evergreen_hw_context.c
index 96769ad..22c1021 100644
--- a/src/gallium/winsys/r600/drm/evergreen_hw_context.c
+++ b/src/gallium/winsys/r600/drm/evergreen_hw_context.c
@@ -613,6 +613,13 @@ int evergreen_context_init(struct r600_context *ctx,
struct radeon *radeon)
r = -ENOMEM;
goto out_err;
}
+
+ /* allocate an array for dirty blocks */
+ ctx->dirty = calloc(ctx->nblocks,sizeof(struct r600_block*));
+ if (ctx->dirty == NULL) {
+ r = -ENOMEM;
+ goto out_err;
+ }
return 0;
out_err:
r600_context_fini(ctx);
@@ -654,6 +661,7 @@ static inline void
evergreen_context_pipe_state_set_resource(struct r600_context
r600_bo_reference(ctx->radeon, &block->reloc[2].bo,
state->regs[3].bo);
}
if (!(block->status & R600_BLOCK_STATUS_DIRTY)) {
+ ctx->dirty[ctx->ndirty++] = block;
block->status |= R600_BLOCK_STATUS_ENABLED;
block->status |= R600_BLOCK_STATUS_DIRTY;
ctx->pm4_dirty_cdwords += block->pm4_ndwords +
block->pm4_flush_ndwords;
@@ -689,6 +697,7 @@ static inline void
evergreen_context_pipe_state_set_sampler(struct r600_context
block->reg[1] = state->regs[1].value;
block->reg[2] = state->regs[2].value;
if (!(block->status & R600_BLOCK_STATUS_DIRTY)) {
+ ctx->dirty[ctx->ndirty++] = block;
block->status |= R600_BLOCK_STATUS_ENABLED;
block->status |= R600_BLOCK_STATUS_DIRTY;
ctx->pm4_dirty_cdwords += block->pm4_ndwords +
block->pm4_flush_ndwords;
@@ -716,6 +725,7 @@ static inline void
evergreen_context_pipe_state_set_sampler_border(struct r600_c
block->reg[3] = state->regs[5].value;
block->reg[4] = state->regs[6].value;
if (!(block->status & R600_BLOCK_STATUS_DIRTY)) {
+ ctx->dirty[ctx->ndirty++] = block;
block->status |= R600_BLOCK_STATUS_ENABLED;
block->status |= R600_BLOCK_STATUS_DIRTY;
ctx->pm4_dirty_cdwords += block->pm4_ndwords +
block->pm4_flush_ndwords;
@@ -800,11 +810,12 @@ void evergreen_context_draw(struct r600_context *ctx,
const struct r600_draw *dr
}
/* enough room to copy packet */
- for (int i = 0; i < ctx->nblocks; i++) {
- if (ctx->blocks[i]->status & R600_BLOCK_STATUS_DIRTY) {
- r600_context_block_emit_dirty(ctx, ctx->blocks[i]);
+ for (int i = 0; i < ctx->ndirty; i++) {
+ if (ctx->dirty[i]->status & R600_BLOCK_STATUS_DIRTY) {
+ r600_context_block_emit_dirty(ctx, ctx->dirty[i]);
}
}
+ ctx->ndirty = 0;
/* draw packet */
ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_INDEX_TYPE, 0);
@@ -881,6 +892,7 @@ static inline void evergreen_resource_set(struct
r600_context *ctx, struct r600_
r600_bo_reference(ctx->radeon, &block->reloc[2].bo,
state->regs[3].bo);
}
if (!(block->status & R600_BLOCK_STATUS_DIRTY)) {
+ ctx->dirty[ctx->ndirty++] = block;
block->status |= R600_BLOCK_STATUS_ENABLED;
block->status |= R600_BLOCK_STATUS_DIRTY;
ctx->pm4_dirty_cdwords += block->pm4_ndwords +
block->pm4_flush_ndwords;
diff --git a/src/gallium/winsys/r600/drm/r600_hw_context.c
b/src/gallium/winsys/r600/drm/r600_hw_context.c
index 84539d1..45c1f0b 100644
--- a/src/gallium/winsys/r600/drm/r600_hw_context.c
+++ b/src/gallium/winsys/r600/drm/r600_hw_context.c
@@ -572,6 +572,7 @@ void r600_context_fini(struct r600_context *ctx)
}
free(ctx->reloc);
free(ctx->pm4);
+ free(ctx->dirty);
memset(ctx, 0, sizeof(struct r600_context));
}
@@ -691,6 +692,13 @@ int r600_context_init(struct r600_context *ctx, struct
radeon *radeon)
r = -ENOMEM;
goto out_err;
}
+
+ /* allocate an array for dirty blocks */
+ ctx->dirty = calloc(ctx->nblocks,sizeof(struct r600_block*));
+ if (ctx->dirty == NULL) {
+ r = -ENOMEM;
+ goto out_err;
+ }
return 0;
out_err:
r600_context_fini(ctx);
@@ -759,6 +767,7 @@ void r600_context_pipe_state_set(struct r600_context *ctx,
struct r600_pipe_stat
r600_bo_reference(ctx->radeon, &block->reloc[id].bo,
state->regs[i].bo);
}
if (!(block->status & R600_BLOCK_STATUS_DIRTY)) {
+ ctx->dirty[ctx->ndirty++] = block;
block->status |= R600_BLOCK_STATUS_ENABLED;
block->status |= R600_BLOCK_STATUS_DIRTY;
ctx->pm4_dirty_cdwords += block->pm4_ndwords +
block->pm4_flush_ndwords;
@@ -800,6 +809,7 @@ static inline void
r600_context_pipe_state_set_resource(struct r600_context *ctx
r600_bo_reference(ctx->radeon, &block->reloc[2].bo,
state->regs[3].bo);
}
if (!(block->status & R600_BLOCK_STATUS_DIRTY)) {
+ ctx->dirty[ctx->ndirty++] = block;
block->status |= R600_BLOCK_STATUS_ENABLED;
block->status |= R600_BLOCK_STATUS_DIRTY;
ctx->pm4_dirty_cdwords += block->pm4_ndwords +
block->pm4_flush_ndwords;
@@ -835,6 +845,7 @@ static inline void
r600_context_pipe_state_set_sampler(struct r600_context *ctx,
block->reg[1] = state->regs[1].value;
block->reg[2] = state->regs[2].value;
if (!(block->status & R600_BLOCK_STATUS_DIRTY)) {
+ ctx->dirty[ctx->ndirty++] = block;
block->status |= R600_BLOCK_STATUS_ENABLED;
block->status |= R600_BLOCK_STATUS_DIRTY;
ctx->pm4_dirty_cdwords += block->pm4_ndwords +
block->pm4_flush_ndwords;
@@ -860,6 +871,7 @@ static inline void
r600_context_pipe_state_set_sampler_border(struct r600_contex
block->reg[2] = state->regs[5].value;
block->reg[3] = state->regs[6].value;
if (!(block->status & R600_BLOCK_STATUS_DIRTY)) {
+ ctx->dirty[ctx->ndirty++] = block;
block->status |= R600_BLOCK_STATUS_ENABLED;
block->status |= R600_BLOCK_STATUS_DIRTY;
ctx->pm4_dirty_cdwords += block->pm4_ndwords +
block->pm4_flush_ndwords;
@@ -959,11 +971,12 @@ void r600_context_draw(struct r600_context *ctx, const
struct r600_draw *draw)
}
/* enough room to copy packet */
- for (int i = 0; i < ctx->nblocks; i++) {
- if (ctx->blocks[i]->status & R600_BLOCK_STATUS_DIRTY) {
- r600_context_block_emit_dirty(ctx, ctx->blocks[i]);
+ for (int i = 0; i < ctx->ndirty; i++) {
+ if (ctx->dirty[i]->status & R600_BLOCK_STATUS_DIRTY) {
+ r600_context_block_emit_dirty(ctx, ctx->dirty[i]);
}
}
+ ctx->ndirty = 0;
/* draw packet */
ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_INDEX_TYPE, 0);
@@ -1052,6 +1065,7 @@ void r600_context_flush(struct r600_context *ctx)
*/
for (int i = 0; i < ctx->nblocks; i++) {
if (ctx->blocks[i]->status & R600_BLOCK_STATUS_ENABLED) {
+ ctx->dirty[ctx->ndirty++] = ctx->blocks[i];
ctx->pm4_dirty_cdwords += ctx->blocks[i]->pm4_ndwords +
ctx->blocks[i]->pm4_flush_ndwords;
ctx->blocks[i]->status |= R600_BLOCK_STATUS_DIRTY;
}
diff --git a/src/gallium/winsys/r600/drm/r600_priv.h
b/src/gallium/winsys/r600/drm/r600_priv.h
index f6ceb0a..1d0c122 100644
--- a/src/gallium/winsys/r600/drm/r600_priv.h
+++ b/src/gallium/winsys/r600/drm/r600_priv.h
@@ -134,6 +134,7 @@ static void inline r600_context_reg(struct r600_context
*ctx,
block->reg[id] &= ~mask;
block->reg[id] |= value;
if (!(block->status & R600_BLOCK_STATUS_DIRTY)) {
+ ctx->dirty[ctx->ndirty++] = block;
ctx->pm4_dirty_cdwords += block->pm4_ndwords;
block->status |= R600_BLOCK_STATUS_ENABLED;
block->status |= R600_BLOCK_STATUS_DIRTY;
--
1.7.1
_______________________________________________
mesa-dev mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/mesa-dev
