From: Marek Olšák <[email protected]>
This just needs to be done for r600g in the screen.
We don't need an IB submission for every new context created for GCN.
---
src/gallium/drivers/r600/r600_pipe.c | 2 +-
src/gallium/drivers/radeon/r600_pipe_common.c | 1 +
src/gallium/drivers/radeon/r600_pipe_common.h | 3 +--
src/gallium/drivers/radeon/r600_query.c | 33 +++++++++++------------
src/gallium/drivers/radeonsi/si_pipe.c | 1 -
src/gallium/winsys/radeon/drm/radeon_drm_winsys.c | 3 +++
6 files changed, 21 insertions(+), 22 deletions(-)
diff --git a/src/gallium/drivers/r600/r600_pipe.c
b/src/gallium/drivers/r600/r600_pipe.c
index 98ceebf..f126af0 100644
--- a/src/gallium/drivers/r600/r600_pipe.c
+++ b/src/gallium/drivers/r600/r600_pipe.c
@@ -200,21 +200,20 @@ static struct pipe_context *r600_create_context(struct
pipe_screen *screen,
if (rscreen->b.debug_flags & DBG_FORCE_DMA)
rctx->b.b.resource_copy_region = rctx->b.dma_copy;
rctx->blitter = util_blitter_create(&rctx->b.b);
if (rctx->blitter == NULL)
goto fail;
util_blitter_set_texture_multisample(rctx->blitter, rscreen->has_msaa);
rctx->blitter->draw_rectangle = r600_draw_rectangle;
r600_begin_new_cs(rctx);
- r600_query_init_backend_mask(&rctx->b); /* this emits commands and must
be last */
rctx->dummy_pixel_shader =
util_make_fragment_cloneinput_shader(&rctx->b.b, 0,
TGSI_SEMANTIC_GENERIC,
TGSI_INTERPOLATE_CONSTANT);
rctx->b.b.bind_fs_state(&rctx->b.b, rctx->dummy_pixel_shader);
return &rctx->b.b;
fail:
@@ -728,12 +727,13 @@ struct pipe_screen *r600_screen_create(struct
radeon_winsys *ws)
for (i = 0; i < 256; i++) {
printf("%02X", map[i]);
if (i % 16 == 15)
printf("\n");
}
#endif
if (rscreen->b.debug_flags & DBG_TEST_DMA)
r600_test_dma(&rscreen->b);
+ r600_query_fix_enabled_rb_mask(rscreen);
return &rscreen->b.b;
}
diff --git a/src/gallium/drivers/radeon/r600_pipe_common.c
b/src/gallium/drivers/radeon/r600_pipe_common.c
index 5113765..5be21b4 100644
--- a/src/gallium/drivers/radeon/r600_pipe_common.c
+++ b/src/gallium/drivers/radeon/r600_pipe_common.c
@@ -1300,20 +1300,21 @@ bool r600_common_screen_init(struct r600_common_screen
*rscreen,
printf("num_good_compute_units = %i\n",
rscreen->info.num_good_compute_units);
printf("max_se = %i\n", rscreen->info.max_se);
printf("max_sh_per_se = %i\n", rscreen->info.max_sh_per_se);
printf("r600_gb_backend_map = %i\n",
rscreen->info.r600_gb_backend_map);
printf("r600_gb_backend_map_valid = %i\n",
rscreen->info.r600_gb_backend_map_valid);
printf("r600_num_banks = %i\n", rscreen->info.r600_num_banks);
printf("num_render_backends = %i\n",
rscreen->info.num_render_backends);
printf("num_tile_pipes = %i\n", rscreen->info.num_tile_pipes);
printf("pipe_interleave_bytes = %i\n",
rscreen->info.pipe_interleave_bytes);
+ printf("enabled_rb_mask = 0x%x\n",
rscreen->info.enabled_rb_mask);
}
return true;
}
void r600_destroy_common_screen(struct r600_common_screen *rscreen)
{
r600_perfcounters_destroy(rscreen);
r600_gpu_load_kill_thread(rscreen);
pipe_mutex_destroy(rscreen->gpu_load_mutex);
diff --git a/src/gallium/drivers/radeon/r600_pipe_common.h
b/src/gallium/drivers/radeon/r600_pipe_common.h
index f3e42ee..bc88fab 100644
--- a/src/gallium/drivers/radeon/r600_pipe_common.h
+++ b/src/gallium/drivers/radeon/r600_pipe_common.h
@@ -574,21 +574,20 @@ struct r600_common_context {
/* Additional context states. */
unsigned flags; /* flush flags */
/* Queries. */
/* Maintain the list of active queries for pausing between IBs. */
int num_occlusion_queries;
int num_perfect_occlusion_queries;
struct list_head active_queries;
unsigned num_cs_dw_queries_suspend;
/* Additional hardware info. */
- unsigned backend_mask;
unsigned max_db; /* for OQ */
/* Misc stats. */
unsigned num_draw_calls;
unsigned num_spill_draw_calls;
unsigned num_compute_calls;
unsigned num_spill_compute_calls;
unsigned num_dma_calls;
unsigned num_cp_dma_calls;
unsigned num_vs_flushes;
unsigned num_ps_flushes;
@@ -763,21 +762,21 @@ unsigned r600_end_counter(struct r600_common_screen
*rscreen, unsigned type,
uint64_t begin);
/* r600_perfcounters.c */
void r600_perfcounters_destroy(struct r600_common_screen *rscreen);
/* r600_query.c */
void r600_init_screen_query_functions(struct r600_common_screen *rscreen);
void r600_query_init(struct r600_common_context *rctx);
void r600_suspend_queries(struct r600_common_context *ctx);
void r600_resume_queries(struct r600_common_context *ctx);
-void r600_query_init_backend_mask(struct r600_common_context *ctx);
+void r600_query_fix_enabled_rb_mask(struct r600_common_screen *rscreen);
/* r600_streamout.c */
void r600_streamout_buffers_dirty(struct r600_common_context *rctx);
void r600_set_streamout_targets(struct pipe_context *ctx,
unsigned num_targets,
struct pipe_stream_output_target **targets,
const unsigned *offset);
void r600_emit_streamout_end(struct r600_common_context *rctx);
void r600_update_prims_generated_query_state(struct r600_common_context *rctx,
unsigned type, int diff);
diff --git a/src/gallium/drivers/radeon/r600_query.c
b/src/gallium/drivers/radeon/r600_query.c
index 25e7f5b..564f59e 100644
--- a/src/gallium/drivers/radeon/r600_query.c
+++ b/src/gallium/drivers/radeon/r600_query.c
@@ -422,28 +422,29 @@ static bool r600_query_hw_prepare_buffer(struct
r600_common_context *ctx,
uint32_t *results = ctx->ws->buffer_map(buffer->buf, NULL,
PIPE_TRANSFER_WRITE |
PIPE_TRANSFER_UNSYNCHRONIZED);
if (!results)
return false;
memset(results, 0, buffer->b.b.width0);
if (query->b.type == PIPE_QUERY_OCCLUSION_COUNTER ||
query->b.type == PIPE_QUERY_OCCLUSION_PREDICATE) {
+ unsigned enabled_rb_mask = ctx->screen->info.enabled_rb_mask;
unsigned num_results;
unsigned i, j;
/* Set top bits for unused backends. */
num_results = buffer->b.b.width0 / query->result_size;
for (j = 0; j < num_results; j++) {
for (i = 0; i < ctx->max_db; i++) {
- if (!(ctx->backend_mask & (1<<i))) {
+ if (!(enabled_rb_mask & (1<<i))) {
results[(i * 4)+1] = 0x80000000;
results[(i * 4)+3] = 0x80000000;
}
}
results += 4 * ctx->max_db;
}
}
return true;
}
@@ -1598,62 +1599,65 @@ void r600_resume_queries(struct r600_common_context
*ctx)
assert(ctx->num_cs_dw_queries_suspend == 0);
/* Check CS space here. Resuming must not be interrupted by flushes. */
ctx->need_gfx_cs_space(&ctx->b, num_cs_dw, true);
LIST_FOR_EACH_ENTRY(query, &ctx->active_queries, list) {
r600_query_hw_emit_start(ctx, query);
}
}
-/* Get backends mask */
-void r600_query_init_backend_mask(struct r600_common_context *ctx)
+/* Fix radeon_info::enabled_rb_mask for R600, R700, EVERGREEN, NI. */
+void r600_query_fix_enabled_rb_mask(struct r600_common_screen *rscreen)
{
+ struct r600_common_context *ctx =
+ (struct r600_common_context*)rscreen->aux_context;
struct radeon_winsys_cs *cs = ctx->gfx.cs;
struct r600_resource *buffer;
uint32_t *results;
- unsigned num_backends = ctx->screen->info.num_render_backends;
unsigned i, mask = 0;
+ assert(rscreen->chip_class <= CAYMAN);
+
/* if backend_map query is supported by the kernel */
- if (ctx->screen->info.r600_gb_backend_map_valid) {
- unsigned num_tile_pipes = ctx->screen->info.num_tile_pipes;
- unsigned backend_map = ctx->screen->info.r600_gb_backend_map;
+ if (rscreen->info.r600_gb_backend_map_valid) {
+ unsigned num_tile_pipes = rscreen->info.num_tile_pipes;
+ unsigned backend_map = rscreen->info.r600_gb_backend_map;
unsigned item_width, item_mask;
if (ctx->chip_class >= EVERGREEN) {
item_width = 4;
item_mask = 0x7;
} else {
item_width = 2;
item_mask = 0x3;
}
while (num_tile_pipes--) {
i = backend_map & item_mask;
mask |= (1<<i);
backend_map >>= item_width;
}
if (mask != 0) {
- ctx->backend_mask = mask;
+ rscreen->info.enabled_rb_mask = mask;
return;
}
}
/* otherwise backup path for older kernels */
/* create buffer for event data */
buffer = (struct r600_resource*)
pipe_buffer_create(ctx->b.screen, 0,
PIPE_USAGE_STAGING, ctx->max_db*16);
if (!buffer)
- goto err;
+ return;
/* initialize buffer with zeroes */
results = r600_buffer_map_sync_with_rings(ctx, buffer,
PIPE_TRANSFER_WRITE);
if (results) {
memset(results, 0, ctx->max_db * 4 * 4);
/* emit EVENT_WRITE for ZPASS_DONE */
radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 2, 0));
radeon_emit(cs, EVENT_TYPE(EVENT_TYPE_ZPASS_DONE) |
EVENT_INDEX(1));
radeon_emit(cs, buffer->gpu_address);
@@ -1668,29 +1672,22 @@ void r600_query_init_backend_mask(struct
r600_common_context *ctx)
for(i = 0; i < ctx->max_db; i++) {
/* at least highest bit will be set if backend
is used */
if (results[i*4 + 1])
mask |= (1<<i);
}
}
}
r600_resource_reference(&buffer, NULL);
- if (mask != 0) {
- ctx->backend_mask = mask;
- return;
- }
-
-err:
- /* fallback to old method - set num_backends lower bits to 1 */
- ctx->backend_mask = (~((uint32_t)0))>>(32-num_backends);
- return;
+ if (mask)
+ rscreen->info.enabled_rb_mask = mask;
}
#define XFULL(name_, query_type_, type_, result_type_, group_id_) \
{ \
.name = name_, \
.query_type = R600_QUERY_##query_type_, \
.type = PIPE_DRIVER_QUERY_TYPE_##type_, \
.result_type = PIPE_DRIVER_QUERY_RESULT_TYPE_##result_type_, \
.group_id = group_id_ \
}
diff --git a/src/gallium/drivers/radeonsi/si_pipe.c
b/src/gallium/drivers/radeonsi/si_pipe.c
index 8c54997..cb31113 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.c
+++ b/src/gallium/drivers/radeonsi/si_pipe.c
@@ -246,21 +246,20 @@ static struct pipe_context *si_create_context(struct
pipe_screen *screen,
sctx->blitter = util_blitter_create(&sctx->b.b);
if (sctx->blitter == NULL)
goto fail;
sctx->blitter->draw_rectangle = r600_draw_rectangle;
sctx->sample_mask.sample_mask = 0xffff;
/* these must be last */
si_begin_new_cs(sctx);
- r600_query_init_backend_mask(&sctx->b); /* this emits commands and must
be last */
/* CIK cannot unbind a constant buffer (S_BUFFER_LOAD doesn't skip loads
* if NUM_RECORDS == 0). We need to use a dummy buffer instead. */
if (sctx->b.chip_class == CIK) {
sctx->null_const_buf.buffer = pipe_buffer_create(screen,
PIPE_BIND_CONSTANT_BUFFER,
PIPE_USAGE_DEFAULT, 16);
if (!sctx->null_const_buf.buffer)
goto fail;
sctx->null_const_buf.buffer_size =
sctx->null_const_buf.buffer->width0;
diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_winsys.c
b/src/gallium/winsys/radeon/drm/radeon_drm_winsys.c
index c85e427..2a34d51 100644
--- a/src/gallium/winsys/radeon/drm/radeon_drm_winsys.c
+++ b/src/gallium/winsys/radeon/drm/radeon_drm_winsys.c
@@ -377,20 +377,23 @@ static bool do_winsys_init(struct radeon_drm_winsys *ws)
* unlikely to succeed. */
ws->info.max_alloc_size = MAX2(ws->info.vram_size, ws->info.gart_size) *
0.7;
if (ws->info.drm_minor < 40)
ws->info.max_alloc_size = MIN2(ws->info.max_alloc_size, 256*1024*1024);
/* Get max clock frequency info and convert it to MHz */
radeon_get_drm_value(ws->fd, RADEON_INFO_MAX_SCLK, NULL,
&ws->info.max_shader_clock);
ws->info.max_shader_clock /= 1000;
+ /* Default value. */
+ ws->info.enabled_rb_mask = u_bit_consecutive(0,
ws->info.num_render_backends);
+ /* This fails on non-GCN or older kernels: */
radeon_get_drm_value(ws->fd, RADEON_INFO_SI_BACKEND_ENABLED_MASK, NULL,
&ws->info.enabled_rb_mask);
ws->num_cpus = sysconf(_SC_NPROCESSORS_ONLN);
/* Generation-specific queries. */
if (ws->gen == DRV_R300) {
if (!radeon_get_drm_value(ws->fd, RADEON_INFO_NUM_GB_PIPES,
"GB pipe count",
&ws->info.r300_num_gb_pipes))
--
2.7.4
_______________________________________________
mesa-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
