[Mesa-dev] [PATCH] [rfc] radv: extend predicating to cache flushing.
Dave Airlie
airlied at gmail.com
Tue Jun 13 00:41:57 UTC 2017
From: Dave Airlie <airlied at redhat.com>
This allows us to skip cache flushing in the predicated area.
This gets about 10fps more on the deferred demo. (670->680)
Signed-off-by: Dave Airlie <airlied at redhat.com>
---
src/amd/vulkan/radv_cmd_buffer.c | 3 ++-
src/amd/vulkan/radv_device.c | 3 +++
src/amd/vulkan/radv_private.h | 3 +++
src/amd/vulkan/radv_query.c | 7 ++++--
src/amd/vulkan/si_cmd_buffer.c | 49 ++++++++++++++++++++++++----------------
5 files changed, 42 insertions(+), 23 deletions(-)
diff --git a/src/amd/vulkan/radv_cmd_buffer.c b/src/amd/vulkan/radv_cmd_buffer.c
index a33d776..69e4832 100644
--- a/src/amd/vulkan/radv_cmd_buffer.c
+++ b/src/amd/vulkan/radv_cmd_buffer.c
@@ -3323,6 +3323,7 @@ static void write_event(struct radv_cmd_buffer *cmd_buffer,
* the stage mask. */
si_cs_emit_write_event_eop(cs,
+ cmd_buffer->state.predicating,
cmd_buffer->device->physical_device->rad_info.chip_class,
false,
EVENT_TYPE_BOTTOM_OF_PIPE_TS, 0,
@@ -3374,7 +3375,7 @@ void radv_CmdWaitEvents(VkCommandBuffer commandBuffer,
MAYBE_UNUSED unsigned cdw_max = radeon_check_space(cmd_buffer->device->ws, cs, 7);
- si_emit_wait_fence(cs, va, 1, 0xffffffff);
+ si_emit_wait_fence(cs, false, va, 1, 0xffffffff);
assert(cmd_buffer->cs->cdw <= cdw_max);
}
diff --git a/src/amd/vulkan/radv_device.c b/src/amd/vulkan/radv_device.c
index 427d357..abbdfdd 100644
--- a/src/amd/vulkan/radv_device.c
+++ b/src/amd/vulkan/radv_device.c
@@ -1095,6 +1095,7 @@ VkResult radv_CreateDevice(
case RADV_QUEUE_GENERAL:
case RADV_QUEUE_COMPUTE:
si_cs_emit_cache_flush(device->flush_cs[family],
+ false,
device->physical_device->rad_info.chip_class,
NULL, 0,
family == RADV_QUEUE_COMPUTE && device->physical_device->rad_info.chip_class >= CIK,
@@ -1111,6 +1112,7 @@ VkResult radv_CreateDevice(
case RADV_QUEUE_GENERAL:
case RADV_QUEUE_COMPUTE:
si_cs_emit_cache_flush(device->flush_shader_cs[family],
+ false,
device->physical_device->rad_info.chip_class,
NULL, 0,
family == RADV_QUEUE_COMPUTE && device->physical_device->rad_info.chip_class >= CIK,
@@ -1761,6 +1763,7 @@ radv_get_preamble_cs(struct radv_queue *queue,
if (!i) {
si_cs_emit_cache_flush(cs,
+ false,
queue->device->physical_device->rad_info.chip_class,
NULL, 0,
queue->queue_family_index == RING_COMPUTE &&
diff --git a/src/amd/vulkan/radv_private.h b/src/amd/vulkan/radv_private.h
index a36ba07..821ac7e 100644
--- a/src/amd/vulkan/radv_private.h
+++ b/src/amd/vulkan/radv_private.h
@@ -849,6 +849,7 @@ uint32_t si_get_ia_multi_vgt_param(struct radv_cmd_buffer *cmd_buffer,
bool instanced_draw, bool indirect_draw,
uint32_t draw_vertex_count);
void si_cs_emit_write_event_eop(struct radeon_winsys_cs *cs,
+ bool predicated,
enum chip_class chip_class,
bool is_mec,
unsigned event, unsigned event_flags,
@@ -858,9 +859,11 @@ void si_cs_emit_write_event_eop(struct radeon_winsys_cs *cs,
uint32_t new_fence);
void si_emit_wait_fence(struct radeon_winsys_cs *cs,
+ bool predicated,
uint64_t va, uint32_t ref,
uint32_t mask);
void si_cs_emit_cache_flush(struct radeon_winsys_cs *cs,
+ bool predicated,
enum chip_class chip_class,
uint32_t *fence_ptr, uint64_t va,
bool is_mec,
diff --git a/src/amd/vulkan/radv_query.c b/src/amd/vulkan/radv_query.c
index 03b5af1..3c40774 100644
--- a/src/amd/vulkan/radv_query.c
+++ b/src/amd/vulkan/radv_query.c
@@ -992,7 +992,7 @@ void radv_CmdCopyQueryPoolResults(
uint64_t avail_va = va + pool->availability_offset + 4 * query;
/* This waits on the ME. All copies below are done on the ME */
- si_emit_wait_fence(cs, avail_va, 1, 0xffffffff);
+ si_emit_wait_fence(cs, false, avail_va, 1, 0xffffffff);
}
}
radv_query_shader(cmd_buffer, cmd_buffer->device->meta_state.query.pipeline_statistics_query_pipeline,
@@ -1015,7 +1015,7 @@ void radv_CmdCopyQueryPoolResults(
uint64_t avail_va = va + pool->availability_offset + 4 * query;
/* This waits on the ME. All copies below are done on the ME */
- si_emit_wait_fence(cs, avail_va, 1, 0xffffffff);
+ si_emit_wait_fence(cs, false, avail_va, 1, 0xffffffff);
}
if (flags & VK_QUERY_RESULT_WITH_AVAILABILITY_BIT) {
uint64_t avail_va = va + pool->availability_offset + 4 * query;
@@ -1149,6 +1149,7 @@ void radv_CmdEndQuery(
radeon_emit(cs, va >> 32);
si_cs_emit_write_event_eop(cs,
+ false,
cmd_buffer->device->physical_device->rad_info.chip_class,
false,
EVENT_TYPE_BOTTOM_OF_PIPE_TS, 0,
@@ -1198,11 +1199,13 @@ void radv_CmdWriteTimestamp(
break;
default:
si_cs_emit_write_event_eop(cs,
+ false,
cmd_buffer->device->physical_device->rad_info.chip_class,
mec,
V_028A90_BOTTOM_OF_PIPE_TS, 0,
3, query_va, 0, 0);
si_cs_emit_write_event_eop(cs,
+ false,
cmd_buffer->device->physical_device->rad_info.chip_class,
mec,
V_028A90_BOTTOM_OF_PIPE_TS, 0,
diff --git a/src/amd/vulkan/si_cmd_buffer.c b/src/amd/vulkan/si_cmd_buffer.c
index f84443b..2edc20a 100644
--- a/src/amd/vulkan/si_cmd_buffer.c
+++ b/src/amd/vulkan/si_cmd_buffer.c
@@ -816,6 +816,7 @@ si_get_ia_multi_vgt_param(struct radv_cmd_buffer *cmd_buffer,
}
void si_cs_emit_write_event_eop(struct radeon_winsys_cs *cs,
+ bool predicated,
enum chip_class chip_class,
bool is_mec,
unsigned event, unsigned event_flags,
@@ -830,7 +831,7 @@ void si_cs_emit_write_event_eop(struct radeon_winsys_cs *cs,
unsigned is_gfx8_mec = is_mec && chip_class < GFX9;
if (chip_class >= GFX9 || is_gfx8_mec) {
- radeon_emit(cs, PKT3(PKT3_RELEASE_MEM, is_gfx8_mec ? 5 : 6, 0));
+ radeon_emit(cs, PKT3(PKT3_RELEASE_MEM, is_gfx8_mec ? 5 : 6, predicated));
radeon_emit(cs, op);
radeon_emit(cs, EOP_DATA_SEL(data_sel));
radeon_emit(cs, va); /* address lo */
@@ -846,7 +847,7 @@ void si_cs_emit_write_event_eop(struct radeon_winsys_cs *cs,
* (and optional cache flushes executed) before the timestamp
* is written.
*/
- radeon_emit(cs, PKT3(PKT3_EVENT_WRITE_EOP, 4, 0));
+ radeon_emit(cs, PKT3(PKT3_EVENT_WRITE_EOP, 4, predicated));
radeon_emit(cs, op);
radeon_emit(cs, va);
radeon_emit(cs, ((va >> 32) & 0xffff) | EOP_DATA_SEL(data_sel));
@@ -854,7 +855,7 @@ void si_cs_emit_write_event_eop(struct radeon_winsys_cs *cs,
radeon_emit(cs, 0); /* unused */
}
- radeon_emit(cs, PKT3(PKT3_EVENT_WRITE_EOP, 4, 0));
+ radeon_emit(cs, PKT3(PKT3_EVENT_WRITE_EOP, 4, predicated));
radeon_emit(cs, op);
radeon_emit(cs, va);
radeon_emit(cs, ((va >> 32) & 0xffff) | EOP_DATA_SEL(data_sel));
@@ -865,10 +866,11 @@ void si_cs_emit_write_event_eop(struct radeon_winsys_cs *cs,
void
si_emit_wait_fence(struct radeon_winsys_cs *cs,
+ bool predicated,
uint64_t va, uint32_t ref,
uint32_t mask)
{
- radeon_emit(cs, PKT3(PKT3_WAIT_REG_MEM, 5, 0));
+ radeon_emit(cs, PKT3(PKT3_WAIT_REG_MEM, 5, predicated));
radeon_emit(cs, WAIT_REG_MEM_EQUAL | WAIT_REG_MEM_MEM_SPACE(1));
radeon_emit(cs, va);
radeon_emit(cs, va >> 32);
@@ -879,12 +881,14 @@ si_emit_wait_fence(struct radeon_winsys_cs *cs,
static void
si_emit_acquire_mem(struct radeon_winsys_cs *cs,
- bool is_mec, bool is_gfx9,
+ bool is_mec,
+ bool predicated,
+ bool is_gfx9,
unsigned cp_coher_cntl)
{
if (is_mec || is_gfx9) {
uint32_t hi_val = is_gfx9 ? 0xffffff : 0xff;
- radeon_emit(cs, PKT3(PKT3_ACQUIRE_MEM, 5, 0) |
+ radeon_emit(cs, PKT3(PKT3_ACQUIRE_MEM, 5, predicated) |
PKT3_SHADER_TYPE_S(is_mec));
radeon_emit(cs, cp_coher_cntl); /* CP_COHER_CNTL */
radeon_emit(cs, 0xffffffff); /* CP_COHER_SIZE */
@@ -894,7 +898,7 @@ si_emit_acquire_mem(struct radeon_winsys_cs *cs,
radeon_emit(cs, 0x0000000A); /* POLL_INTERVAL */
} else {
/* ACQUIRE_MEM is only required on a compute ring. */
- radeon_emit(cs, PKT3(PKT3_SURFACE_SYNC, 3, 0));
+ radeon_emit(cs, PKT3(PKT3_SURFACE_SYNC, 3, predicated));
radeon_emit(cs, cp_coher_cntl); /* CP_COHER_CNTL */
radeon_emit(cs, 0xffffffff); /* CP_COHER_SIZE */
radeon_emit(cs, 0); /* CP_COHER_BASE */
@@ -904,6 +908,7 @@ si_emit_acquire_mem(struct radeon_winsys_cs *cs,
void
si_cs_emit_cache_flush(struct radeon_winsys_cs *cs,
+ bool predicated,
enum chip_class chip_class,
uint32_t *flush_cnt,
uint64_t flush_va,
@@ -934,6 +939,7 @@ si_cs_emit_cache_flush(struct radeon_winsys_cs *cs,
/* Necessary for DCC */
if (chip_class >= VI) {
si_cs_emit_write_event_eop(cs,
+ predicated,
chip_class,
is_mec,
V_028A90_FLUSH_AND_INV_CB_DATA_TS,
@@ -947,27 +953,27 @@ si_cs_emit_cache_flush(struct radeon_winsys_cs *cs,
}
if (flush_bits & RADV_CMD_FLAG_FLUSH_AND_INV_CB_META) {
- radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
+ radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, predicated));
radeon_emit(cs, EVENT_TYPE(V_028A90_FLUSH_AND_INV_CB_META) | EVENT_INDEX(0));
}
if (flush_bits & RADV_CMD_FLAG_FLUSH_AND_INV_DB_META) {
- radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
+ radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, predicated));
radeon_emit(cs, EVENT_TYPE(V_028A90_FLUSH_AND_INV_DB_META) | EVENT_INDEX(0));
}
if (!flush_cb_db) {
if (flush_bits & RADV_CMD_FLAG_PS_PARTIAL_FLUSH) {
- radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
+ radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, predicated));
radeon_emit(cs, EVENT_TYPE(V_028A90_PS_PARTIAL_FLUSH) | EVENT_INDEX(4));
} else if (flush_bits & RADV_CMD_FLAG_VS_PARTIAL_FLUSH) {
- radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
+ radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, predicated));
radeon_emit(cs, EVENT_TYPE(V_028A90_VS_PARTIAL_FLUSH) | EVENT_INDEX(4));
}
}
if (flush_bits & RADV_CMD_FLAG_CS_PARTIAL_FLUSH) {
- radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
+ radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, predicated));
radeon_emit(cs, EVENT_TYPE(V_028A90_CS_PARTIAL_FLUSH) | EVENT_INDEX(4));
}
@@ -1014,14 +1020,14 @@ si_cs_emit_cache_flush(struct radeon_winsys_cs *cs,
assert(flush_cnt);
uint32_t old_fence = (*flush_cnt)++;
- si_cs_emit_write_event_eop(cs, chip_class, false, cb_db_event, tc_flags, 1,
+ si_cs_emit_write_event_eop(cs, predicated, chip_class, false, cb_db_event, tc_flags, 1,
flush_va, old_fence, *flush_cnt);
- si_emit_wait_fence(cs, flush_va, *flush_cnt, 0xffffffff);
+ si_emit_wait_fence(cs, predicated, flush_va, *flush_cnt, 0xffffffff);
}
/* VGT state sync */
if (flush_bits & RADV_CMD_FLAG_VGT_FLUSH) {
- radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
+ radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, predicated));
radeon_emit(cs, EVENT_TYPE(V_028A90_VGT_FLUSH) | EVENT_INDEX(0));
}
@@ -1034,13 +1040,13 @@ si_cs_emit_cache_flush(struct radeon_winsys_cs *cs,
RADV_CMD_FLAG_INV_GLOBAL_L2 |
RADV_CMD_FLAG_WRITEBACK_GLOBAL_L2))) &&
!is_mec) {
- radeon_emit(cs, PKT3(PKT3_PFP_SYNC_ME, 0, 0));
+ radeon_emit(cs, PKT3(PKT3_PFP_SYNC_ME, 0, predicated));
radeon_emit(cs, 0);
}
if ((flush_bits & RADV_CMD_FLAG_INV_GLOBAL_L2) ||
(chip_class <= CIK && (flush_bits & RADV_CMD_FLAG_WRITEBACK_GLOBAL_L2))) {
- si_emit_acquire_mem(cs, is_mec, chip_class >= GFX9,
+ si_emit_acquire_mem(cs, is_mec, predicated, chip_class >= GFX9,
cp_coher_cntl |
S_0085F0_TC_ACTION_ENA(1) |
S_0085F0_TCL1_ACTION_ENA(1) |
@@ -1054,14 +1060,16 @@ si_cs_emit_cache_flush(struct radeon_winsys_cs *cs,
*
* WB doesn't work without NC.
*/
- si_emit_acquire_mem(cs, is_mec, chip_class >= GFX9,
+ si_emit_acquire_mem(cs, is_mec, predicated,
+ chip_class >= GFX9,
cp_coher_cntl |
S_0301F0_TC_WB_ACTION_ENA(1) |
S_0301F0_TC_NC_ACTION_ENA(1));
cp_coher_cntl = 0;
}
if (flush_bits & RADV_CMD_FLAG_INV_VMEM_L1) {
- si_emit_acquire_mem(cs, is_mec, chip_class >= GFX9,
+ si_emit_acquire_mem(cs, is_mec,
+ predicated, chip_class >= GFX9,
cp_coher_cntl |
S_0085F0_TCL1_ACTION_ENA(1));
cp_coher_cntl = 0;
@@ -1072,7 +1080,7 @@ si_cs_emit_cache_flush(struct radeon_winsys_cs *cs,
* Therefore, it should be last. Done in PFP.
*/
if (cp_coher_cntl)
- si_emit_acquire_mem(cs, is_mec, chip_class >= GFX9, cp_coher_cntl);
+ si_emit_acquire_mem(cs, is_mec, predicated, chip_class >= GFX9, cp_coher_cntl);
}
void
@@ -1102,6 +1110,7 @@ si_emit_cache_flush(struct radv_cmd_buffer *cmd_buffer)
ptr = &cmd_buffer->gfx9_fence_idx;
}
si_cs_emit_cache_flush(cmd_buffer->cs,
+ cmd_buffer->state.predicating,
cmd_buffer->device->physical_device->rad_info.chip_class,
ptr, va,
radv_cmd_buffer_uses_mec(cmd_buffer),
--
2.9.4
More information about the mesa-dev
mailing list