[Mesa-dev] [PATCH 1/5] anv: Move push constant allocation to the command buffer
Jason Ekstrand
jason at jlekstrand.net
Fri May 20 19:41:04 UTC 2016
Instead of blasting it out as part of the pipeline, we put it in the
command buffer and only blast it out when it's really needed. Since the
PUSH_CONSTANT_ALLOC commands aren't pipelined, they immediately cause a
stall which we would like to avoid.
---
src/intel/vulkan/anv_cmd_buffer.c | 1 +
src/intel/vulkan/anv_pipeline.c | 22 ----------
src/intel/vulkan/anv_private.h | 2 +-
src/intel/vulkan/genX_cmd_buffer.c | 78 +++++++++++++++++++++++++++++++----
src/intel/vulkan/genX_pipeline_util.h | 12 ------
5 files changed, 71 insertions(+), 44 deletions(-)
diff --git a/src/intel/vulkan/anv_cmd_buffer.c b/src/intel/vulkan/anv_cmd_buffer.c
index bba24e8..db7793e 100644
--- a/src/intel/vulkan/anv_cmd_buffer.c
+++ b/src/intel/vulkan/anv_cmd_buffer.c
@@ -130,6 +130,7 @@ anv_cmd_state_reset(struct anv_cmd_buffer *cmd_buffer)
state->descriptors_dirty = 0;
state->push_constants_dirty = 0;
state->pipeline = NULL;
+ state->push_constant_stages = 0;
state->restart_index = UINT32_MAX;
state->dynamic = default_dynamic_state;
state->need_query_wa = true;
diff --git a/src/intel/vulkan/anv_pipeline.c b/src/intel/vulkan/anv_pipeline.c
index faa0adb..c010f0f 100644
--- a/src/intel/vulkan/anv_pipeline.c
+++ b/src/intel/vulkan/anv_pipeline.c
@@ -939,28 +939,6 @@ anv_compute_urb_partition(struct anv_pipeline *pipeline)
pipeline->urb.start[MESA_SHADER_TESS_EVAL] = push_constant_chunks;
pipeline->urb.size[MESA_SHADER_TESS_EVAL] = 1;
pipeline->urb.entries[MESA_SHADER_TESS_EVAL] = 0;
-
- const unsigned stages =
- _mesa_bitcount(pipeline->active_stages & VK_SHADER_STAGE_ALL_GRAPHICS);
- unsigned size_per_stage = stages ? (push_constant_kb / stages) : 0;
- unsigned used_kb = 0;
-
- /* Broadwell+ and Haswell gt3 require that the push constant sizes be in
- * units of 2KB. Incidentally, these are the same platforms that have
- * 32KB worth of push constant space.
- */
- if (push_constant_kb == 32)
- size_per_stage &= ~1u;
-
- for (int i = MESA_SHADER_VERTEX; i < MESA_SHADER_FRAGMENT; i++) {
- pipeline->urb.push_size[i] =
- (pipeline->active_stages & (1 << i)) ? size_per_stage : 0;
- used_kb += pipeline->urb.push_size[i];
- assert(used_kb <= push_constant_kb);
- }
-
- pipeline->urb.push_size[MESA_SHADER_FRAGMENT] =
- push_constant_kb - used_kb;
}
static void
diff --git a/src/intel/vulkan/anv_private.h b/src/intel/vulkan/anv_private.h
index 33cbff9..80e768d 100644
--- a/src/intel/vulkan/anv_private.h
+++ b/src/intel/vulkan/anv_private.h
@@ -1177,6 +1177,7 @@ struct anv_cmd_state {
uint32_t restart_index;
struct anv_vertex_binding vertex_bindings[MAX_VBS];
struct anv_descriptor_set * descriptors[MAX_SETS];
+ VkShaderStageFlags push_constant_stages;
struct anv_push_constants * push_constants[MESA_SHADER_STAGES];
struct anv_state binding_tables[MESA_SHADER_STAGES];
struct anv_state samplers[MESA_SHADER_STAGES];
@@ -1411,7 +1412,6 @@ struct anv_pipeline {
uint32_t scratch_start[MESA_SHADER_STAGES];
uint32_t total_scratch;
struct {
- uint8_t push_size[MESA_SHADER_FRAGMENT + 1];
uint32_t start[MESA_SHADER_GEOMETRY + 1];
uint32_t size[MESA_SHADER_GEOMETRY + 1];
uint32_t entries[MESA_SHADER_GEOMETRY + 1];
diff --git a/src/intel/vulkan/genX_cmd_buffer.c b/src/intel/vulkan/genX_cmd_buffer.c
index ee47c29..f5e3530 100644
--- a/src/intel/vulkan/genX_cmd_buffer.c
+++ b/src/intel/vulkan/genX_cmd_buffer.c
@@ -274,6 +274,72 @@ void genX(CmdPipelineBarrier)(
}
}
+static void
+cmd_buffer_alloc_push_constants(struct anv_cmd_buffer *cmd_buffer)
+{
+ VkShaderStageFlags stages = cmd_buffer->state.pipeline->active_stages;
+
+ /* In order to avoid thrash, we assume that vertex and fragment stages
+ * always exist. In the rare case where one is missing *and* the other
+ * uses push concstants, this may be suboptimal. However, avoiding stalls
+ * seems more important.
+ */
+ stages |= VK_SHADER_STAGE_FRAGMENT_BIT | VK_SHADER_STAGE_VERTEX_BIT;
+
+ if (stages == cmd_buffer->state.push_constant_stages)
+ return;
+
+#if GEN_GEN >= 8
+ const unsigned push_constant_kb = 32;
+#elif GEN_IS_HASWELL
+ const unsigned push_constant_kb = cmd_buffer->device->info.gt == 3 ? 32 : 16;
+#else
+ const unsigned push_constant_kb = 16;
+#endif
+
+ const unsigned num_stages =
+ _mesa_bitcount(stages & VK_SHADER_STAGE_ALL_GRAPHICS);
+ unsigned size_per_stage = push_constant_kb / num_stages;
+
+ /* Broadwell+ and Haswell gt3 require that the push constant sizes be in
+ * units of 2KB. Incidentally, these are the same platforms that have
+ * 32KB worth of push constant space.
+ */
+ if (push_constant_kb == 32)
+ size_per_stage &= ~1u;
+
+ uint32_t kb_used = 0;
+ for (int i = MESA_SHADER_VERTEX; i < MESA_SHADER_FRAGMENT; i++) {
+ unsigned push_size = (stages & (1 << i)) ? size_per_stage : 0;
+ anv_batch_emit(&cmd_buffer->batch,
+ GENX(3DSTATE_PUSH_CONSTANT_ALLOC_VS), alloc) {
+ alloc._3DCommandSubOpcode = 18 + i;
+ alloc.ConstantBufferOffset = (push_size > 0) ? kb_used : 0;
+ alloc.ConstantBufferSize = push_size;
+ }
+ kb_used += push_size;
+ }
+
+ anv_batch_emit(&cmd_buffer->batch,
+ GENX(3DSTATE_PUSH_CONSTANT_ALLOC_PS), alloc) {
+ alloc.ConstantBufferOffset = kb_used;
+ alloc.ConstantBufferSize = push_constant_kb - kb_used;
+ }
+
+ cmd_buffer->state.push_constant_stages = stages;
+
+ /* From the BDW PRM for 3DSTATE_PUSH_CONSTANT_ALLOC_VS:
+ *
+ * "The 3DSTATE_CONSTANT_VS must be reprogrammed prior to
+ * the next 3DPRIMITIVE command after programming the
+ * 3DSTATE_PUSH_CONSTANT_ALLOC_VS"
+ *
+ * Since 3DSTATE_PUSH_CONSTANT_ALLOC_VS is programmed as part of
+ * pipeline setup, we need to dirty push constants.
+ */
+ cmd_buffer->state.push_constants_dirty |= VK_SHADER_STAGE_ALL_GRAPHICS;
+}
+
static uint32_t
cmd_buffer_flush_push_constants(struct anv_cmd_buffer *cmd_buffer)
{
@@ -384,16 +450,10 @@ genX(cmd_buffer_flush_state)(struct anv_cmd_buffer *cmd_buffer)
anv_batch_emit_batch(&cmd_buffer->batch, &pipeline->batch);
- /* From the BDW PRM for 3DSTATE_PUSH_CONSTANT_ALLOC_VS:
- *
- * "The 3DSTATE_CONSTANT_VS must be reprogrammed prior to
- * the next 3DPRIMITIVE command after programming the
- * 3DSTATE_PUSH_CONSTANT_ALLOC_VS"
- *
- * Since 3DSTATE_PUSH_CONSTANT_ALLOC_VS is programmed as part of
- * pipeline setup, we need to dirty push constants.
+ /* If the pipeline changed, we may need to re-allocate push constant
+ * space in the URB.
*/
- cmd_buffer->state.push_constants_dirty |= VK_SHADER_STAGE_ALL_GRAPHICS;
+ cmd_buffer_alloc_push_constants(cmd_buffer);
}
#if GEN_GEN <= 7
diff --git a/src/intel/vulkan/genX_pipeline_util.h b/src/intel/vulkan/genX_pipeline_util.h
index ecbe436..2ed55e0 100644
--- a/src/intel/vulkan/genX_pipeline_util.h
+++ b/src/intel/vulkan/genX_pipeline_util.h
@@ -205,18 +205,6 @@ emit_urb_setup(struct anv_pipeline *pipeline)
}
#endif
- unsigned push_start = 0;
- for (int i = MESA_SHADER_VERTEX; i <= MESA_SHADER_FRAGMENT; i++) {
- unsigned push_size = pipeline->urb.push_size[i];
- anv_batch_emit(&pipeline->batch,
- GENX(3DSTATE_PUSH_CONSTANT_ALLOC_VS), alloc) {
- alloc._3DCommandSubOpcode = 18 + i;
- alloc.ConstantBufferOffset = (push_size > 0) ? push_start : 0;
- alloc.ConstantBufferSize = push_size;
- }
- push_start += pipeline->urb.push_size[i];
- }
-
for (int i = MESA_SHADER_VERTEX; i <= MESA_SHADER_GEOMETRY; i++) {
anv_batch_emit(&pipeline->batch, GENX(3DSTATE_URB_VS), urb) {
urb._3DCommandSubOpcode = 48 + i;
--
2.5.0.400.gff86faf
More information about the mesa-dev
mailing list