[Mesa-dev] [PATCH v4 09/12] i965: Support new local ID push constant & cross-thread constants
Jordan Justen
jordan.l.justen at intel.com
Wed Jun 1 22:04:16 UTC 2016
The cross thread constant support appears on Haswell. It allows us to
upload a set of uniform data for all threads without duplicating it
per thread.
We also support per-thread data which allows us to store a per-thread
ID in one of the uniforms that can be used to calculate the
gl_LocalInvocationIndex and gl_LocalInvocationID variables.
v4:
* Support the old local ID push constant layout as well (Jason)
Signed-off-by: Jordan Justen <jordan.l.justen at intel.com>
---
src/mesa/drivers/dri/i965/brw_defines.h | 3 +
src/mesa/drivers/dri/i965/gen7_cs_state.c | 99 +++++++++++++++++--------------
2 files changed, 56 insertions(+), 46 deletions(-)
diff --git a/src/mesa/drivers/dri/i965/brw_defines.h b/src/mesa/drivers/dri/i965/brw_defines.h
index 4eb6b1f..e7d1a9f 100644
--- a/src/mesa/drivers/dri/i965/brw_defines.h
+++ b/src/mesa/drivers/dri/i965/brw_defines.h
@@ -2943,6 +2943,9 @@ enum brw_wm_barycentric_interp_mode {
# define MEDIA_GPGPU_THREAD_COUNT_MASK INTEL_MASK(7, 0)
# define GEN8_MEDIA_GPGPU_THREAD_COUNT_SHIFT 0
# define GEN8_MEDIA_GPGPU_THREAD_COUNT_MASK INTEL_MASK(9, 0)
+/* GEN7 DW6, GEN8+ DW7 */
+# define CROSS_THREAD_READ_LENGTH_SHIFT 0
+# define CROSS_THREAD_READ_LENGTH_MASK INTEL_MASK(7, 0)
#define MEDIA_STATE_FLUSH 0x7004
#define GPGPU_WALKER 0x7105
/* GEN7 DW0 */
diff --git a/src/mesa/drivers/dri/i965/gen7_cs_state.c b/src/mesa/drivers/dri/i965/gen7_cs_state.c
index 619edfb..2fee02d 100644
--- a/src/mesa/drivers/dri/i965/gen7_cs_state.c
+++ b/src/mesa/drivers/dri/i965/gen7_cs_state.c
@@ -42,7 +42,6 @@ brw_upload_cs_state(struct brw_context *brw)
uint32_t offset;
uint32_t *desc = (uint32_t*) brw_state_batch(brw, AUB_TRACE_SURFACE_STATE,
8 * 4, 64, &offset);
- struct gl_program *prog = (struct gl_program *) brw->compute_program;
struct brw_stage_state *stage_state = &brw->cs.base;
struct brw_cs_prog_data *cs_prog_data = brw->cs.prog_data;
struct brw_stage_prog_data *prog_data = &cs_prog_data->base;
@@ -59,16 +58,6 @@ brw_upload_cs_state(struct brw_context *brw)
prog_data->binding_table.size_bytes,
32, &stage_state->bind_bo_offset);
- unsigned local_id_dwords = 0;
-
- if (prog->SystemValuesRead & SYSTEM_BIT_LOCAL_INVOCATION_ID)
- local_id_dwords = cs_prog_data->local_invocation_id_regs * 8;
-
- unsigned push_constant_data_size =
- (prog_data->nr_params + local_id_dwords) * sizeof(gl_constant_value);
- unsigned reg_aligned_constant_size = ALIGN(push_constant_data_size, 32);
- unsigned push_constant_regs = reg_aligned_constant_size / 32;
-
uint32_t dwords = brw->gen < 8 ? 8 : 9;
BEGIN_BATCH(dwords);
OUT_BATCH(MEDIA_VFE_STATE << 16 | (dwords - 2));
@@ -118,7 +107,8 @@ brw_upload_cs_state(struct brw_context *brw)
* Note: The constant data is built in brw_upload_cs_push_constants below.
*/
const uint32_t vfe_curbe_allocation =
- push_constant_regs * cs_prog_data->threads;
+ ALIGN(cs_prog_data->push.per_thread.regs * cs_prog_data->threads +
+ cs_prog_data->push.cross_thread.regs, 2);
OUT_BATCH(SET_FIELD(vfe_urb_allocation, MEDIA_VFE_STATE_URB_ALLOC) |
SET_FIELD(vfe_curbe_allocation, MEDIA_VFE_STATE_CURBE_ALLOC));
OUT_BATCH(0);
@@ -126,11 +116,11 @@ brw_upload_cs_state(struct brw_context *brw)
OUT_BATCH(0);
ADVANCE_BATCH();
- if (reg_aligned_constant_size > 0) {
+ if (cs_prog_data->push.total.size > 0) {
BEGIN_BATCH(4);
OUT_BATCH(MEDIA_CURBE_LOAD << 16 | (4 - 2));
OUT_BATCH(0);
- OUT_BATCH(ALIGN(reg_aligned_constant_size * cs_prog_data->threads, 64));
+ OUT_BATCH(ALIGN(cs_prog_data->push.total.size, 64));
OUT_BATCH(stage_state->push_const_offset);
ADVANCE_BATCH();
}
@@ -149,7 +139,8 @@ brw_upload_cs_state(struct brw_context *brw)
desc[dw++] = stage_state->sampler_offset |
((stage_state->sampler_count + 3) / 4);
desc[dw++] = stage_state->bind_bo_offset;
- desc[dw++] = SET_FIELD(push_constant_regs, MEDIA_CURBE_READ_LENGTH);
+ desc[dw++] = SET_FIELD(cs_prog_data->push.per_thread.regs,
+ MEDIA_CURBE_READ_LENGTH);
const uint32_t media_threads =
brw->gen >= 8 ?
SET_FIELD(cs_prog_data->threads, GEN8_MEDIA_GPGPU_THREAD_COUNT) :
@@ -171,6 +162,10 @@ brw_upload_cs_state(struct brw_context *brw)
SET_FIELD(slm_size, MEDIA_SHARED_LOCAL_MEMORY_SIZE) |
media_threads;
+ desc[dw++] =
+ SET_FIELD(cs_prog_data->push.cross_thread.regs,
+ CROSS_THREAD_READ_LENGTH);
+
BEGIN_BATCH(4);
OUT_BATCH(MEDIA_INTERFACE_DESCRIPTOR_LOAD << 16 | (4 - 2));
OUT_BATCH(0);
@@ -213,10 +208,6 @@ brw_upload_cs_push_constants(struct brw_context *brw,
struct gl_context *ctx = &brw->ctx;
const struct brw_stage_prog_data *prog_data =
(struct brw_stage_prog_data*) cs_prog_data;
- unsigned local_id_dwords = 0;
-
- if (prog->SystemValuesRead & SYSTEM_BIT_LOCAL_INVOCATION_ID)
- local_id_dwords = cs_prog_data->local_invocation_id_regs * 8;
/* Updates the ParamaterValues[i] pointers for all parameters of the
* basic type of PROGRAM_STATE_VAR.
@@ -224,41 +215,57 @@ brw_upload_cs_push_constants(struct brw_context *brw,
/* XXX: Should this happen somewhere before to get our state flag set? */
_mesa_load_state_parameters(ctx, prog->Parameters);
- if (prog_data->nr_params == 0 && local_id_dwords == 0) {
+ if (cs_prog_data->push.total.size == 0) {
stage_state->push_const_size = 0;
- } else {
- gl_constant_value *param;
- unsigned i, t;
-
- const unsigned push_constant_data_size =
- (local_id_dwords + prog_data->nr_params) * sizeof(gl_constant_value);
- const unsigned reg_aligned_constant_size = ALIGN(push_constant_data_size, 32);
- const unsigned param_aligned_count =
- reg_aligned_constant_size / sizeof(*param);
+ return;
+ }
- param = (gl_constant_value*)
- brw_state_batch(brw, type,
- ALIGN(reg_aligned_constant_size *
- cs_prog_data->threads, 64),
- 64, &stage_state->push_const_offset);
- assert(param);
- STATIC_ASSERT(sizeof(gl_constant_value) == sizeof(float));
+ gl_constant_value *param;
+ param = (gl_constant_value*)
+ brw_state_batch(brw, type,
+ ALIGN(cs_prog_data->push.total.size, 64),
+ 64, &stage_state->push_const_offset);
+ assert(param);
+
+ STATIC_ASSERT(sizeof(gl_constant_value) == sizeof(float));
+
+ if (cs_prog_data->push.cross_thread.size > 0) {
+ gl_constant_value *param_copy = param;
+ assert(cs_prog_data->thread_local_id_index < 0 ||
+ cs_prog_data->thread_local_id_index >=
+ cs_prog_data->push.cross_thread.dwords);
+ for (unsigned i = 0;
+ i < cs_prog_data->push.cross_thread.dwords;
+ i++) {
+ param_copy[i] = *prog_data->param[i];
+ }
+ }
+ gl_constant_value thread_id;
+ if (cs_prog_data->push.per_thread.size > 0) {
brw_cs_fill_local_id_payload(cs_prog_data, param, cs_prog_data->threads,
- reg_aligned_constant_size);
-
- /* _NEW_PROGRAM_CONSTANTS */
- for (t = 0; t < cs_prog_data->threads; t++) {
- gl_constant_value *next_param =
- ¶m[t * param_aligned_count + local_id_dwords];
- for (i = 0; i < prog_data->nr_params; i++) {
- next_param[i] = *prog_data->param[i];
+ cs_prog_data->push.per_thread.size);
+ for (unsigned t = 0; t < cs_prog_data->threads; t++) {
+ unsigned dst =
+ 8 * (cs_prog_data->push.per_thread.regs * t +
+ cs_prog_data->push.cross_thread.regs +
+ cs_prog_data->local_invocation_id_regs);
+ unsigned src = cs_prog_data->push.cross_thread.dwords;
+ for ( ; src < prog_data->nr_params; src++, dst++) {
+ if (src != cs_prog_data->thread_local_id_index)
+ param[dst] = *prog_data->param[src];
+ else {
+ thread_id.u = t * cs_prog_data->simd_size;
+ param[dst] = thread_id;
+ }
}
}
-
- stage_state->push_const_size = ALIGN(prog_data->nr_params, 8) / 8;
}
+
+ stage_state->push_const_size =
+ cs_prog_data->push.cross_thread.regs +
+ cs_prog_data->push.per_thread.regs;
}
--
2.8.1
More information about the mesa-dev
mailing list