summaryrefslogtreecommitdiffstats
path: root/src
diff options
context:
space:
mode:
authorRob Clark <[email protected]>2018-03-06 08:30:41 -0500
committerRob Clark <[email protected]>2018-03-31 15:10:56 -0400
commit4f783838099eec11d38a13dd021d500d812cf410 (patch)
treee4ae15b51b29fa38d0d6bb34b125e5c98861e263 /src
parent96e7927fb2234bd2358a0da11ca0ce47e3b3c82b (diff)
freedreno/ir3: add local_group_size
Signed-off-by: Rob Clark <[email protected]>
Diffstat (limited to 'src')
-rw-r--r--src/gallium/drivers/freedreno/ir3/ir3_compiler_nir.c5
-rw-r--r--src/gallium/drivers/freedreno/ir3/ir3_shader.c4
-rw-r--r--src/gallium/drivers/freedreno/ir3/ir3_shader.h5
3 files changed, 12 insertions, 2 deletions
diff --git a/src/gallium/drivers/freedreno/ir3/ir3_compiler_nir.c b/src/gallium/drivers/freedreno/ir3/ir3_compiler_nir.c
index a3e82ab593b..bca1a822a9f 100644
--- a/src/gallium/drivers/freedreno/ir3/ir3_compiler_nir.c
+++ b/src/gallium/drivers/freedreno/ir3/ir3_compiler_nir.c
@@ -2194,6 +2194,11 @@ emit_intrinsic(struct ir3_context *ctx, nir_intrinsic_instr *intr)
dst[i] = create_driver_param(ctx, IR3_DP_NUM_WORK_GROUPS_X + i);
}
break;
+ case nir_intrinsic_load_local_group_size:
+ for (int i = 0; i < intr->num_components; i++) {
+ dst[i] = create_driver_param(ctx, IR3_DP_LOCAL_GROUP_SIZE_X + i);
+ }
+ break;
case nir_intrinsic_discard_if:
case nir_intrinsic_discard: {
struct ir3_instruction *cond, *kill;
diff --git a/src/gallium/drivers/freedreno/ir3/ir3_shader.c b/src/gallium/drivers/freedreno/ir3/ir3_shader.c
index 555c654374e..8ed7f5673c8 100644
--- a/src/gallium/drivers/freedreno/ir3/ir3_shader.c
+++ b/src/gallium/drivers/freedreno/ir3/ir3_shader.c
@@ -949,7 +949,9 @@ ir3_emit_cs_consts(const struct ir3_shader_variant *v, struct fd_ringbuffer *rin
[IR3_DP_NUM_WORK_GROUPS_X] = info->grid[0],
[IR3_DP_NUM_WORK_GROUPS_Y] = info->grid[1],
[IR3_DP_NUM_WORK_GROUPS_Z] = info->grid[2],
- /* do we need work-group-size? */
+ [IR3_DP_LOCAL_GROUP_SIZE_X] = info->block[0],
+ [IR3_DP_LOCAL_GROUP_SIZE_Y] = info->block[1],
+ [IR3_DP_LOCAL_GROUP_SIZE_Z] = info->block[2],
};
ctx->emit_const(ring, SHADER_COMPUTE, offset * 4, 0,
diff --git a/src/gallium/drivers/freedreno/ir3/ir3_shader.h b/src/gallium/drivers/freedreno/ir3/ir3_shader.h
index 040ea4cd9d8..272368c6980 100644
--- a/src/gallium/drivers/freedreno/ir3/ir3_shader.h
+++ b/src/gallium/drivers/freedreno/ir3/ir3_shader.h
@@ -44,12 +44,15 @@ enum ir3_driver_param {
IR3_DP_NUM_WORK_GROUPS_X = 0,
IR3_DP_NUM_WORK_GROUPS_Y = 1,
IR3_DP_NUM_WORK_GROUPS_Z = 2,
+ IR3_DP_LOCAL_GROUP_SIZE_X = 4,
+ IR3_DP_LOCAL_GROUP_SIZE_Y = 5,
+ IR3_DP_LOCAL_GROUP_SIZE_Z = 6,
/* NOTE: gl_NumWorkGroups should be vec4 aligned because
* glDispatchComputeIndirect() needs to load these from
* the info->indirect buffer. Keep that in mind when/if
* adding any addition CS driver params.
*/
- IR3_DP_CS_COUNT = 4, /* must be aligned to vec4 */
+ IR3_DP_CS_COUNT = 8, /* must be aligned to vec4 */
/* vertex shader driver params: */
IR3_DP_VTXID_BASE = 0,