Mesa (master): freedreno/ir3: add local_group_size

[Rob Clark]

Module: Mesa
Branch: master
Commit: 4f783838099eec11d38a13dd021d500d812cf410
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=4f783838099eec11d38a13dd021d500d812cf410

Author: Rob Clark <robdclark at gmail.com>
Date:   Tue Mar  6 08:30:41 2018 -0500

freedreno/ir3: add local_group_size

Signed-off-by: Rob Clark <robdclark at gmail.com>

---

 src/gallium/drivers/freedreno/ir3/ir3_compiler_nir.c | 5 +++++
 src/gallium/drivers/freedreno/ir3/ir3_shader.c       | 4 +++-
 src/gallium/drivers/freedreno/ir3/ir3_shader.h       | 5 ++++-
 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 a3e82ab593..bca1a822a9 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 555c654374..8ed7f5673c 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 040ea4cd9d..272368c698 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,