This introduces pipe_grid_info which contains all information to
describe a launch_grid call. This will be used to implement indirect
compute in the same fashion as indirect draw.
Signed-off-by: Samuel Pitoiset <[email protected]>
Reviewed-by: Marek Olšák <[email protected]>
Reviewed-by: Ilia Mirkin <[email protected]>
---
src/gallium/drivers/ilo/ilo_gpgpu.c | 8 ++----
src/gallium/drivers/nouveau/nv50/nv50_compute.c | 16 +++++------
src/gallium/drivers/nouveau/nv50/nv50_context.h | 3 +-
.../drivers/nouveau/nv50/nv50_query_hw_sm.c | 12 ++++++--
src/gallium/drivers/nouveau/nvc0/nvc0_compute.c | 19 ++++++-------
src/gallium/drivers/nouveau/nvc0/nvc0_context.h | 6 ++--
.../drivers/nouveau/nvc0/nvc0_query_hw_sm.c | 12 ++++++--
src/gallium/drivers/nouveau/nvc0/nve4_compute.c | 10 +++----
src/gallium/drivers/r600/evergreen_compute.c | 19 ++++++-------
src/gallium/drivers/radeonsi/si_compute.c | 33
+++++++++++-----------
src/gallium/include/pipe/p_context.h | 17 ++---------
src/gallium/include/pipe/p_state.h | 27
++++++++++++++++++
src/gallium/state_trackers/clover/core/kernel.cpp | 13 +++++----
src/gallium/tests/trivial/compute.c | 11 +++++++-
14 files changed, 117 insertions(+), 89 deletions(-)
diff --git a/src/gallium/drivers/ilo/ilo_gpgpu.c
b/src/gallium/drivers/ilo/ilo_gpgpu.c
index b741590..ab165b6 100644
--- a/src/gallium/drivers/ilo/ilo_gpgpu.c
+++ b/src/gallium/drivers/ilo/ilo_gpgpu.c
@@ -79,9 +79,7 @@ launch_grid(struct ilo_context *ilo,
}
static void
-ilo_launch_grid(struct pipe_context *pipe,
- const uint *block_layout, const uint *grid_layout,
- uint32_t pc, const void *input)
+ilo_launch_grid(struct pipe_context *pipe, const struct pipe_grid_info
*info)
{
struct ilo_context *ilo = ilo_context(pipe);
struct ilo_shader_state *cs = ilo->state_vector.cs;
@@ -92,13 +90,13 @@ ilo_launch_grid(struct pipe_context *pipe,
input_buf.buffer_size =
ilo_shader_get_kernel_param(cs, ILO_KERNEL_CS_INPUT_SIZE);
if (input_buf.buffer_size) {
- u_upload_data(ilo->uploader, 0, input_buf.buffer_size, 16, input,
+ u_upload_data(ilo->uploader, 0, input_buf.buffer_size, 16,
info->input,
&input_buf.buffer_offset, &input_buf.buffer);
}
ilo_shader_cache_upload(ilo->shader_cache, &ilo->cp->builder);
- launch_grid(ilo, block_layout, grid_layout, &input_buf, pc);
+ launch_grid(ilo, info->block, info->grid, &input_buf, info->pc);
ilo_render_invalidate_hw(ilo->render);
diff --git a/src/gallium/drivers/nouveau/nv50/nv50_compute.c
b/src/gallium/drivers/nouveau/nv50/nv50_compute.c
index 6d23fd6..04488d6 100644
--- a/src/gallium/drivers/nouveau/nv50/nv50_compute.c
+++ b/src/gallium/drivers/nouveau/nv50/nv50_compute.c
@@ -270,13 +270,11 @@ nv50_compute_find_symbol(struct nv50_context *nv50,
uint32_t label)
}
void
-nv50_launch_grid(struct pipe_context *pipe,
- const uint *block_layout, const uint *grid_layout,
- uint32_t label, const void *input)
+nv50_launch_grid(struct pipe_context *pipe, const struct pipe_grid_info
*info)
{
struct nv50_context *nv50 = nv50_context(pipe);
struct nouveau_pushbuf *push = nv50->base.pushbuf;
- unsigned block_size = block_layout[0] * block_layout[1] *
block_layout[2];
+ unsigned block_size = info->block[0] * info->block[1] *
info->block[2];
struct nv50_program *cp = nv50->compprog;
bool ret;
@@ -286,10 +284,10 @@ nv50_launch_grid(struct pipe_context *pipe,
return;
}
- nv50_compute_upload_input(nv50, input);
+ nv50_compute_upload_input(nv50, info->input);
BEGIN_NV04(push, NV50_COMPUTE(CP_START_ID), 1);
- PUSH_DATA (push, nv50_compute_find_symbol(nv50, label));
+ PUSH_DATA (push, nv50_compute_find_symbol(nv50, info->pc));
BEGIN_NV04(push, NV50_COMPUTE(SHARED_SIZE), 1);
PUSH_DATA (push, align(cp->cp.smem_size + cp->parm_size + 0x10,
0x40));
@@ -298,14 +296,14 @@ nv50_launch_grid(struct pipe_context *pipe,
/* grid/block setup */
BEGIN_NV04(push, NV50_COMPUTE(BLOCKDIM_XY), 2);
- PUSH_DATA (push, block_layout[1] << 16 | block_layout[0]);
- PUSH_DATA (push, block_layout[2]);
+ PUSH_DATA (push, info->block[1] << 16 | info->block[0]);
+ PUSH_DATA (push, info->block[2]);
BEGIN_NV04(push, NV50_COMPUTE(BLOCK_ALLOC), 1);
PUSH_DATA (push, 1 << 16 | block_size);
BEGIN_NV04(push, NV50_COMPUTE(BLOCKDIM_LATCH), 1);
PUSH_DATA (push, 1);
BEGIN_NV04(push, NV50_COMPUTE(GRIDDIM), 1);
- PUSH_DATA (push, grid_layout[1] << 16 | grid_layout[0]);
+ PUSH_DATA (push, info->grid[1] << 16 | info->grid[0]);
BEGIN_NV04(push, NV50_COMPUTE(GRIDID), 1);
PUSH_DATA (push, 1);
diff --git a/src/gallium/drivers/nouveau/nv50/nv50_context.h
b/src/gallium/drivers/nouveau/nv50/nv50_context.h
index 712d00e..cb18df3 100644
--- a/src/gallium/drivers/nouveau/nv50/nv50_context.h
+++ b/src/gallium/drivers/nouveau/nv50/nv50_context.h
@@ -324,7 +324,6 @@ nv98_video_buffer_create(struct pipe_context *pipe,
/* nv50_compute.c */
void
-nv50_launch_grid(struct pipe_context *, const uint *, const uint *,
- uint32_t, const void *);
+nv50_launch_grid(struct pipe_context *, const struct pipe_grid_info *);
#endif
diff --git a/src/gallium/drivers/nouveau/nv50/nv50_query_hw_sm.c
b/src/gallium/drivers/nouveau/nv50/nv50_query_hw_sm.c
index 79c7023..da3970e 100644
--- a/src/gallium/drivers/nouveau/nv50/nv50_query_hw_sm.c
+++ b/src/gallium/drivers/nouveau/nv50/nv50_query_hw_sm.c
@@ -218,11 +218,12 @@ nv50_hw_sm_end_query(struct nv50_context *nv50,
struct nv50_hw_query *hq)
struct pipe_context *pipe = &nv50->base.pipe;
struct nouveau_pushbuf *push = nv50->base.pushbuf;
struct nv50_hw_sm_query *hsq = nv50_hw_sm_query(hq);
+ struct pipe_grid_info info;
