gallium: add a new interface for pipe_context::launch_grid()

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.

Changes from v2:
 - correctly initialize pipe_grid_info for nv50/nvc0

Signed-off-by: Samuel Pitoiset <[email protected]>
Reviewed-by: Marek Olšák <[email protected]>
Reviewed-by: Ilia Mirkin <[email protected]>

3 files changed, 18 insertions, 13 deletions

diff --git a/src/gallium/drivers/nouveau/nv50/nv50_compute.c b/src/gallium/drivers/nouveau/nv50/nv50_compute.c
index 6d23fd66945..04488d6d0a6 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 342ec96d62c..23d82103332 100644
--- a/src/gallium/drivers/nouveau/nv50/nv50_context.h
+++ b/src/gallium/drivers/nouveau/nv50/nv50_context.h
@@ -322,7 +322,6 @@ 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 79c7023b2d4..be19c0fdc85 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 = {};
    uint32_t mask;
    uint32_t input[3];
    const uint block[3] = { 32, 1, 1 };
    const uint grid[3] = { screen->MPsInTP, screen->TPs, 1 };
-   int c;
+   int c, i;
 
    if (unlikely(!screen->pm.prog)) {
       struct nv50_program *prog = CALLOC_STRUCT(nv50_program);
@@ -262,7 +263,14 @@ nv50_hw_sm_end_query(struct nv50_context *nv50, struct nv50_hw_query *hq)
    pipe->bind_compute_state(pipe, screen->pm.prog);
    input[0] = hq->bo->offset + hq->base_offset;
    input[1] = hq->sequence;
-   pipe->launch_grid(pipe, block, grid, 0, input);
+
+   for (i = 0; i < 3; i++) {
+      info.block[i] = block[i];
+      info.grid[i] = grid[i];
+   }
+   info.pc = 0;
+   info.input = input;
+   pipe->launch_grid(pipe, &info);
 
    nouveau_bufctx_reset(nv50->bufctx_cp, NV50_BIND_CP_QUERY);