static void
nv50_compute_upload_input(struct nv50_context *nv50, const uint32_t *input)
{
struct nv50_screen *screen = nv50->screen;
struct nouveau_pushbuf *push = screen->base.pushbuf;
unsigned size = align(nv50->compprog->parm_size, 0x4);
BEGIN_NV04(push, NV50_COMPUTE(USER_PARAM_COUNT), 1);
PUSH_DATA (push, (size / 4) << 8);
if (size) {
struct nouveau_mm_allocation *mm;
struct nouveau_bo *bo = NULL;
unsigned offset;
mm = nouveau_mm_allocate(screen->base.mm_GART, size, &bo, &offset);
assert(mm);
nouveau_bo_map(bo, 0, screen->base.client);
memcpy(bo->map + offset, input, size);
nouveau_bufctx_refn(nv50->bufctx, 0, bo, NOUVEAU_BO_GART | NOUVEAU_BO_RD);
nouveau_pushbuf_bufctx(push, nv50->bufctx);
nouveau_pushbuf_validate(push);
BEGIN_NV04(push, NV50_COMPUTE(USER_PARAM(0)), size / 4);
nouveau_pushbuf_data(push, bo, offset, size);
nouveau_fence_work(screen->base.fence.current, nouveau_mm_free_work, mm);
nouveau_bo_ref(NULL, &bo);
nouveau_bufctx_reset(nv50->bufctx, 0);
}
}
void
nvc0_query_pushbuf_submit(struct nouveau_pushbuf *push,
struct pipe_query *pq, unsigned result_offset)
{
struct nvc0_query *q = nvc0_query(pq);
#define NVC0_IB_ENTRY_1_NO_PREFETCH (1 << (31 - 8))
nouveau_pushbuf_space(push, 0, 0, 1);
nouveau_pushbuf_data(push, q->bo, q->offset + result_offset, 4 |
NVC0_IB_ENTRY_1_NO_PREFETCH);
}
void
nv50_query_pushbuf_submit(struct nouveau_pushbuf *push,
struct pipe_query *pq, unsigned result_offset)
{
struct nv50_query *q = nv50_query(pq);
/* XXX: does this exist ? */
#define NV50_IB_ENTRY_1_NO_PREFETCH (0 << (31 - 8))
nouveau_pushbuf_space(push, 0, 0, 1);
nouveau_pushbuf_data(push, q->bo, q->offset + result_offset, 4 |
NV50_IB_ENTRY_1_NO_PREFETCH);
}
void
nvc0_launch_grid(struct pipe_context *pipe, const struct pipe_grid_info *info)
{
struct nvc0_context *nvc0 = nvc0_context(pipe);
struct nouveau_pushbuf *push = nvc0->base.pushbuf;
struct nvc0_program *cp = nvc0->compprog;
int ret;
ret = !nvc0_state_validate_cp(nvc0, ~0);
if (ret) {
NOUVEAU_ERR("Failed to launch grid !\n");
return;
}
nvc0_compute_upload_input(nvc0, info);
BEGIN_NVC0(push, NVC0_CP(CP_START_ID), 1);
PUSH_DATA (push, nvc0_program_symbol_offset(cp, info->pc));
BEGIN_NVC0(push, NVC0_CP(LOCAL_POS_ALLOC), 3);
PUSH_DATA (push, (cp->hdr[1] & 0xfffff0) + align(cp->cp.lmem_size, 0x10));
PUSH_DATA (push, 0);
PUSH_DATA (push, 0x800); /* WARP_CSTACK_SIZE */
BEGIN_NVC0(push, NVC0_CP(SHARED_SIZE), 3);
PUSH_DATA (push, align(cp->cp.smem_size, 0x100));
PUSH_DATA (push, info->block[0] * info->block[1] * info->block[2]);
PUSH_DATA (push, cp->num_barriers);
BEGIN_NVC0(push, NVC0_CP(CP_GPR_ALLOC), 1);
PUSH_DATA (push, cp->num_gprs);
/* launch preliminary setup */
BEGIN_NVC0(push, NVC0_CP(GRIDID), 1);
PUSH_DATA (push, 0x1);
BEGIN_NVC0(push, SUBC_CP(0x036c), 1);
PUSH_DATA (push, 0);
BEGIN_NVC0(push, NVC0_CP(FLUSH), 1);
PUSH_DATA (push, NVC0_COMPUTE_FLUSH_GLOBAL | NVC0_COMPUTE_FLUSH_UNK8);
/* block setup */
BEGIN_NVC0(push, NVC0_CP(BLOCKDIM_YX), 2);
PUSH_DATA (push, (info->block[1] << 16) | info->block[0]);
PUSH_DATA (push, info->block[2]);
if (unlikely(info->indirect)) {
struct nv04_resource *res = nv04_resource(info->indirect);
uint32_t offset = res->offset + info->indirect_offset;
unsigned macro = NVC0_CP_MACRO_LAUNCH_GRID_INDIRECT;
nouveau_pushbuf_space(push, 16, 0, 1);
PUSH_REFN(push, res->bo, NOUVEAU_BO_RD | res->domain);
PUSH_DATA(push, NVC0_FIFO_PKHDR_1I(1, macro, 3));
nouveau_pushbuf_data(push, res->bo, offset,
NVC0_IB_ENTRY_1_NO_PREFETCH | 3 * 4);
} else {
/* grid setup */
BEGIN_NVC0(push, NVC0_CP(GRIDDIM_YX), 2);
PUSH_DATA (push, (info->grid[1] << 16) | info->grid[0]);
PUSH_DATA (push, info->grid[2]);
/* kernel launching */
BEGIN_NVC0(push, NVC0_CP(COMPUTE_BEGIN), 1);
PUSH_DATA (push, 0);
BEGIN_NVC0(push, SUBC_CP(0x0a08), 1);
PUSH_DATA (push, 0);
BEGIN_NVC0(push, NVC0_CP(LAUNCH), 1);
PUSH_DATA (push, 0x1000);
BEGIN_NVC0(push, NVC0_CP(COMPUTE_END), 1);
PUSH_DATA (push, 0);
BEGIN_NVC0(push, SUBC_CP(0x0360), 1);
PUSH_DATA (push, 0x1);
}
/* TODO: Not sure if this is really necessary. */
nvc0_compute_invalidate_surfaces(nvc0, 5);
nouveau_bufctx_reset(nvc0->bufctx_cp, NVC0_BIND_CP_SUF);
nvc0->dirty_cp |= NVC0_NEW_CP_SURFACES;
nvc0->images_dirty[5] |= nvc0->images_valid[5];
}