X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fdrivers%2Fr600%2Fevergreen_compute.c;h=ce17d3a61ece49e31134cd7d3bbcd7bee0aefe31;hb=480e1463057c70ca1b197cbba5a9b1c153d8b052;hp=fe9e2f98ce0495cfce7b559351badf909eb4ef29;hpb=d0349c91c85c929de0502c4aeddae5b0ee3a1b7b;p=mesa.git diff --git a/src/gallium/drivers/r600/evergreen_compute.c b/src/gallium/drivers/r600/evergreen_compute.c index fe9e2f98ce0..ce17d3a61ec 100644 --- a/src/gallium/drivers/r600/evergreen_compute.c +++ b/src/gallium/drivers/r600/evergreen_compute.c @@ -101,15 +101,13 @@ static void evergreen_cs_set_vertex_buffer( rctx->flags |= R600_CONTEXT_TEX_FLUSH; state->enabled_mask |= 1 << vb_index; state->dirty_mask |= 1 << vb_index; - r600_atom_dirty(rctx, &state->atom); + state->atom.dirty = true; } -const struct u_resource_vtbl r600_global_buffer_vtbl = +static const struct u_resource_vtbl r600_global_buffer_vtbl = { u_default_resource_get_handle, /* get_handle */ r600_compute_global_buffer_destroy, /* resource_destroy */ - r600_compute_global_get_transfer, /* get_transfer */ - r600_compute_global_transfer_destroy, /* transfer_destroy */ r600_compute_global_transfer_map, /* transfer_map */ r600_compute_global_transfer_flush_region,/* transfer_flush_region */ r600_compute_global_transfer_unmap, /* transfer_unmap */ @@ -329,7 +327,7 @@ static void compute_emit_cs(struct r600_context *ctx, const uint *block_layout, * See evergreen_init_atom_start_compute_cs() in this file for the list * of registers initialized by the start_compute_cs_cmd atom. */ - r600_emit_atom(ctx, &ctx->start_compute_cs_cmd.atom); + r600_emit_command_buffer(ctx->cs, &ctx->start_compute_cs_cmd); ctx->flags |= R600_CONTEXT_CB_FLUSH; r600_flush_emit(ctx); @@ -480,10 +478,12 @@ static void evergreen_launch_grid( { struct r600_context *ctx = (struct r600_context *)ctx_; +#ifdef HAVE_OPENCL COMPUTE_DBG("*** evergreen_launch_grid: pc = %u\n", pc); -#ifdef HAVE_OPENCL + struct r600_pipe_compute *shader = ctx->cs_shader_state.shader; if (!shader->kernels[pc].code_bo) { + void *p; struct r600_kernel *kernel = &shader->kernels[pc]; r600_compute_shader_create(ctx_, kernel->llvm_module, &kernel->bc); kernel->code_bo = r600_compute_buffer_alloc_vram(ctx->screen, @@ -623,9 +623,18 @@ void evergreen_init_atom_start_compute_cs(struct r600_context *ctx) /* since all required registers are initialised in the * start_compute_cs_cmd atom, we can EMIT_EARLY here. */ - r600_init_command_buffer(ctx, cb, 1, 256); + r600_init_command_buffer(cb, 256); cb->pkt_flags = RADEON_CP_PACKET3_COMPUTE_MODE; + /* This must be first. */ + r600_store_value(cb, PKT3(PKT3_CONTEXT_CONTROL, 1, 0)); + r600_store_value(cb, 0x80000000); + r600_store_value(cb, 0x80000000); + + /* We're setting config registers here. */ + r600_store_value(cb, PKT3(PKT3_EVENT_WRITE, 0, 0)); + r600_store_value(cb, EVENT_TYPE(EVENT_TYPE_CS_PARTIAL_FLUSH) | EVENT_INDEX(4)); + switch (ctx->family) { case CHIP_CEDAR: default: @@ -839,30 +848,56 @@ void r600_compute_global_buffer_destroy( free(res); } -void* r600_compute_global_transfer_map( +void *r600_compute_global_transfer_map( struct pipe_context *ctx_, - struct pipe_transfer* transfer) + struct pipe_resource *resource, + unsigned level, + unsigned usage, + const struct pipe_box *box, + struct pipe_transfer **ptransfer) { + struct r600_context *rctx = (struct r600_context*)ctx_; + struct compute_memory_pool *pool = rctx->screen->global_pool; + struct pipe_transfer *transfer = util_slab_alloc(&rctx->pool_transfers); + struct r600_resource_global* buffer = + (struct r600_resource_global*)resource; + uint32_t* map; + + compute_memory_finalize_pending(pool, ctx_); + + assert(resource->target == PIPE_BUFFER); + + COMPUTE_DBG("* r600_compute_global_get_transfer()\n" + "level = %u, usage = %u, box(x = %u, y = %u, z = %u " + "width = %u, height = %u, depth = %u)\n", level, usage, + box->x, box->y, box->z, box->width, box->height, + box->depth); + + transfer->resource = resource; + transfer->level = level; + transfer->usage = usage; + transfer->box = *box; + transfer->stride = 0; + transfer->layer_stride = 0; + assert(transfer->resource->target == PIPE_BUFFER); assert(transfer->resource->bind & PIPE_BIND_GLOBAL); assert(transfer->box.x >= 0); assert(transfer->box.y == 0); assert(transfer->box.z == 0); - struct r600_context *ctx = (struct r600_context *)ctx_; - struct r600_resource_global* buffer = - (struct r600_resource_global*)transfer->resource; - - uint32_t* map; ///TODO: do it better, mapping is not possible if the pool is too big COMPUTE_DBG("* r600_compute_global_transfer_map()\n"); - if (!(map = ctx->ws->buffer_map(buffer->chunk->pool->bo->cs_buf, - ctx->cs, transfer->usage))) { + if (!(map = rctx->ws->buffer_map(buffer->chunk->pool->bo->cs_buf, + rctx->cs, transfer->usage))) { + util_slab_free(&rctx->pool_transfers, transfer); return NULL; } + *ptransfer = transfer; + COMPUTE_DBG("Buffer: %p + %u (buffer offset in global memory) " "+ %u (box.x)\n", map, buffer->chunk->start_in_dw, transfer->box.x); return ((char*)(map + buffer->chunk->start_in_dw)) + transfer->box.x; @@ -882,50 +917,7 @@ void r600_compute_global_transfer_unmap( COMPUTE_DBG("* r600_compute_global_transfer_unmap()\n"); ctx->ws->buffer_unmap(buffer->chunk->pool->bo->cs_buf); -} - -struct pipe_transfer * r600_compute_global_get_transfer( - struct pipe_context *ctx_, - struct pipe_resource *resource, - unsigned level, - unsigned usage, - const struct pipe_box *box) -{ - struct r600_context *ctx = (struct r600_context *)ctx_; - struct compute_memory_pool *pool = ctx->screen->global_pool; - - compute_memory_finalize_pending(pool, ctx_); - - assert(resource->target == PIPE_BUFFER); - struct r600_context *rctx = (struct r600_context*)ctx_; - struct pipe_transfer *transfer = util_slab_alloc(&rctx->pool_transfers); - - COMPUTE_DBG("* r600_compute_global_get_transfer()\n" - "level = %u, usage = %u, box(x = %u, y = %u, z = %u " - "width = %u, height = %u, depth = %u)\n", level, usage, - box->x, box->y, box->z, box->width, box->height, - box->depth); - - transfer->resource = resource; - transfer->level = level; - transfer->usage = usage; - transfer->box = *box; - transfer->stride = 0; - transfer->layer_stride = 0; - transfer->data = NULL; - - /* Note strides are zero, this is ok for buffers, but not for - * textures 2d & higher at least. - */ - return transfer; -} - -void r600_compute_global_transfer_destroy( - struct pipe_context *ctx_, - struct pipe_transfer *transfer) -{ - struct r600_context *rctx = (struct r600_context*)ctx_; - util_slab_free(&rctx->pool_transfers, transfer); + util_slab_free(&ctx->pool_transfers, transfer); } void r600_compute_global_transfer_flush_region(