28 Apr
2022
28 Apr
'22
3:19 p.m.
On Thu, Apr 28, 2022 at 05:18:13PM +0200, Thierry Reding wrote:
On Tue, Apr 26, 2022 at 02:07:59PM +0800, Cai Huoqing wrote: [...]
diff --git a/drivers/gpu/drm/nvdla/nvdla_drv.c b/drivers/gpu/drm/nvdla/nvdla_drv.c
I'll look at this from an architectural level and leave it to other experts to review the more technical things.
[...]
+static struct nvdla_config nvdla_config_os_initial = {
- .atom_size = 32,
- .bdma_enable = true,
- .rubik_enable = true,
- .weight_compress_support = true,
+};
+static struct nvdla_config nvdla_config_small = {
- //.atom_size = 8,
- .atom_size = 32, // nv_large config
- .bdma_enable = false,
- .rubik_enable = false,
- .weight_compress_support = false,
+};
[...]
+static union nvdla_operation_container operation_desc[NVDLA_OP_NUM][NVDLA_NUM_GROUPS]; +static union nvdla_surface_container surface_desc[NVDLA_OP_NUM][NVDLA_NUM_GROUPS];
+static struct nvdla_task_desc global_task;
+static struct nvdla_engine engine = {
- .processors[NVDLA_OP_BDMA] = {
.name = "BDMA",.op_type = NVDLA_OP_BDMA,.program = nvdla_bdma_program,.enable = nvdla_bdma_enable,.set_producer = nvdla_bdma_set_producer,.is_ready = nvdla_bdma_is_ready,.dump_config = nvdla_bdma_dump_config,.rdma_check = nvdla_bdma_rdma_check,.consumer_ptr = 0,.roi_index = 0,.group_status = 0,.rdma_status = 0,.last_group = 1,.groups[0] = {.id = 0,.rdma_id = 0,.active = 0,.events = 0,.roi_index = 0,.is_rdma_needed = 0,.lut_index = -1,.operation_desc = &operation_desc[NVDLA_OP_BDMA][0],.surface_desc = &surface_desc[NVDLA_OP_BDMA][0],},.groups[1] = {.id = 1,.rdma_id = 0,.active = 0,.events = 0,.roi_index = 0,.is_rdma_needed = 0,.lut_index = -1,.operation_desc = &operation_desc[NVDLA_OP_BDMA][1],.surface_desc = &surface_desc[NVDLA_OP_BDMA][1],},- },
- .processors[NVDLA_OP_CONV] = {
.name = "Convolution",.op_type = NVDLA_OP_CONV,.program = nvdla_conv_program,.enable = nvdla_conv_enable,.set_producer = nvdla_conv_set_producer,.is_ready = nvdla_conv_is_ready,.dump_config = nvdla_conv_dump_config,.rdma_check = nvdla_conv_rdma_check,.consumer_ptr = 0,.roi_index = 0,.group_status = 0,.rdma_status = 0,.last_group = 1,.groups[0] = {.id = 0,.rdma_id = 0,.active = 0,.events = 0,.roi_index = 0,.is_rdma_needed = 0,.lut_index = -1,.operation_desc = &operation_desc[NVDLA_OP_CONV][0],.surface_desc = &surface_desc[NVDLA_OP_CONV][0],},.groups[1] = {.id = 1,.rdma_id = 0,.active = 0,.events = 0,.roi_index = 0,.is_rdma_needed = 0,.lut_index = -1,.operation_desc = &operation_desc[NVDLA_OP_CONV][1],.surface_desc = &surface_desc[NVDLA_OP_CONV][1],},- },
- .processors[NVDLA_OP_SDP] = {
.name = "SDP",.op_type = NVDLA_OP_SDP,.program = nvdla_sdp_program,.enable = nvdla_sdp_enable,.set_producer = nvdla_sdp_set_producer,.is_ready = nvdla_sdp_is_ready,.dump_config = nvdla_sdp_dump_config,.rdma_check = nvdla_sdp_rdma_check,.consumer_ptr = 0,.roi_index = 0,.group_status = 0,.rdma_status = 0,.last_group = 1,.groups[0] = {.id = 0,.rdma_id = 0,.active = 0,.events = 0,.roi_index = 0,.is_rdma_needed = 0,.lut_index = -1,.operation_desc = &operation_desc[NVDLA_OP_SDP][0],.surface_desc = &surface_desc[NVDLA_OP_SDP][0],},.groups[1] = {.id = 1,.rdma_id = 0,.active = 0,.events = 0,.roi_index = 0,.is_rdma_needed = 0,.lut_index = -1,.operation_desc = &operation_desc[NVDLA_OP_SDP][1],.surface_desc = &surface_desc[NVDLA_OP_SDP][1],},- },
- .processors[NVDLA_OP_PDP] = {
.name = "PDP",.op_type = NVDLA_OP_PDP,.program = nvdla_pdp_program,.enable = nvdla_pdp_enable,.set_producer = nvdla_pdp_set_producer,.is_ready = nvdla_pdp_is_ready,.dump_config = nvdla_pdp_dump_config,.rdma_check = nvdla_pdp_rdma_check,.consumer_ptr = 0,.roi_index = 0,.group_status = 0,.rdma_status = 0,.last_group = 1,.groups[0] = {.id = 0,.rdma_id = 0,.active = 0,.events = 0,.roi_index = 0,.is_rdma_needed = 0,.lut_index = -1,.operation_desc = &operation_desc[NVDLA_OP_PDP][0],.surface_desc = &surface_desc[NVDLA_OP_PDP][0],},.groups[1] = {.id = 1,.rdma_id = 0,.active = 0,.events = 0,.roi_index = 0,.is_rdma_needed = 0,.lut_index = -1,.operation_desc = &operation_desc[NVDLA_OP_PDP][1],.surface_desc = &surface_desc[NVDLA_OP_PDP][1],},- },
- .processors[NVDLA_OP_CDP] = {
.name = "CDP",.op_type = NVDLA_OP_CDP,.program = nvdla_cdp_program,.enable = nvdla_cdp_enable,.set_producer = nvdla_cdp_set_producer,.is_ready = nvdla_cdp_is_ready,.dump_config = nvdla_cdp_dump_config,.rdma_check = nvdla_cdp_rdma_check,.consumer_ptr = 0,.roi_index = 0,.group_status = 0,.rdma_status = 0,.last_group = 1,.groups[0] = {.id = 0,.rdma_id = 0,.active = 0,.events = 0,.roi_index = 0,.is_rdma_needed = 0,.lut_index = -1,.operation_desc = &operation_desc[NVDLA_OP_CDP][0],.surface_desc = &surface_desc[NVDLA_OP_CDP][0],},.groups[1] = {.id = 1,.rdma_id = 0,.active = 0,.events = 0,.roi_index = 0,.is_rdma_needed = 0,.lut_index = -1,.operation_desc = &operation_desc[NVDLA_OP_CDP][1],.surface_desc = &surface_desc[NVDLA_OP_CDP][1],},- },
- .processors[NVDLA_OP_RUBIK] = {
.name = "RUBIK",.op_type = NVDLA_OP_RUBIK,.program = nvdla_rubik_program,.enable = nvdla_rubik_enable,.set_producer = nvdla_rubik_set_producer,.is_ready = nvdla_rubik_is_ready,.dump_config = nvdla_rubik_dump_config,.rdma_check = nvdla_rubik_rdma_check,.consumer_ptr = 0,.roi_index = 0,.group_status = 0,.rdma_status = 0,.last_group = 1,.groups[0] = {.id = 0,.rdma_id = 0,.active = 0,.events = 0,.roi_index = 0,.is_rdma_needed = 0,.lut_index = -1,.operation_desc = &operation_desc[NVDLA_OP_RUBIK][0],.surface_desc = &surface_desc[NVDLA_OP_RUBIK][0],},.groups[1] = {.id = 1,.rdma_id = 0,.active = 0,.events = 0,.roi_index = 0,.is_rdma_needed = 0,.lut_index = -1,.operation_desc = &operation_desc[NVDLA_OP_RUBIK][1],.surface_desc = &surface_desc[NVDLA_OP_RUBIK][1],},- },
+};
These global variables aren't going to work because Tegra234 (Tegra194's successor) has two instances of NVDLA.
Small correction: I just recalled that even Tegra194 has two DLA instances, so if we want both supported we'll need to get rid of those global variables for that chip already.
Thierry