Path: blob/21.2-virgl/src/gallium/drivers/nouveau/nv50/nv98_video_bsp.c
4574 views
/*1* Copyright 2011-2013 Maarten Lankhorst, Ilia Mirkin2*3* Permission is hereby granted, free of charge, to any person obtaining a4* copy of this software and associated documentation files (the "Software"),5* to deal in the Software without restriction, including without limitation6* the rights to use, copy, modify, merge, publish, distribute, sublicense,7* and/or sell copies of the Software, and to permit persons to whom the8* Software is furnished to do so, subject to the following conditions:9*10* The above copyright notice and this permission notice shall be included in11* all copies or substantial portions of the Software.12*13* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR14* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,15* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL16* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR17* OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,18* ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR19* OTHER DEALINGS IN THE SOFTWARE.20*/2122#include "nv50/nv98_video.h"2324#if NOUVEAU_VP3_DEBUG_FENCE25static void dump_comm_bsp(struct comm *comm)26{27unsigned idx = comm->bsp_cur_index & 0xf;28debug_printf("Cur seq: %x, bsp byte ofs: %x\n", comm->bsp_cur_index, comm->byte_ofs);29debug_printf("Status: %08x, pos: %08x\n", comm->status[idx], comm->pos[idx]);30}31#endif3233unsigned34nv98_decoder_bsp(struct nouveau_vp3_decoder *dec, union pipe_desc desc,35struct nouveau_vp3_video_buffer *target,36unsigned comm_seq, unsigned num_buffers,37const void *const *data, const unsigned *num_bytes,38unsigned *vp_caps, unsigned *is_ref,39struct nouveau_vp3_video_buffer *refs[16])40{41struct nouveau_pushbuf *push = dec->pushbuf[0];42enum pipe_video_format codec = u_reduce_video_profile(dec->base.profile);43uint32_t bsp_addr, comm_addr, inter_addr;44uint32_t slice_size, bucket_size, ring_size, bsp_size;45uint32_t caps, i;46int ret;47struct nouveau_bo *bsp_bo = dec->bsp_bo[comm_seq % NOUVEAU_VP3_VIDEO_QDEPTH];48struct nouveau_bo *inter_bo = dec->inter_bo[comm_seq & 1];49struct nouveau_pushbuf_refn bo_refs[] = {50{ bsp_bo, NOUVEAU_BO_RD | NOUVEAU_BO_VRAM },51{ inter_bo, NOUVEAU_BO_WR | NOUVEAU_BO_VRAM },52#if NOUVEAU_VP3_DEBUG_FENCE53{ dec->fence_bo, NOUVEAU_BO_WR | NOUVEAU_BO_GART },54#endif55{ dec->bitplane_bo, NOUVEAU_BO_RDWR | NOUVEAU_BO_VRAM },56};57int num_refs = ARRAY_SIZE(bo_refs);5859if (!dec->bitplane_bo)60num_refs--;6162bsp_size = NOUVEAU_VP3_BSP_RESERVED_SIZE;63for (i = 0; i < num_buffers; i++)64bsp_size += num_bytes[i];65bsp_size += 256; /* the 4 end markers */6667if (!bsp_bo || bsp_size > bsp_bo->size) {68struct nouveau_bo *tmp_bo = NULL;6970/* round up to the nearest mb */71bsp_size += (1 << 20) - 1;72bsp_size &= ~((1 << 20) - 1);7374ret = nouveau_bo_new(dec->client->device, NOUVEAU_BO_VRAM, 0, bsp_size, NULL, &tmp_bo);75if (ret) {76debug_printf("reallocating bsp %u -> %u failed with %i\n",77bsp_bo ? (unsigned)bsp_bo->size : 0, bsp_size, ret);78return -1;79}80nouveau_bo_ref(NULL, &bsp_bo);81bo_refs[0].bo = dec->bsp_bo[comm_seq % NOUVEAU_VP3_VIDEO_QDEPTH] = bsp_bo = tmp_bo;82}8384if (!inter_bo || bsp_bo->size * 4 > inter_bo->size) {85struct nouveau_bo *tmp_bo = NULL;8687ret = nouveau_bo_new(dec->client->device, NOUVEAU_BO_VRAM, 0, bsp_bo->size * 4, NULL, &tmp_bo);88if (ret) {89debug_printf("reallocating inter %u -> %u failed with %i\n",90inter_bo ? (unsigned)inter_bo->size : 0, (unsigned)bsp_bo->size * 4, ret);91return -1;92}93nouveau_bo_ref(NULL, &inter_bo);94bo_refs[1].bo = dec->inter_bo[comm_seq & 1] = inter_bo = tmp_bo;95}9697ret = nouveau_bo_map(bsp_bo, NOUVEAU_BO_WR, dec->client);98if (ret) {99debug_printf("map failed: %i %s\n", ret, strerror(-ret));100return -1;101}102103nouveau_vp3_bsp_begin(dec);104nouveau_vp3_bsp_next(dec, num_buffers, data, num_bytes);105caps = nouveau_vp3_bsp_end(dec, desc);106107nouveau_vp3_vp_caps(dec, desc, target, comm_seq, vp_caps, is_ref, refs);108109nouveau_pushbuf_space(push, 32, num_refs, 0);110nouveau_pushbuf_refn(push, bo_refs, num_refs);111112bsp_addr = bsp_bo->offset >> 8;113inter_addr = inter_bo->offset >> 8;114115#if NOUVEAU_VP3_DEBUG_FENCE116memset(dec->comm, 0, 0x200);117comm_addr = (dec->fence_bo->offset + COMM_OFFSET) >> 8;118#else119comm_addr = bsp_addr + (COMM_OFFSET>>8);120#endif121122BEGIN_NV04(push, SUBC_BSP(0x700), 5);123PUSH_DATA (push, caps); // 700 cmd124PUSH_DATA (push, bsp_addr + 1); // 704 strparm_bsp125PUSH_DATA (push, bsp_addr + 7); // 708 str addr126PUSH_DATA (push, comm_addr); // 70c comm127PUSH_DATA (push, comm_seq); // 710 seq128129if (codec != PIPE_VIDEO_FORMAT_MPEG4_AVC) {130u32 bitplane_addr;131int mpeg12 = (codec == PIPE_VIDEO_FORMAT_MPEG12);132133bitplane_addr = dec->bitplane_bo->offset >> 8;134135nouveau_vp3_inter_sizes(dec, 1, &slice_size, &bucket_size, &ring_size);136BEGIN_NV04(push, SUBC_BSP(0x400), mpeg12 ? 5 : 7);137PUSH_DATA (push, bsp_addr); // 400 picparm addr138PUSH_DATA (push, inter_addr); // 404 interparm addr139PUSH_DATA (push, inter_addr + slice_size + bucket_size); // 408 interdata addr140PUSH_DATA (push, ring_size << 8); // 40c interdata_size141if (!mpeg12) {142PUSH_DATA (push, bitplane_addr); // 410 BITPLANE_DATA143PUSH_DATA (push, 0x400); // 414 BITPLANE_DATA_SIZE144}145PUSH_DATA (push, 0); // dma idx146} else {147nouveau_vp3_inter_sizes(dec, desc.h264->slice_count, &slice_size, &bucket_size, &ring_size);148BEGIN_NV04(push, SUBC_BSP(0x400), 8);149PUSH_DATA (push, bsp_addr); // 400 picparm addr150PUSH_DATA (push, inter_addr); // 404 interparm addr151PUSH_DATA (push, slice_size << 8); // 408 interparm size?152PUSH_DATA (push, inter_addr + slice_size + bucket_size); // 40c interdata addr153PUSH_DATA (push, ring_size << 8); // 410 interdata size154PUSH_DATA (push, inter_addr + slice_size); // 414 bucket?155PUSH_DATA (push, bucket_size << 8); // 418 bucket size? unshifted..156PUSH_DATA (push, 0); // 41c targets157// TODO: Double check 414 / 418 with nvidia trace158}159160#if NOUVEAU_VP3_DEBUG_FENCE161BEGIN_NV04(push, SUBC_BSP(0x240), 3);162PUSH_DATAh(push, dec->fence_bo->offset);163PUSH_DATA (push, dec->fence_bo->offset);164PUSH_DATA (push, dec->fence_seq);165166BEGIN_NV04(push, SUBC_BSP(0x300), 1);167PUSH_DATA (push, 1);168PUSH_KICK (push);169170{171unsigned spin = 0;172do {173usleep(100);174if ((spin++ & 0xff) == 0xff) {175debug_printf("b%u: %u\n", dec->fence_seq, dec->fence_map[0]);176dump_comm_bsp(dec->comm);177}178} while (dec->fence_seq > dec->fence_map[0]);179}180181dump_comm_bsp(dec->comm);182return dec->comm->status[comm_seq & 0xf];183#else184BEGIN_NV04(push, SUBC_BSP(0x300), 1);185PUSH_DATA (push, 0);186PUSH_KICK (push);187return 2;188#endif189}190191192