/* Initialize kernel GPU context and state map */
static int initialize_gpu_context(gcoCONTEXT vctx)
{
/* First build context state map from compressed representation */
size_t contextbuf_addr_size = sizeof(contextbuf_addr)/sizeof(address_index_t);
size_t state_count = contextbuf_addr[contextbuf_addr_size - 1].address / 4 + 1;
uint32_t *context_map = malloc(state_count * 4);
if(context_map == NULL)
{
return ETNA_OUT_OF_MEMORY;
}
memset(context_map, 0, state_count*4);
for(int idx=0; idx<contextbuf_addr_size; ++idx)
{
context_map[contextbuf_addr[idx].address / 4] = contextbuf_addr[idx].index;
}
#ifdef DEBUG
printf("Initialized state map (%x)\n", state_count);
#endif
/* fill in context */
vctx->object.type = gcvOBJ_CONTEXT;
vctx->id = 0x0; // Actual ID will be returned here by kernel
vctx->map = context_map;
vctx->stateCount = state_count;
vctx->buffer = malloc(sizeof(contextbuf));
memcpy(vctx->buffer, contextbuf, sizeof(contextbuf)); /* copy over hardcoded context command buffer */
vctx->pipe3DIndex = 0x2d6; // XXX should not be hardcoded
vctx->pipe2DIndex = 0x106e; // XXX should not be hardcoded
vctx->linkIndex = 0x1076; // XXX should not be hardcoded
vctx->inUseIndex = 0x1078; // XXX should not be hardcoded
vctx->bufferSize = sizeof(contextbuf);
vctx->bytes = 0x0; // Number of bytes at actually allocated for physical, logical
vctx->physical = (void*)0x0;
vctx->logical = (void*)0x0;
vctx->link = (void*)0x0; // Logical address of link (within consecutive array)
vctx->initialPipe = ETNA_PIPE_2D;
vctx->entryPipe = ETNA_PIPE_3D;
vctx->currentPipe = ETNA_PIPE_3D; // not used by kernel
vctx->postCommit = 1; // not used by kernel
vctx->inUse = (int*)0x0; // Logical address of inUse (within consecutive array)
viv_addr_t cbuf0_physical = 0;
void *cbuf0_logical = 0;
size_t cbuf0_bytes = 0;
if(viv_alloc_contiguous(vctx->bufferSize, &cbuf0_physical, &cbuf0_logical, &cbuf0_bytes)!=0)
{
#ifdef DEBUG
fprintf(stderr, "Error allocating contiguous host memory for context\n");
#endif
free(context_map);
return ETNA_OUT_OF_MEMORY;
}
#ifdef DEBUG
printf("Allocated buffer (size 0x%x) for context: phys=%08x log=%08x\n", (int)cbuf0_bytes, (int)cbuf0_physical, (int)cbuf0_logical);
#endif
vctx->bytes = cbuf0_bytes; /* actual size of buffer */
vctx->physical = (void*)cbuf0_physical;
vctx->logical = cbuf0_logical;
vctx->link = ((uint32_t*)cbuf0_logical) + vctx->linkIndex;
vctx->inUse = (gctBOOL*)(((uint32_t*)cbuf0_logical) + vctx->inUseIndex);
/* copy over context buffer to contiguous memory, clear in-use flag */
memcpy(vctx->logical, vctx->buffer, vctx->bufferSize);
*vctx->inUse = 0;
return ETNA_OK;
}
int main(int argc, char **argv)
{
int rv;
fb_info fb;
rv = fb_open(0, &fb);
if(rv!=0)
{
exit(1);
}
fb_set_buffer(&fb, 0);
rv = viv_open();
if(rv!=0)
{
fprintf(stderr, "Error opening device\n");
exit(1);
}
printf("Succesfully opened device\n");
/* allocate command buffer (blob uses four command buffers, but we don't even fill one) */
viv_addr_t buf0_physical = 0;
void *buf0_logical = 0;
if(viv_alloc_contiguous(0x8000, &buf0_physical, &buf0_logical, NULL)!=0)
{
fprintf(stderr, "Error allocating host memory\n");
exit(1);
}
printf("Allocated buffer: phys=%08x log=%08x\n", (uint32_t)buf0_physical, (uint32_t)buf0_logical);
/* allocate main render target */
gcuVIDMEM_NODE_PTR rt_node = 0;
if(viv_alloc_linear_vidmem(0x70000, 0x40, gcvSURF_RENDER_TARGET, gcvPOOL_DEFAULT, &rt_node, NULL)!=0)
{
fprintf(stderr, "Error allocating render target buffer memory\n");
exit(1);
}
printf("Allocated render target node: node=%08x\n", (uint32_t)rt_node);
viv_addr_t rt_physical = 0;
void *rt_logical = 0;
if(viv_lock_vidmem(rt_node, &rt_physical, &rt_logical)!=0)
{
fprintf(stderr, "Error locking render target memory\n");
exit(1);
}
printf("Locked render target: phys=%08x log=%08x\n", (uint32_t)rt_physical, (uint32_t)rt_logical);
memset(rt_logical, 0xff, 0x70000); /* clear previous result just in case, test that clearing works */
/* allocate tile status for main render target */
gcuVIDMEM_NODE_PTR rt_ts_node = 0;
if(viv_alloc_linear_vidmem(0x700, 0x40, gcvSURF_TILE_STATUS, gcvPOOL_DEFAULT, &rt_ts_node, NULL)!=0)
{
fprintf(stderr, "Error allocating render target tile status memory\n");
exit(1);
}
printf("Allocated render target tile status node: node=%08x\n", (uint32_t)rt_ts_node);
viv_addr_t rt_ts_physical = 0;
void *rt_ts_logical = 0;
if(viv_lock_vidmem(rt_ts_node, &rt_ts_physical, &rt_ts_logical)!=0)
{
fprintf(stderr, "Error locking render target memory\n");
exit(1);
}
printf("Locked render target ts: phys=%08x log=%08x\n", (uint32_t)rt_ts_physical, (uint32_t)rt_ts_logical);
/* allocate depth for main render target */
gcuVIDMEM_NODE_PTR z_node = 0;
if(viv_alloc_linear_vidmem(0x38000, 0x40, gcvSURF_DEPTH, gcvPOOL_DEFAULT, &z_node, NULL)!=0)
{
fprintf(stderr, "Error allocating depth memory\n");
exit(1);
}
printf("Allocated depth node: node=%08x\n", (uint32_t)z_node);
viv_addr_t z_physical = 0;
void *z_logical = 0;
if(viv_lock_vidmem(z_node, &z_physical, &z_logical)!=0)
{
fprintf(stderr, "Error locking depth target memory\n");
exit(1);
}
printf("Locked depth target: phys=%08x log=%08x\n", (uint32_t)z_physical, (uint32_t)z_logical);
/* allocate depth ts for main render target */
gcuVIDMEM_NODE_PTR z_ts_node = 0;
if(viv_alloc_linear_vidmem(0x400, 0x40, gcvSURF_TILE_STATUS, gcvPOOL_DEFAULT, &z_ts_node, NULL)!=0)
{
fprintf(stderr, "Error allocating depth memory\n");
exit(1);
}
printf("Allocated depth ts node: node=%08x\n", (uint32_t)z_ts_node);
viv_addr_t z_ts_physical = 0;
void *z_ts_logical = 0;
if(viv_lock_vidmem(z_ts_node, &z_ts_physical, &z_ts_logical)!=0)
{
fprintf(stderr, "Error locking depth target ts memory\n");
exit(1);
}
printf("Locked depth ts target: phys=%08x log=%08x\n", (uint32_t)z_ts_physical, (uint32_t)z_ts_logical);
/* allocate vertex buffer */
gcuVIDMEM_NODE_PTR vtx_node = 0;
if(viv_alloc_linear_vidmem(0x60000, 0x40, gcvSURF_VERTEX, gcvPOOL_DEFAULT, &vtx_node, NULL)!=0)
{
fprintf(stderr, "Error allocating vertex memory\n");
exit(1);
}
printf("Allocated vertex node: node=%08x\n", (uint32_t)vtx_node);
viv_addr_t vtx_physical = 0;
void *vtx_logical = 0;
if(viv_lock_vidmem(vtx_node, &vtx_physical, &vtx_logical)!=0)
{
fprintf(stderr, "Error locking vertex memory\n");
exit(1);
}
printf("Locked vertex memory: phys=%08x log=%08x\n", (uint32_t)vtx_physical, (uint32_t)vtx_logical);
/* allocate aux render target */
gcuVIDMEM_NODE_PTR aux_rt_node = 0;
if(viv_alloc_linear_vidmem(0x4000, 0x40, gcvSURF_RENDER_TARGET, gcvPOOL_SYSTEM /*why?*/, &aux_rt_node, NULL)!=0)
{
fprintf(stderr, "Error allocating aux render target buffer memory\n");
exit(1);
}
printf("Allocated aux render target node: node=%08x\n", (uint32_t)aux_rt_node);
viv_addr_t aux_rt_physical = 0;
void *aux_rt_logical = 0;
if(viv_lock_vidmem(aux_rt_node, &aux_rt_physical, &aux_rt_logical)!=0)
{
fprintf(stderr, "Error locking aux render target memory\n");
exit(1);
}
printf("Locked aux render target: phys=%08x log=%08x\n", (uint32_t)aux_rt_physical, (uint32_t)aux_rt_logical);
/* allocate tile status for aux render target */
gcuVIDMEM_NODE_PTR aux_rt_ts_node = 0;
if(viv_alloc_linear_vidmem(0x100, 0x40, gcvSURF_TILE_STATUS, gcvPOOL_DEFAULT, &aux_rt_ts_node, NULL)!=0)
{
fprintf(stderr, "Error allocating aux render target tile status memory\n");
exit(1);
}
printf("Allocated aux render target tile status node: node=%08x\n", (uint32_t)aux_rt_ts_node);
viv_addr_t aux_rt_ts_physical = 0;
void *aux_rt_ts_logical = 0;
if(viv_lock_vidmem(aux_rt_ts_node, &aux_rt_ts_physical, &aux_rt_ts_logical)!=0)
{
fprintf(stderr, "Error locking aux ts render target memory\n");
exit(1);
}
printf("Locked aux render target ts: phys=%08x log=%08x\n", (uint32_t)aux_rt_ts_physical, (uint32_t)aux_rt_ts_logical);
/* Phew, now we got all the memory we need.
* Write interleaved attribute vertex stream.
* Unlike the GL example we only do this once, not every time glDrawArrays is called, the same would be accomplished
* from GL by using a vertex buffer object.
*/
for(int vert=0; vert<NUM_VERTICES; ++vert)
{
int src_idx = vert * COMPONENTS_PER_VERTEX;
int dest_idx = vert * COMPONENTS_PER_VERTEX * 3;
for(int comp=0; comp<COMPONENTS_PER_VERTEX; ++comp)
{
((float*)vtx_logical)[dest_idx+comp+0] = vVertices[src_idx + comp]; /* 0 */
((float*)vtx_logical)[dest_idx+comp+3] = vNormals[src_idx + comp]; /* 1 */
((float*)vtx_logical)[dest_idx+comp+6] = vColors[src_idx + comp]; /* 2 */
}
}
/*
for(int idx=0; idx<NUM_VERTICES*3*3; ++idx)
{
printf("%i %f\n", idx, ((float*)vtx_logical)[idx]);
}*/
/* Load the command buffer and send the commit command. */
/* First build context state map */
size_t stateCount = 0x1d00;
uint32_t *contextMap = malloc(stateCount * 4);
memset(contextMap, 0, stateCount*4);
for(int idx=0; idx<sizeof(contextbuf_addr)/sizeof(address_index_t); ++idx)
{
contextMap[contextbuf_addr[idx].address / 4] = contextbuf_addr[idx].index;
}
struct _gcoCMDBUF commandBuffer = {
.object = {
.type = gcvOBJ_COMMANDBUFFER
},
//.os = (_gcoOS*)0xbf7488,
//.hardware = (_gcoHARDWARE*)0x402694e0,
.physical = (void*)buf0_physical,
.logical = (void*)buf0_logical,
.bytes = 0x8000,
.startOffset = 0x0,
//.offset = 0xac0,
//.free = 0x7520,
//.hintTable = (unsigned int*)0x0, // Used when gcdSECURE
//.hintIndex = (unsigned int*)0x58, // Used when gcdSECURE
//.hintCommit = (unsigned int*)0xffffffff // Used when gcdSECURE
};
struct _gcoCONTEXT contextBuffer = {
.object = {
.type = gcvOBJ_CONTEXT
},
//.os = (_gcoOS*)0xbf7488,
//.hardware = (_gcoHARDWARE*)0x402694e0,
.id = 0x0, // Actual ID will be returned here
.map = contextMap,
.stateCount = stateCount,
//.hint = (unsigned char*)0x0, // Used when gcdSECURE
//.hintValue = 2, // Used when gcdSECURE
//.hintCount = 0xca, // Used when gcdSECURE
.buffer = contextbuf,
.pipe3DIndex = 0x2d6, // XXX should not be hardcoded
.pipe2DIndex = 0x106e,
.linkIndex = 0x1076,
.inUseIndex = 0x1078,
.bufferSize = 0x41e4,
.bytes = 0x0, // Number of bytes at physical, logical
.physical = (void*)0x0,
.logical = (void*)0x0,
.link = (void*)0x0, // Logical address of link
.initialPipe = 0x1,
.entryPipe = 0x0,
.currentPipe = 0x0,
.postCommit = 1,
.inUse = (int*)0x0, // Logical address of inUse
.lastAddress = 0xffffffff, // Not used by kernel
.lastSize = 0x2, // Not used by kernel
.lastIndex = 0x106a, // Not used by kernel
.lastFixed = 0, // Not used by kernel
//.hintArray = (unsigned int*)0x0, // Used when gcdSECURE
//.hintIndex = (unsigned int*)0x0 // Used when gcdSECURE
};
commandBuffer.free = commandBuffer.bytes - 0x8; /* Always keep 0x8 at end of buffer for kernel driver */
/* Set addresses in first command buffer */
cmdbuf1[0x57] = cmdbuf1[0x67] = cmdbuf1[0x9f] = cmdbuf1[0xbb] = cmdbuf1[0xd9] = cmdbuf1[0xfb] = rt_physical;
cmdbuf1[0x65] = cmdbuf1[0x9d] = cmdbuf1[0xb9] = cmdbuf1[0xd7] = cmdbuf1[0xe5] = cmdbuf1[0xf9] = rt_ts_physical;
cmdbuf1[0x6d] = cmdbuf1[0x7f] = z_physical;
cmdbuf1[0x7d] = z_ts_physical;
cmdbuf1[0x87] = cmdbuf1[0xa3] = cmdbuf1[0xc1] = aux_rt_ts_physical;
cmdbuf1[0x89] = cmdbuf1[0x8f] = cmdbuf1[0x93]
= cmdbuf1[0xa5] = cmdbuf1[0xab] = cmdbuf1[0xaf]
= cmdbuf1[0xc3] = cmdbuf1[0xc9] = cmdbuf1[0xcd] = aux_rt_physical;
cmdbuf1[0x1f3] = cmdbuf1[0x215] = cmdbuf1[0x237]
= cmdbuf1[0x259] = cmdbuf1[0x27b] = cmdbuf1[0x29d] = vtx_physical;
/* Submit first command buffer */
commandBuffer.startOffset = 0;
memcpy((void*)((size_t)commandBuffer.logical + commandBuffer.startOffset), cmdbuf1, sizeof(cmdbuf1));
commandBuffer.offset = commandBuffer.startOffset + sizeof(cmdbuf1);
commandBuffer.free -= sizeof(cmdbuf1) + 0x18;
printf("[1] startOffset=%08x, offset=%08x, free=%08x\n", (uint32_t)commandBuffer.startOffset, (uint32_t)commandBuffer.offset, (uint32_t)commandBuffer.free);
if(viv_commit(&commandBuffer, &contextBuffer) != 0)
{
fprintf(stderr, "Error committing first command buffer\n");
exit(1);
}
/* After the first COMMIT, allocate contiguous memory for context and set
* bytes, physical, logical, link, inUse */
printf("Context assigned index: %i\n", (uint32_t)contextBuffer.id);
viv_addr_t cbuf0_physical = 0;
void *cbuf0_logical = 0;
size_t cbuf0_bytes = 0;
if(viv_alloc_contiguous(contextBuffer.bufferSize, &cbuf0_physical, &cbuf0_logical, &cbuf0_bytes)!=0)
{
fprintf(stderr, "Error allocating contiguous host memory for context\n");
exit(1);
}
printf("Allocated buffer (size 0x%x) for context: phys=%08x log=%08x\n", (int)cbuf0_bytes, (int)cbuf0_physical, (int)cbuf0_logical);
contextBuffer.bytes = cbuf0_bytes; /* actual size of buffer */
contextBuffer.physical = (void*)cbuf0_physical;
contextBuffer.logical = cbuf0_logical;
contextBuffer.link = ((uint32_t*)cbuf0_logical) + contextBuffer.linkIndex;
contextBuffer.inUse = (gctBOOL*)(((uint32_t*)cbuf0_logical) + contextBuffer.inUseIndex);
*contextBuffer.inUse = 0;
/* Submit second command buffer, with updated context.
* Second command buffer fills the background.
*/
cmdbuf2[0x1d] = cmdbuf2[0x1f] = rt_physical;
commandBuffer.startOffset = commandBuffer.offset + 0x18; /* Make space for LINK */
memcpy((void*)((size_t)commandBuffer.logical + commandBuffer.startOffset), cmdbuf2, sizeof(cmdbuf2));
commandBuffer.offset = commandBuffer.startOffset + sizeof(cmdbuf2);
commandBuffer.free -= sizeof(cmdbuf2) + 0x18;
printf("[2] startOffset=%08x, offset=%08x, free=%08x\n", (uint32_t)commandBuffer.startOffset, (uint32_t)commandBuffer.offset, (uint32_t)commandBuffer.free);
if(viv_commit(&commandBuffer, &contextBuffer) != 0)
{
fprintf(stderr, "Error committing second command buffer\n");
exit(1);
}
/* Submit third command buffer, with updated context
* Third command buffer does some cache flush trick?
* It can be left out without any visible harm.
**/
cmdbuf3[0x9] = aux_rt_ts_physical;
cmdbuf3[0xb] = cmdbuf3[0x11] = cmdbuf3[0x15] = aux_rt_physical;
cmdbuf3[0x1f] = rt_ts_physical;
cmdbuf3[0x21] = rt_physical;
commandBuffer.startOffset = commandBuffer.offset + 0x18;
memcpy((void*)((size_t)commandBuffer.logical + commandBuffer.startOffset), cmdbuf3, sizeof(cmdbuf3));
commandBuffer.offset = commandBuffer.startOffset + sizeof(cmdbuf3);
commandBuffer.free -= sizeof(cmdbuf3) + 0x18;
printf("[3] startOffset=%08x, offset=%08x, free=%08x\n", (uint32_t)commandBuffer.startOffset, (uint32_t)commandBuffer.offset, (uint32_t)commandBuffer.free);
if(viv_commit(&commandBuffer, &contextBuffer) != 0)
{
fprintf(stderr, "Error committing third command buffer\n");
exit(1);
}
/* Submit event queue with SIGNAL, fromWhere=gcvKERNEL_PIXEL (wait for pixel engine to finish) */
int sig_id = 0;
if(viv_user_signal_create(0, &sig_id) != 0) /* automatic resetting signal */
{
fprintf(stderr, "Cannot create user signal\n");
exit(1);
}
printf("Created user signal %i\n", sig_id);
if(viv_event_queue_signal(sig_id, gcvKERNEL_PIXEL) != 0)
{
fprintf(stderr, "Cannot queue GPU signal\n");
exit(1);
}
/* Wait for signal */
if(viv_user_signal_wait(sig_id, SIG_WAIT_INDEFINITE) != 0)
{
fprintf(stderr, "Cannot wait for signal\n");
exit(1);
}
/* Allocate video memory for BITMAP, lock */
gcuVIDMEM_NODE_PTR bmp_node = 0;
if(viv_alloc_linear_vidmem(0x5dc00, 0x40, gcvSURF_BITMAP, gcvPOOL_DEFAULT, &bmp_node, NULL)!=0)
{
fprintf(stderr, "Error allocating bitmap status memory\n");
exit(1);
}
printf("Allocated bitmap node: node=%08x\n", (uint32_t)bmp_node);
viv_addr_t bmp_physical = 0;
void *bmp_logical = 0;
if(viv_lock_vidmem(bmp_node, &bmp_physical, &bmp_logical)!=0)
{
fprintf(stderr, "Error locking bmp memory\n");
exit(1);
}
memset(bmp_logical, 0xff, 0x5dc00); /* clear previous result */
printf("Locked bmp: phys=%08x log=%08x\n", (uint32_t)bmp_physical, (uint32_t)bmp_logical);
/* Submit fourth command buffer, updating context.
* Fourth command buffer copies render result to bitmap, detiling along the way.
*/
cmdbuf4[0x0f] = fb.fb_fix.line_length;
cmdbuf4[0x19] = rt_physical;
cmdbuf4[0x1b] = fb.physical[0];
fb_set_buffer(&fb, 0);
/* XXX gcvHAL_MAP_USER_MEMORY to get dma-able address, or does this work as-is? */
commandBuffer.startOffset = commandBuffer.offset + 0x18;
memcpy((void*)((size_t)commandBuffer.logical + commandBuffer.startOffset), cmdbuf4, sizeof(cmdbuf4));
commandBuffer.offset = commandBuffer.startOffset + sizeof(cmdbuf4);
commandBuffer.free -= sizeof(cmdbuf4) + 0x18;
printf("[4] startOffset=%08x, offset=%08x, free=%08x\n", (uint32_t)commandBuffer.startOffset, (uint32_t)commandBuffer.offset, (uint32_t)commandBuffer.free);
if(viv_commit(&commandBuffer, &contextBuffer) != 0)
{
fprintf(stderr, "Error committing fourth command buffer\n");
exit(1);
}
/* Submit event queue with SIGNAL, fromWhere=gcvKERNEL_PIXEL */
if(viv_event_queue_signal(sig_id, gcvKERNEL_PIXEL) != 0)
{
fprintf(stderr, "Cannot queue GPU signal\n");
exit(1);
}
/* Wait for signal */
if(viv_user_signal_wait(sig_id, SIG_WAIT_INDEFINITE) != 0)
{
fprintf(stderr, "Cannot wait for signal\n");
exit(1);
}
bmp_dump32(bmp_logical, 400, 240, false, "/mnt/sdcard/replay.bmp");
/* Unlock video memory */
if(viv_unlock_vidmem(bmp_node, gcvSURF_BITMAP, 1) != 0)
{
fprintf(stderr, "Cannot unlock vidmem\n");
exit(1);
}
/*
for(int x=0; x<0x700; ++x)
{
uint32_t value = ((uint32_t*)rt_ts_logical)[x];
printf("Sample ts: %x %08x\n", x*4, value);
}*/
printf("Contextbuffer used %i\n", *contextBuffer.inUse);
viv_close();
return 0;
}
int etna_create(struct viv_conn *conn, struct etna_ctx **ctx_out)
{
int rv;
if(ctx_out == NULL) return ETNA_INVALID_ADDR;
struct etna_ctx *ctx = ETNA_CALLOC_STRUCT(etna_ctx);
if(ctx == NULL) return ETNA_OUT_OF_MEMORY;
ctx->conn = conn;
if(gpu_context_initialize(ctx) != ETNA_OK)
{
ETNA_FREE(ctx);
return ETNA_INTERNAL_ERROR;
}
/* Create synchronization signal */
if(viv_user_signal_create(conn, 0, &ctx->sig_id) != 0) /* automatic resetting signal */
{
#ifdef DEBUG
fprintf(stderr, "Cannot create user signal\n");
#endif
return ETNA_INTERNAL_ERROR;
}
#ifdef DEBUG
printf("Created user signal %i\n", ctx->sig_id);
#endif
/* Allocate command buffers, and create a synchronization signal for each.
* Also signal the synchronization signal for the buffers to tell that the buffers are ready for use.
*/
for(int x=0; x<NUM_COMMAND_BUFFERS; ++x)
{
ctx->cmdbuf[x] = ETNA_CALLOC_STRUCT(_gcoCMDBUF);
if(viv_alloc_contiguous(conn, COMMAND_BUFFER_SIZE, &ctx->cmdbufi[x].physical, &ctx->cmdbufi[x].logical, &ctx->cmdbufi[x].bytes)!=0)
{
#ifdef DEBUG
fprintf(stderr, "Error allocating host memory for command buffer\n");
#endif
return ETNA_OUT_OF_MEMORY;
}
ctx->cmdbuf[x]->object.type = gcvOBJ_COMMANDBUFFER;
#ifdef GCABI_CMDBUF_HAS_PHYSICAL
ctx->cmdbuf[x]->physical = PTR_TO_VIV((void*)ctx->cmdbufi[x].physical);
ctx->cmdbuf[x]->bytes = ctx->cmdbufi[x].bytes;
#endif
ctx->cmdbuf[x]->logical = PTR_TO_VIV((void*)ctx->cmdbufi[x].logical);
if(viv_user_signal_create(conn, 0, &ctx->cmdbufi[x].sig_id) != 0 ||
viv_user_signal_signal(conn, ctx->cmdbufi[x].sig_id, 1) != 0)
{
#ifdef DEBUG
fprintf(stderr, "Cannot create user signal\n");
#endif
return ETNA_INTERNAL_ERROR;
}
#ifdef DEBUG
printf("Allocated buffer %i: phys=%08x log=%08x bytes=%08x [signal %i]\n", x,
(uint32_t)buf0_physical, (uint32_t)buf0_logical, buf0_bytes, ctx->cmdbufi[x].sig);
#endif
}
/* Allocate command queue */
if((rv = etna_queue_create(ctx, &ctx->queue)) != ETNA_OK)
{
#ifdef DEBUG
fprintf(stderr, "Error allocating kernel command queue\n");
#endif
return rv;
}
/* Set current buffer to ETNA_NO_BUFFER, to signify that we need to switch to buffer 0 before
* queueing of commands can be started.
*/
ctx->cur_buf = ETNA_NO_BUFFER;
*ctx_out = ctx;
return ETNA_OK;
}
int main(int argc, char **argv)
{
int rv;
struct viv_conn *conn = 0;
rv = viv_open(VIV_HW_3D, &conn);
if(rv!=0)
{
fprintf(stderr, "Error opening device\n");
exit(1);
}
printf("Succesfully opened device\n");
viv_show_chip_info(conn);
/* allocate command buffer (blob uses four command buffers, but we don't even fill one) */
viv_addr_t buf0_physical = 0;
void *buf0_logical = 0;
if(viv_alloc_contiguous(conn, 0x20000, &buf0_physical, &buf0_logical, NULL)!=0)
{
fprintf(stderr, "Error allocating host memory\n");
exit(1);
}
printf("Allocated buffer: phys=%08x log=%08x\n", (uint32_t)buf0_physical, (uint32_t)buf0_logical);
/* allocate main render target */
gcuVIDMEM_NODE_PTR color_surface_node = 0;
if(viv_alloc_linear_vidmem(conn, 0x73000, 0x40, gcvSURF_RENDER_TARGET, gcvPOOL_SYSTEM /*why?*/, &color_surface_node, NULL)!=0)
{
fprintf(stderr, "Error allocating render target buffer memory\n");
exit(1);
}
printf("Allocated render target node: node=%08x\n", (uint32_t)color_surface_node);
viv_addr_t color_surface_physical = 0;
void *color_surface_logical = 0;
if(viv_lock_vidmem(conn, color_surface_node, &color_surface_physical, &color_surface_logical)!=0)
{
fprintf(stderr, "Error locking render target memory\n");
exit(1);
}
printf("Locked render target: phys=%08x log=%08x\n", (uint32_t)color_surface_physical, (uint32_t)color_surface_logical);
/* allocate tile status for main render target */
gcuVIDMEM_NODE_PTR color_status_node = 0;
if(viv_alloc_linear_vidmem(conn, 0x800, 0x40, gcvSURF_TILE_STATUS, gcvPOOL_DEFAULT, &color_status_node, NULL)!=0)
{
fprintf(stderr, "Error allocating render target tile status memory\n");
exit(1);
}
printf("Allocated render target tile status node: node=%08x\n", (uint32_t)color_status_node);
viv_addr_t color_status_physical = 0;
void *color_status_logical = 0;
if(viv_lock_vidmem(conn, color_status_node, &color_status_physical, &color_status_logical)!=0)
{
fprintf(stderr, "Error locking render target memory\n");
exit(1);
}
printf("Locked render target ts: phys=%08x log=%08x\n", (uint32_t)color_status_physical, (uint32_t)color_status_logical);
/* allocate depth for main render target */
gcuVIDMEM_NODE_PTR depth_surface_node = 0;
if(viv_alloc_linear_vidmem(conn, 0x45000, 0x40, gcvSURF_DEPTH, gcvPOOL_DEFAULT, &depth_surface_node, NULL)!=0)
{
fprintf(stderr, "Error allocating depth memory\n");
exit(1);
}
printf("Allocated depth node: node=%08x\n", (uint32_t)depth_surface_node);
viv_addr_t depth_surface_physical = 0;
void *depth_surface_logical = 0;
if(viv_lock_vidmem(conn, depth_surface_node, &depth_surface_physical, &depth_surface_logical)!=0)
{
fprintf(stderr, "Error locking depth target memory\n");
exit(1);
}
printf("Locked depth target: phys=%08x log=%08x\n", (uint32_t)depth_surface_physical, (uint32_t)depth_surface_logical);
/* allocate depth ts for main render target */
gcuVIDMEM_NODE_PTR depth_status_node = 0;
if(viv_alloc_linear_vidmem(conn, 0x500, 0x40, gcvSURF_TILE_STATUS, gcvPOOL_DEFAULT, &depth_status_node, NULL)!=0)
{
fprintf(stderr, "Error allocating depth memory\n");
exit(1);
}
printf("Allocated depth ts node: node=%08x\n", (uint32_t)depth_status_node);
viv_addr_t depth_status_physical = 0;
void *depth_status_logical = 0;
if(viv_lock_vidmem(conn, depth_status_node, &depth_status_physical, &depth_status_logical)!=0)
{
fprintf(stderr, "Error locking depth target ts memory\n");
exit(1);
}
printf("Locked depth ts target: phys=%08x log=%08x\n", (uint32_t)depth_status_physical, (uint32_t)depth_status_logical);
/* allocate tile status for aux render target */
gcuVIDMEM_NODE_PTR rs_dest_node = 0;
if(viv_alloc_linear_vidmem(conn, 0x70000, 0x40, gcvSURF_BITMAP, gcvPOOL_DEFAULT, &rs_dest_node, NULL)!=0)
{
fprintf(stderr, "Error allocating aux render target tile status memory\n");
exit(1);
}
printf("Allocated aux render target tile status node: node=%08x\n", (uint32_t)rs_dest_node);
viv_addr_t rs_dest_physical = 0;
void *rs_dest_logical = 0;
if(viv_lock_vidmem(conn, rs_dest_node, &rs_dest_physical, &rs_dest_logical)!=0)
{
fprintf(stderr, "Error locking aux ts render target memory\n");
exit(1);
}
printf("Locked aux render target ts: phys=%08x log=%08x\n", (uint32_t)rs_dest_physical, (uint32_t)rs_dest_logical);
/* allocate vertex buffer */
gcuVIDMEM_NODE_PTR vtx_node = 0;
if(viv_alloc_linear_vidmem(conn, 0x100000, 0x40, gcvSURF_VERTEX, gcvPOOL_DEFAULT, &vtx_node, NULL)!=0)
{
fprintf(stderr, "Error allocating vertex memory\n");
exit(1);
}
printf("Allocated vertex node: node=%08x\n", (uint32_t)vtx_node);
viv_addr_t vtx_physical = 0;
void *vtx_logical = 0;
if(viv_lock_vidmem(conn, vtx_node, &vtx_physical, &vtx_logical)!=0)
{
fprintf(stderr, "Error locking vertex memory\n");
exit(1);
}
printf("Locked vertex memory: phys=%08x log=%08x\n", (uint32_t)vtx_physical, (uint32_t)vtx_logical);
/* Phew, now we got all the memory we need.
* Write interleaved attribute vertex stream.
* Unlike the GL example we only do this once, not every time glDrawArrays is called, the same would be accomplished
* from GL by using a vertex buffer object.
*/
int dest_idx = 0;
int v_src_idx = 0;
int n_src_idx = 0;
int c_src_idx = 0;
for(int jj=0; jj<DRAW_COUNT; jj++)
{
for(int vert=0; vert<VERTICES_PER_DRAW*3; ++vert)
{
((float*)vtx_logical)[dest_idx] = vVertices[v_src_idx];
dest_idx++;
v_src_idx++;
}
for(int vert=0; vert<VERTICES_PER_DRAW*3; ++vert)
{
((float*)vtx_logical)[dest_idx] = vNormals[n_src_idx];
dest_idx++;
n_src_idx++;
}
for(int vert=0; vert<VERTICES_PER_DRAW*3; ++vert)
{
((float*)vtx_logical)[dest_idx] = vColors[c_src_idx];
dest_idx++;
c_src_idx++;
}
}
/*
* for(int idx=0; idx<NUM_VERTICES*3*3; ++idx)
* {
* printf("%i %f\n", idx, ((float*)vtx_logical)[idx]);
}*/
/* Load the command buffer and send the commit command. */
/* First build context state map */
size_t stateCount = 0x1d00;
uint32_t *contextMap = malloc(stateCount * 4);
memset(contextMap, 0, stateCount*4);
for(int idx=0; idx<sizeof(contextbuf_addr)/sizeof(address_index_t); ++idx)
{
contextMap[contextbuf_addr[idx].address / 4] = contextbuf_addr[idx].index;
}
struct _gcoCMDBUF commandBuffer = {
.object = {
.type = gcvOBJ_COMMANDBUFFER
},
//.os = (_gcoOS*)0xbf7488,
//.hardware = (_gcoHARDWARE*)0x402694e0,
.physical = (void*)buf0_physical,
.logical = (void*)buf0_logical,
.bytes = 0x20000,
.startOffset = 0x0,
//.offset = 0xac0,
//.free = 0x7520,
//.hintTable = (unsigned int*)0x0, // Used when gcdSECURE
//.hintIndex = (unsigned int*)0x58, // Used when gcdSECURE
//.hintCommit = (unsigned int*)0xffffffff // Used when gcdSECURE
};
gcsHAL_INTERFACE id = {};
id.command = gcvHAL_ATTACH;
if((viv_invoke(conn, &id)) != gcvSTATUS_OK)
{
#ifdef DEBUG
fprintf(stderr, "Error attaching to GPU\n");
#endif
exit(1);
}
else
{
fprintf(stderr, "gcvHAL_ATTACHed to GPU\n");
}
gckCONTEXT context = id.u.Attach.context;
commandBuffer.free = commandBuffer.bytes - 0x8; /* Always keep 0x8 at end of buffer for kernel driver */
/* Set addresses in first command buffer */
cmdbuf1[37] = cmdbuf1[87] = cmdbuf1[109] = color_status_physical;
cmdbuf1[38] = cmdbuf1[110] = cmdbuf1[213] = cmdbuf1[215] = color_surface_physical;
cmdbuf1[47] = depth_status_physical; //ADDR_J */ 0x500 gcvSURF_TILE_STATUS
cmdbuf1[48] = cmdbuf1[225] = cmdbuf1[227] = depth_surface_physical; //DDR_I */ 0x45000 gcvSURF_DEPTH
cmdbuf1[169] = vtx_physical;
cmdbuf1[170] = vtx_physical + 0x030;
cmdbuf1[171] = vtx_physical + 0x060;
cmdbuf1[413] = vtx_physical + 0x060;
cmdbuf1[414] = vtx_physical + 0x090;
cmdbuf1[415] = vtx_physical + 0x0c0;
cmdbuf1[435] = vtx_physical + 0x0c0;
cmdbuf1[436] = vtx_physical + 0x0f0;
cmdbuf1[437] = vtx_physical + 0x120;
cmdbuf1[457] = vtx_physical + 0x120;
cmdbuf1[458] = vtx_physical + 0x150;
cmdbuf1[459] = vtx_physical + 0x180;
cmdbuf1[479] = vtx_physical + 0x180;
cmdbuf1[480] = vtx_physical + 0x1b0;
cmdbuf1[481] = vtx_physical + 0x1e0;
cmdbuf1[501] = vtx_physical + 0x1e0;
cmdbuf1[502] = vtx_physical + 0x210;
cmdbuf1[503] = vtx_physical + 0x240;
/* Submit first command buffer */
commandBuffer.startOffset = 0;
memcpy((void*)((size_t)commandBuffer.logical + commandBuffer.startOffset), cmdbuf1, sizeof(cmdbuf1));
commandBuffer.offset = commandBuffer.startOffset + sizeof(cmdbuf1);
commandBuffer.free -= sizeof(cmdbuf1) + 0x08;
printf("[1] startOffset=%08x, offset=%08x, free=%08x\n", (uint32_t)commandBuffer.startOffset, (uint32_t)commandBuffer.offset, (uint32_t)commandBuffer.free);
if(viv_commit(conn, &commandBuffer, context) != 0)
{
fprintf(stderr, "Error committing first command buffer\n");
exit(1);
}
/*
* What does it do? Can be skipped.
*/
cmdbuf2[35] = color_surface_physical;
cmdbuf2[37] = color_surface_physical;
commandBuffer.startOffset = commandBuffer.offset + 0x08; /* Make space for LINK */
memcpy((void*)((size_t)commandBuffer.logical + commandBuffer.startOffset), cmdbuf2, sizeof(cmdbuf2));
commandBuffer.offset = commandBuffer.startOffset + sizeof(cmdbuf2);
commandBuffer.free -= sizeof(cmdbuf2) + 0x08;
printf("[2] startOffset=%08x, offset=%08x, free=%08x\n", (uint32_t)commandBuffer.startOffset, (uint32_t)commandBuffer.offset, (uint32_t)commandBuffer.free);
if(viv_commit(conn, &commandBuffer, context) != 0)
{
fprintf(stderr, "Error committing second command buffer\n");
exit(1);
}
/* Submit third command buffer - SWAP_RB=1 - swaps red and blue
**/
cmdbuf3[35] = color_surface_physical;
cmdbuf3[37] = rs_dest_physical;
commandBuffer.startOffset = commandBuffer.offset + 0x08;
memcpy((void*)((size_t)commandBuffer.logical + commandBuffer.startOffset), cmdbuf3, sizeof(cmdbuf3));
commandBuffer.offset = commandBuffer.startOffset + sizeof(cmdbuf3);
commandBuffer.free -= sizeof(cmdbuf3) + 0x08;
printf("[3] startOffset=%08x, offset=%08x, free=%08x\n", (uint32_t)commandBuffer.startOffset, (uint32_t)commandBuffer.offset, (uint32_t)commandBuffer.free);
if(viv_commit(conn, &commandBuffer, context) != 0)
{
fprintf(stderr, "Error committing third command buffer\n");
exit(1);
}
/* Submit event queue with SIGNAL, fromWhere=gcvKERNEL_PIXEL (wait for pixel engine to finish) */
int sig_id = 0;
if(viv_user_signal_create(conn, 0, &sig_id) != 0) /* automatic resetting signal */
{
fprintf(stderr, "Cannot create user signal\n");
exit(1);
}
printf("Created user signal %i\n", sig_id);
if(viv_event_queue_signal(conn, sig_id, gcvKERNEL_PIXEL) != 0)
{
fprintf(stderr, "Cannot queue GPU signal\n");
exit(1);
}
/* Wait for signal */
if(viv_user_signal_wait(conn, sig_id, VIV_WAIT_INDEFINITE) != 0)
{
fprintf(stderr, "Cannot wait for signal\n");
exit(1);
}
/* Allocate video memory for BITMAP, lock */
gcuVIDMEM_NODE_PTR bmp_node = 0;
if(viv_alloc_linear_vidmem(conn, 0x5dc00, 0x40, gcvSURF_BITMAP, gcvPOOL_DEFAULT, &bmp_node, NULL)!=0)
{
fprintf(stderr, "Error allocating bitmap status memory\n");
exit(1);
}
printf("Allocated bitmap node: node=%08x\n", (uint32_t)bmp_node);
viv_addr_t bmp_physical = 0;
void *bmp_logical = 0;
if(viv_lock_vidmem(conn, bmp_node, &bmp_physical, &bmp_logical)!=0)
{
fprintf(stderr, "Error locking bmp memory\n");
exit(1);
}
memset(bmp_logical, 0xff, 0x5dc00); /* clear previous result */
printf("Locked bmp: phys=%08x log=%08x\n", (uint32_t)bmp_physical, (uint32_t)bmp_logical);
/* Submit fourth command buffer, updating context.
* Fourth command buffer copies render result to bitmap, detiling along the way.
*/
/* color_surface_physical = cmdbuf2 or cmdbuf1 result, rs_dest_physical - cmdbuf3 result
* FIXME rs_dest_physical result is bad... why?
* turning off source tilling in cmdbuf4 helps but don't solve problem. */
cmdbuf4[0x19] = rs_dest_physical; //color_surface_physical rs_dest_physical
cmdbuf4[0x1b] = bmp_physical;
commandBuffer.startOffset = commandBuffer.offset + 0x08;
memcpy((void*)((size_t)commandBuffer.logical + commandBuffer.startOffset), cmdbuf4, sizeof(cmdbuf4));
commandBuffer.offset = commandBuffer.startOffset + sizeof(cmdbuf4);
commandBuffer.free -= sizeof(cmdbuf4) + 0x08;
printf("[4] startOffset=%08x, offset=%08x, free=%08x\n", (uint32_t)commandBuffer.startOffset, (uint32_t)commandBuffer.offset, (uint32_t)commandBuffer.free);
if(viv_commit(conn, &commandBuffer, context) != 0)
{
fprintf(stderr, "Error committing fourth command buffer\n");
exit(1);
}
/* Submit event queue with SIGNAL, fromWhere=gcvKERNEL_PIXEL */
if(viv_event_queue_signal(conn, sig_id, gcvKERNEL_PIXEL) != 0)
{
fprintf(stderr, "Cannot queue GPU signal\n");
exit(1);
}
/* Wait for signal */
if(viv_user_signal_wait(conn, sig_id, VIV_WAIT_INDEFINITE) != 0)
{
fprintf(stderr, "Cannot wait for signal\n");
exit(1);
}
bmp_dump32(bmp_logical, 400, 240, false, "/home/linaro/replay.bmp");
/* Unlock video memory */
if(viv_unlock_vidmem(conn, bmp_node, gcvSURF_BITMAP, 1) != 0)
{
fprintf(stderr, "Cannot unlock vidmem\n");
exit(1);
}
/*
* for(int x=0; x<0x700; ++x)
* {
* uint32_t value = ((uint32_t*)rt_ts_logical)[x];
* printf("Sample ts: %x %08x\n", x*4, value);
}*/
//printf("Contextbuffer used %i\n", *contextBuffer.inUse);
viv_close(conn);
return 0;
}
