static void
swr_resource_copy(struct pipe_context *pipe,
struct pipe_resource *dst,
unsigned dst_level,
unsigned dstx,
unsigned dsty,
unsigned dstz,
struct pipe_resource *src,
unsigned src_level,
const struct pipe_box *src_box)
{
struct swr_screen *screen = swr_screen(pipe->screen);
/* If either the src or dst is a renderTarget, store tiles before copy */
swr_store_dirty_resource(pipe, src, SWR_TILE_RESOLVED);
swr_store_dirty_resource(pipe, dst, SWR_TILE_RESOLVED);
swr_fence_finish(pipe->screen, NULL, screen->flush_fence, 0);
swr_resource_unused(src);
swr_resource_unused(dst);
if ((dst->target == PIPE_BUFFER && src->target == PIPE_BUFFER)
|| (dst->target != PIPE_BUFFER && src->target != PIPE_BUFFER)) {
util_resource_copy_region(
pipe, dst, dst_level, dstx, dsty, dstz, src, src_level, src_box);
return;
}
debug_printf("unhandled swr_resource_copy\n");
}
void
swr_store_dirty_resource(struct pipe_context *pipe,
struct pipe_resource *resource,
enum SWR_TILE_STATE post_tile_state)
{
/* Only store resource if it has been written to */
if (swr_resource(resource)->status & SWR_RESOURCE_WRITE) {
struct swr_context *ctx = swr_context(pipe);
struct swr_screen *screen = swr_screen(pipe->screen);
struct swr_resource *spr = swr_resource(resource);
swr_draw_context *pDC = &ctx->swrDC;
SWR_SURFACE_STATE *renderTargets = pDC->renderTargets;
for (uint32_t i = 0; i < SWR_NUM_ATTACHMENTS; i++)
if (renderTargets[i].pBaseAddress == spr->swr.pBaseAddress) {
swr_store_render_target(pipe, i, post_tile_state);
/* Mesa thinks depth/stencil are fused, so we'll never get an
* explicit resource for stencil. So, if checking depth, then
* also check for stencil. */
if (spr->has_stencil && (i == SWR_ATTACHMENT_DEPTH)) {
swr_store_render_target(
pipe, SWR_ATTACHMENT_STENCIL, post_tile_state);
}
/* This fence signals StoreTiles completion */
swr_fence_submit(ctx, screen->flush_fence);
break;
}
}
}
static void
swr_resource_destroy(struct pipe_screen *p_screen, struct pipe_resource *pt)
{
struct swr_screen *screen = swr_screen(p_screen);
struct swr_resource *spr = swr_resource(pt);
struct pipe_context *pipe = screen->pipe;
/* Only wait on fence if the resource is being used */
if (pipe && spr->status) {
/* But, if there's no fence pending, submit one.
* XXX: Remove once draw timestamps are implmented. */
if (!swr_is_fence_pending(screen->flush_fence))
swr_fence_submit(swr_context(pipe), screen->flush_fence);
swr_fence_finish(p_screen, screen->flush_fence, 0);
swr_resource_unused(pt);
}
/*
* Free resource primary surface. If resource is display target, winsys
* manages the buffer and will free it on displaytarget_destroy.
*/
if (spr->display_target) {
/* display target */
struct sw_winsys *winsys = screen->winsys;
winsys->displaytarget_destroy(winsys, spr->display_target);
} else
AlignedFree(spr->swr.pBaseAddress);
AlignedFree(spr->secondary.pBaseAddress);
FREE(spr);
}
static void
swr_resource_destroy(struct pipe_screen *p_screen, struct pipe_resource *pt)
{
struct swr_screen *screen = swr_screen(p_screen);
struct swr_resource *spr = swr_resource(pt);
struct pipe_context *pipe = screen->pipe;
if (spr->display_target) {
/* If resource is display target, winsys manages the buffer and will
* free it on displaytarget_destroy. */
swr_fence_finish(p_screen, NULL, screen->flush_fence, 0);
struct sw_winsys *winsys = screen->winsys;
winsys->displaytarget_destroy(winsys, spr->display_target);
} else {
/* For regular resources, if the resource is being used, defer deletion
* (use aligned-free) */
if (pipe && spr->status) {
swr_resource_unused(pt);
swr_fence_work_free(screen->flush_fence,
spr->swr.pBaseAddress, true);
swr_fence_work_free(screen->flush_fence,
spr->secondary.pBaseAddress, true);
} else {
AlignedFree(spr->swr.pBaseAddress);
AlignedFree(spr->secondary.pBaseAddress);
}
}
FREE(spr);
}
static boolean
swr_is_format_supported(struct pipe_screen *screen,
enum pipe_format format,
enum pipe_texture_target target,
unsigned sample_count,
unsigned bind)
{
struct sw_winsys *winsys = swr_screen(screen)->winsys;
const struct util_format_description *format_desc;
assert(target == PIPE_BUFFER || target == PIPE_TEXTURE_1D
|| target == PIPE_TEXTURE_1D_ARRAY
|| target == PIPE_TEXTURE_2D
|| target == PIPE_TEXTURE_2D_ARRAY
|| target == PIPE_TEXTURE_RECT
|| target == PIPE_TEXTURE_3D
|| target == PIPE_TEXTURE_CUBE
|| target == PIPE_TEXTURE_CUBE_ARRAY);
format_desc = util_format_description(format);
if (!format_desc)
return FALSE;
if (sample_count > 1)
return FALSE;
if (bind
& (PIPE_BIND_DISPLAY_TARGET | PIPE_BIND_SCANOUT | PIPE_BIND_SHARED)) {
if (!winsys->is_displaytarget_format_supported(winsys, bind, format))
return FALSE;
}
if (bind & PIPE_BIND_RENDER_TARGET) {
if (format_desc->colorspace == UTIL_FORMAT_COLORSPACE_ZS)
return FALSE;
if (mesa_to_swr_format(format) == (SWR_FORMAT)-1)
return FALSE;
/*
* Although possible, it is unnatural to render into compressed or YUV
* surfaces. So disable these here to avoid going into weird paths
* inside the state trackers.
*/
if (format_desc->block.width != 1 || format_desc->block.height != 1)
return FALSE;
}
if (bind & PIPE_BIND_DEPTH_STENCIL) {
if (format_desc->colorspace != UTIL_FORMAT_COLORSPACE_ZS)
return FALSE;
if (mesa_to_swr_format(format) == (SWR_FORMAT)-1)
return FALSE;
}
return TRUE;
}
static boolean
swr_can_create_resource(struct pipe_screen *screen,
const struct pipe_resource *templat)
{
struct swr_resource res;
memset(&res, 0, sizeof(res));
res.base = *templat;
return swr_texture_layout(swr_screen(screen), &res, false);
}
static void
swr_destroy(struct pipe_context *pipe)
{
struct swr_context *ctx = swr_context(pipe);
struct swr_screen *screen = swr_screen(pipe->screen);
if (ctx->blitter)
util_blitter_destroy(ctx->blitter);
for (unsigned i = 0; i < PIPE_MAX_COLOR_BUFS; i++) {
if (ctx->framebuffer.cbufs[i]) {
struct swr_resource *res = swr_resource(ctx->framebuffer.cbufs[i]->texture);
/* NULL curr_pipe, so we don't have a reference to a deleted pipe */
res->curr_pipe = NULL;
pipe_surface_reference(&ctx->framebuffer.cbufs[i], NULL);
}
}
if (ctx->framebuffer.zsbuf) {
struct swr_resource *res = swr_resource(ctx->framebuffer.zsbuf->texture);
/* NULL curr_pipe, so we don't have a reference to a deleted pipe */
res->curr_pipe = NULL;
pipe_surface_reference(&ctx->framebuffer.zsbuf, NULL);
}
for (unsigned i = 0; i < ARRAY_SIZE(ctx->sampler_views[0]); i++) {
pipe_sampler_view_reference(&ctx->sampler_views[PIPE_SHADER_FRAGMENT][i], NULL);
}
for (unsigned i = 0; i < ARRAY_SIZE(ctx->sampler_views[0]); i++) {
pipe_sampler_view_reference(&ctx->sampler_views[PIPE_SHADER_VERTEX][i], NULL);
}
if (ctx->pipe.stream_uploader)
u_upload_destroy(ctx->pipe.stream_uploader);
/* Idle core after destroying buffer resources, but before deleting
* context. Destroying resources has potentially called StoreTiles.*/
ctx->api.pfnSwrWaitForIdle(ctx->swrContext);
if (ctx->swrContext)
ctx->api.pfnSwrDestroyContext(ctx->swrContext);
delete ctx->blendJIT;
swr_destroy_scratch_buffers(ctx);
/* Only update screen->pipe if current context is being destroyed */
assert(screen);
if (screen->pipe == pipe)
screen->pipe = NULL;
AlignedFree(ctx);
}
static struct pipe_resource *
swr_resource_create(struct pipe_screen *_screen,
const struct pipe_resource *templat)
{
struct swr_screen *screen = swr_screen(_screen);
struct swr_resource *res = CALLOC_STRUCT(swr_resource);
if (!res)
return NULL;
res->base = *templat;
pipe_reference_init(&res->base.reference, 1);
res->base.screen = &screen->base;
if (swr_resource_is_texture(&res->base)) {
if (res->base.bind & (PIPE_BIND_DISPLAY_TARGET | PIPE_BIND_SCANOUT
| PIPE_BIND_SHARED)) {
/* displayable surface
* first call swr_texture_layout without allocating to finish
* filling out the SWR_SURFAE_STATE in res */
swr_texture_layout(screen, res, false);
if (!swr_displaytarget_layout(screen, res))
goto fail;
} else {
/* texture map */
if (!swr_texture_layout(screen, res, true))
goto fail;
}
} else {
/* other data (vertex buffer, const buffer, etc) */
assert(util_format_get_blocksize(templat->format) == 1);
assert(templat->height0 == 1);
assert(templat->depth0 == 1);
assert(templat->last_level == 0);
/* Easiest to just call swr_texture_layout, as it sets up
* SWR_SURFAE_STATE in res */
if (!swr_texture_layout(screen, res, true))
goto fail;
}
return &res->base;
fail:
FREE(res);
return NULL;
}
static void
swr_destroy_screen(struct pipe_screen *p_screen)
{
struct swr_screen *screen = swr_screen(p_screen);
struct sw_winsys *winsys = screen->winsys;
fprintf(stderr, "SWR destroy screen!\n");
swr_fence_finish(p_screen, screen->flush_fence, 0);
swr_fence_reference(p_screen, &screen->flush_fence, NULL);
JitDestroyContext(screen->hJitMgr);
if (winsys->destroy)
winsys->destroy(winsys);
FREE(screen);
}
static void
swr_flush(struct pipe_context *pipe,
struct pipe_fence_handle **fence,
unsigned flags)
{
struct swr_context *ctx = swr_context(pipe);
struct swr_screen *screen = swr_screen(pipe->screen);
struct pipe_surface *cb = ctx->framebuffer.cbufs[0];
/* If the current renderTarget is the display surface, store tiles back to
* the surface, in preparation for present (swr_flush_frontbuffer).
* Other renderTargets get stored back when attachment changes or
* swr_surface_destroy */
if (cb && swr_resource(cb->texture)->display_target)
swr_store_dirty_resource(pipe, cb->texture, SWR_TILE_RESOLVED);
if (fence)
swr_fence_reference(pipe->screen, fence, screen->flush_fence);
}
static void
swr_destroy(struct pipe_context *pipe)
{
struct swr_context *ctx = swr_context(pipe);
struct swr_screen *screen = swr_screen(pipe->screen);
if (ctx->blitter)
util_blitter_destroy(ctx->blitter);
/* Idle core before deleting context */
SwrWaitForIdle(ctx->swrContext);
for (unsigned i = 0; i < PIPE_MAX_COLOR_BUFS; i++) {
pipe_surface_reference(&ctx->framebuffer.cbufs[i], NULL);
}
pipe_surface_reference(&ctx->framebuffer.zsbuf, NULL);
for (unsigned i = 0; i < ARRAY_SIZE(ctx->sampler_views[0]); i++) {
pipe_sampler_view_reference(&ctx->sampler_views[PIPE_SHADER_FRAGMENT][i], NULL);
}
for (unsigned i = 0; i < ARRAY_SIZE(ctx->sampler_views[0]); i++) {
pipe_sampler_view_reference(&ctx->sampler_views[PIPE_SHADER_VERTEX][i], NULL);
}
if (ctx->swrContext)
SwrDestroyContext(ctx->swrContext);
delete ctx->blendJIT;
swr_destroy_scratch_buffers(ctx);
/* Only update screen->pipe if current context is being destroyed */
assert(screen);
if (screen->pipe == pipe)
screen->pipe = NULL;
FREE(ctx);
}
static void
swr_flush(struct pipe_context *pipe,
struct pipe_fence_handle **fence,
unsigned flags)
{
struct swr_context *ctx = swr_context(pipe);
struct swr_screen *screen = swr_screen(pipe->screen);
for (int i=0; i < ctx->framebuffer.nr_cbufs; i++) {
struct pipe_surface *cb = ctx->framebuffer.cbufs[i];
if (cb) {
swr_store_dirty_resource(pipe, cb->texture, SWR_TILE_RESOLVED);
}
}
if (ctx->framebuffer.zsbuf) {
swr_store_dirty_resource(pipe, ctx->framebuffer.zsbuf->texture,
SWR_TILE_RESOLVED);
}
if (fence)
swr_fence_reference(pipe->screen, fence, screen->flush_fence);
}
static void
swr_flush_frontbuffer(struct pipe_screen *p_screen,
struct pipe_resource *resource,
unsigned level,
unsigned layer,
void *context_private,
struct pipe_box *sub_box)
{
struct swr_screen *screen = swr_screen(p_screen);
struct sw_winsys *winsys = screen->winsys;
struct swr_resource *spr = swr_resource(resource);
struct pipe_context *pipe = screen->pipe;
if (pipe) {
swr_fence_finish(p_screen, screen->flush_fence, 0);
swr_resource_unused(resource);
SwrEndFrame(swr_context(pipe)->swrContext);
}
debug_assert(spr->display_target);
if (spr->display_target)
winsys->displaytarget_display(
winsys, spr->display_target, context_private, sub_box);
}
struct pipe_context *
swr_create_context(struct pipe_screen *p_screen, void *priv, unsigned flags)
{
struct swr_context *ctx = (struct swr_context *)
AlignedMalloc(sizeof(struct swr_context), KNOB_SIMD_BYTES);
memset(ctx, 0, sizeof(struct swr_context));
swr_screen(p_screen)->pfnSwrGetInterface(ctx->api);
ctx->swrDC.pAPI = &ctx->api;
ctx->blendJIT =
new std::unordered_map<BLEND_COMPILE_STATE, PFN_BLEND_JIT_FUNC>;
ctx->max_draws_in_flight = KNOB_MAX_DRAWS_IN_FLIGHT;
SWR_CREATECONTEXT_INFO createInfo;
memset(&createInfo, 0, sizeof(createInfo));
createInfo.privateStateSize = sizeof(swr_draw_context);
createInfo.pfnLoadTile = swr_LoadHotTile;
createInfo.pfnStoreTile = swr_StoreHotTile;
createInfo.pfnClearTile = swr_StoreHotTileClear;
createInfo.pfnUpdateStats = swr_UpdateStats;
createInfo.pfnUpdateStatsFE = swr_UpdateStatsFE;
SWR_THREADING_INFO threadingInfo {0};
threadingInfo.MAX_WORKER_THREADS = KNOB_MAX_WORKER_THREADS;
threadingInfo.MAX_NUMA_NODES = KNOB_MAX_NUMA_NODES;
threadingInfo.MAX_CORES_PER_NUMA_NODE = KNOB_MAX_CORES_PER_NUMA_NODE;
threadingInfo.MAX_THREADS_PER_CORE = KNOB_MAX_THREADS_PER_CORE;
threadingInfo.SINGLE_THREADED = KNOB_SINGLE_THREADED;
// Use non-standard settings for KNL
if (swr_screen(p_screen)->is_knl)
{
if (nullptr == getenv("KNOB_MAX_THREADS_PER_CORE"))
threadingInfo.MAX_THREADS_PER_CORE = 2;
if (nullptr == getenv("KNOB_MAX_DRAWS_IN_FLIGHT"))
{
ctx->max_draws_in_flight = 2048;
createInfo.MAX_DRAWS_IN_FLIGHT = ctx->max_draws_in_flight;
}
}
createInfo.pThreadInfo = &threadingInfo;
ctx->swrContext = ctx->api.pfnSwrCreateContext(&createInfo);
ctx->api.pfnSwrInit();
if (ctx->swrContext == NULL)
goto fail;
ctx->pipe.screen = p_screen;
ctx->pipe.destroy = swr_destroy;
ctx->pipe.priv = priv;
ctx->pipe.create_surface = swr_create_surface;
ctx->pipe.surface_destroy = swr_surface_destroy;
ctx->pipe.transfer_map = swr_transfer_map;
ctx->pipe.transfer_unmap = swr_transfer_unmap;
ctx->pipe.transfer_flush_region = swr_transfer_flush_region;
ctx->pipe.buffer_subdata = u_default_buffer_subdata;
ctx->pipe.texture_subdata = u_default_texture_subdata;
ctx->pipe.clear_texture = util_clear_texture;
ctx->pipe.resource_copy_region = swr_resource_copy;
ctx->pipe.render_condition = swr_render_condition;
swr_state_init(&ctx->pipe);
swr_clear_init(&ctx->pipe);
swr_draw_init(&ctx->pipe);
swr_query_init(&ctx->pipe);
ctx->pipe.stream_uploader = u_upload_create_default(&ctx->pipe);
if (!ctx->pipe.stream_uploader)
goto fail;
ctx->pipe.const_uploader = ctx->pipe.stream_uploader;
ctx->pipe.blit = swr_blit;
ctx->blitter = util_blitter_create(&ctx->pipe);
if (!ctx->blitter)
goto fail;
swr_init_scratch_buffers(ctx);
return &ctx->pipe;
fail:
/* Should really validate the init steps and fail gracefully */
swr_destroy(&ctx->pipe);
return NULL;
}
static void *
swr_transfer_map(struct pipe_context *pipe,
struct pipe_resource *resource,
unsigned level,
unsigned usage,
const struct pipe_box *box,
struct pipe_transfer **transfer)
{
struct swr_screen *screen = swr_screen(pipe->screen);
struct swr_resource *spr = swr_resource(resource);
struct pipe_transfer *pt;
enum pipe_format format = resource->format;
assert(resource);
assert(level <= resource->last_level);
/* If mapping an attached rendertarget, store tiles to surface and set
* postStoreTileState to SWR_TILE_INVALID so tiles get reloaded on next use
* and nothing needs to be done at unmap. */
swr_store_dirty_resource(pipe, resource, SWR_TILE_INVALID);
if (!(usage & PIPE_TRANSFER_UNSYNCHRONIZED)) {
/* If resource is in use, finish fence before mapping.
* Unless requested not to block, then if not done return NULL map */
if (usage & PIPE_TRANSFER_DONTBLOCK) {
if (swr_is_fence_pending(screen->flush_fence))
return NULL;
} else {
if (spr->status) {
/* But, if there's no fence pending, submit one.
* XXX: Remove once draw timestamps are finished. */
if (!swr_is_fence_pending(screen->flush_fence))
swr_fence_submit(swr_context(pipe), screen->flush_fence);
swr_fence_finish(pipe->screen, screen->flush_fence, 0);
swr_resource_unused(resource);
}
}
}
pt = CALLOC_STRUCT(pipe_transfer);
if (!pt)
return NULL;
pipe_resource_reference(&pt->resource, resource);
pt->level = level;
pt->box = *box;
pt->stride = spr->row_stride[level];
pt->layer_stride = spr->img_stride[level];
/* if we're mapping the depth/stencil, copy in stencil */
if (spr->base.format == PIPE_FORMAT_Z24_UNORM_S8_UINT
&& spr->has_stencil) {
for (unsigned i = 0; i < spr->alignedWidth * spr->alignedHeight; i++) {
spr->swr.pBaseAddress[4 * i + 3] = spr->secondary.pBaseAddress[i];
}
} else if (spr->base.format == PIPE_FORMAT_Z32_FLOAT_S8X24_UINT
&& spr->has_stencil) {
for (unsigned i = 0; i < spr->alignedWidth * spr->alignedHeight; i++) {
spr->swr.pBaseAddress[8 * i + 4] = spr->secondary.pBaseAddress[i];
}
}
unsigned offset = box->z * pt->layer_stride + box->y * pt->stride
+ box->x * util_format_get_blocksize(format);
*transfer = pt;
return spr->swr.pBaseAddress + offset + spr->mip_offsets[level];
}
/*
* Draw vertex arrays, with optional indexing, optional instancing.
*/
static void
swr_draw_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info)
{
struct swr_context *ctx = swr_context(pipe);
if (!info->count_from_stream_output && !info->indirect &&
!info->primitive_restart &&
!u_trim_pipe_prim(info->mode, (unsigned*)&info->count))
return;
if (!swr_check_render_cond(pipe))
return;
if (info->indirect) {
util_draw_indirect(pipe, info);
return;
}
/* If indexed draw, force vertex validation since index buffer comes
* from draw info. */
if (info->index_size)
ctx->dirty |= SWR_NEW_VERTEX;
/* Update derived state, pass draw info to update function. */
swr_update_derived(pipe, info);
swr_update_draw_context(ctx);
if (ctx->vs->pipe.stream_output.num_outputs) {
if (!ctx->vs->soFunc[info->mode]) {
STREAMOUT_COMPILE_STATE state = {0};
struct pipe_stream_output_info *so = &ctx->vs->pipe.stream_output;
state.numVertsPerPrim = u_vertices_per_prim(info->mode);
uint32_t offsets[MAX_SO_STREAMS] = {0};
uint32_t num = 0;
for (uint32_t i = 0; i < so->num_outputs; i++) {
assert(so->output[i].stream == 0); // @todo
uint32_t output_buffer = so->output[i].output_buffer;
if (so->output[i].dst_offset != offsets[output_buffer]) {
// hole - need to fill
state.stream.decl[num].bufferIndex = output_buffer;
state.stream.decl[num].hole = true;
state.stream.decl[num].componentMask =
(1 << (so->output[i].dst_offset - offsets[output_buffer]))
- 1;
num++;
offsets[output_buffer] = so->output[i].dst_offset;
}
unsigned attrib_slot = so->output[i].register_index;
attrib_slot = swr_so_adjust_attrib(attrib_slot, ctx->vs);
state.stream.decl[num].bufferIndex = output_buffer;
state.stream.decl[num].attribSlot = attrib_slot;
state.stream.decl[num].componentMask =
((1 << so->output[i].num_components) - 1)
<< so->output[i].start_component;
state.stream.decl[num].hole = false;
num++;
offsets[output_buffer] += so->output[i].num_components;
}
state.stream.numDecls = num;
HANDLE hJitMgr = swr_screen(pipe->screen)->hJitMgr;
ctx->vs->soFunc[info->mode] = JitCompileStreamout(hJitMgr, state);
debug_printf("so shader %p\n", ctx->vs->soFunc[info->mode]);
assert(ctx->vs->soFunc[info->mode] && "Error: SoShader = NULL");
}
ctx->api.pfnSwrSetSoFunc(ctx->swrContext, ctx->vs->soFunc[info->mode], 0);
}
struct swr_vertex_element_state *velems = ctx->velems;
if (info->primitive_restart)
velems->fsState.cutIndex = info->restart_index;
else
velems->fsState.cutIndex = 0;
velems->fsState.bEnableCutIndex = info->primitive_restart;
velems->fsState.bPartialVertexBuffer = (info->min_index > 0);
swr_jit_fetch_key key;
swr_generate_fetch_key(key, velems);
auto search = velems->map.find(key);
if (search != velems->map.end()) {
velems->fsFunc = search->second;
} else {
HANDLE hJitMgr = swr_screen(ctx->pipe.screen)->hJitMgr;
velems->fsFunc = JitCompileFetch(hJitMgr, velems->fsState);
debug_printf("fetch shader %p\n", velems->fsFunc);
assert(velems->fsFunc && "Error: FetchShader = NULL");
velems->map.insert(std::make_pair(key, velems->fsFunc));
}
ctx->api.pfnSwrSetFetchFunc(ctx->swrContext, velems->fsFunc);
/* Set up frontend state
* XXX setup provokingVertex & topologyProvokingVertex */
SWR_FRONTEND_STATE feState = {0};
// feState.vsVertexSize seeds the PA size that is used as an interface
// between all the shader stages, so it has to be large enough to
// incorporate all interfaces between stages
// max of gs and vs num_outputs
feState.vsVertexSize = ctx->vs->info.base.num_outputs;
if (ctx->gs &&
ctx->gs->info.base.num_outputs > feState.vsVertexSize) {
feState.vsVertexSize = ctx->gs->info.base.num_outputs;
}
if (ctx->vs->info.base.num_outputs) {
// gs does not adjust for position in SGV slot at input from vs
if (!ctx->gs)
feState.vsVertexSize--;
}
// other (non-SGV) slots start at VERTEX_ATTRIB_START_SLOT
feState.vsVertexSize += VERTEX_ATTRIB_START_SLOT;
// The PA in the clipper does not handle BE vertex sizes
// different from FE. Increase vertexsize only for the cases that needed it
// primid needs a slot
if (ctx->fs->info.base.uses_primid)
feState.vsVertexSize++;
// sprite coord enable
if (ctx->rasterizer->sprite_coord_enable)
feState.vsVertexSize++;
if (ctx->rasterizer->flatshade_first) {
feState.provokingVertex = {1, 0, 0};
} else {
feState.provokingVertex = {2, 1, 2};
}
enum pipe_prim_type topology;
if (ctx->gs)
topology = (pipe_prim_type)ctx->gs->info.base.properties[TGSI_PROPERTY_GS_OUTPUT_PRIM];
else
topology = info->mode;
switch (topology) {
case PIPE_PRIM_TRIANGLE_FAN:
feState.topologyProvokingVertex = feState.provokingVertex.triFan;
break;
case PIPE_PRIM_TRIANGLE_STRIP:
case PIPE_PRIM_TRIANGLES:
feState.topologyProvokingVertex = feState.provokingVertex.triStripList;
break;
case PIPE_PRIM_QUAD_STRIP:
case PIPE_PRIM_QUADS:
if (ctx->rasterizer->flatshade_first)
feState.topologyProvokingVertex = 0;
else
feState.topologyProvokingVertex = 3;
break;
case PIPE_PRIM_LINES:
case PIPE_PRIM_LINE_LOOP:
case PIPE_PRIM_LINE_STRIP:
feState.topologyProvokingVertex = feState.provokingVertex.lineStripList;
break;
default:
feState.topologyProvokingVertex = 0;
}
feState.bEnableCutIndex = info->primitive_restart;
ctx->api.pfnSwrSetFrontendState(ctx->swrContext, &feState);
if (info->index_size)
ctx->api.pfnSwrDrawIndexedInstanced(ctx->swrContext,
swr_convert_prim_topology(info->mode),
info->count,
info->instance_count,
info->start,
info->index_bias,
info->start_instance);
else
ctx->api.pfnSwrDrawInstanced(ctx->swrContext,
swr_convert_prim_topology(info->mode),
info->count,
info->instance_count,
info->start,
info->start_instance);
/* On large client-buffer draw, we used client buffer directly, without
* copy. Block until draw is finished.
* VMD is an example application that benefits from this. */
if (ctx->dirty & SWR_LARGE_CLIENT_DRAW) {
struct swr_screen *screen = swr_screen(pipe->screen);
swr_fence_submit(ctx, screen->flush_fence);
swr_fence_finish(pipe->screen, NULL, screen->flush_fence, 0);
}
}
static void *
swr_transfer_map(struct pipe_context *pipe,
struct pipe_resource *resource,
unsigned level,
unsigned usage,
const struct pipe_box *box,
struct pipe_transfer **transfer)
{
struct swr_screen *screen = swr_screen(pipe->screen);
struct swr_resource *spr = swr_resource(resource);
struct pipe_transfer *pt;
enum pipe_format format = resource->format;
assert(resource);
assert(level <= resource->last_level);
/* If mapping an attached rendertarget, store tiles to surface and set
* postStoreTileState to SWR_TILE_INVALID so tiles get reloaded on next use
* and nothing needs to be done at unmap. */
swr_store_dirty_resource(pipe, resource, SWR_TILE_INVALID);
if (!(usage & PIPE_TRANSFER_UNSYNCHRONIZED)) {
/* If resource is in use, finish fence before mapping.
* Unless requested not to block, then if not done return NULL map */
if (usage & PIPE_TRANSFER_DONTBLOCK) {
if (swr_is_fence_pending(screen->flush_fence))
return NULL;
} else {
if (spr->status) {
/* But, if there's no fence pending, submit one.
* XXX: Remove once draw timestamps are finished. */
if (!swr_is_fence_pending(screen->flush_fence))
swr_fence_submit(swr_context(pipe), screen->flush_fence);
swr_fence_finish(pipe->screen, NULL, screen->flush_fence, 0);
swr_resource_unused(resource);
}
}
}
pt = CALLOC_STRUCT(pipe_transfer);
if (!pt)
return NULL;
pipe_resource_reference(&pt->resource, resource);
pt->usage = (pipe_transfer_usage)usage;
pt->level = level;
pt->box = *box;
pt->stride = spr->swr.pitch;
pt->layer_stride = spr->swr.qpitch * spr->swr.pitch;
/* if we're mapping the depth/stencil, copy in stencil for the section
* being read in
*/
if (usage & PIPE_TRANSFER_READ && spr->has_depth && spr->has_stencil) {
size_t zbase, sbase;
for (int z = box->z; z < box->z + box->depth; z++) {
zbase = (z * spr->swr.qpitch + box->y) * spr->swr.pitch +
spr->mip_offsets[level];
sbase = (z * spr->secondary.qpitch + box->y) * spr->secondary.pitch +
spr->secondary_mip_offsets[level];
for (int y = box->y; y < box->y + box->height; y++) {
if (spr->base.format == PIPE_FORMAT_Z24_UNORM_S8_UINT) {
for (int x = box->x; x < box->x + box->width; x++)
((uint8_t*)(spr->swr.xpBaseAddress))[zbase + 4 * x + 3] =
((uint8_t*)(spr->secondary.xpBaseAddress))[sbase + x];
} else if (spr->base.format == PIPE_FORMAT_Z32_FLOAT_S8X24_UINT) {
for (int x = box->x; x < box->x + box->width; x++)
((uint8_t*)(spr->swr.xpBaseAddress))[zbase + 8 * x + 4] =
((uint8_t*)(spr->secondary.xpBaseAddress))[sbase + x];
}
zbase += spr->swr.pitch;
sbase += spr->secondary.pitch;
}
}
}
unsigned offset = box->z * pt->layer_stride +
util_format_get_nblocksy(format, box->y) * pt->stride +
util_format_get_stride(format, box->x);
*transfer = pt;
return (void*)(spr->swr.xpBaseAddress + offset + spr->mip_offsets[level]);
}
/*
* Draw vertex arrays, with optional indexing, optional instancing.
*/
static void
swr_draw_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info)
{
struct swr_context *ctx = swr_context(pipe);
if (!swr_check_render_cond(pipe))
return;
if (info->indirect) {
util_draw_indirect(pipe, info);
return;
}
/* Update derived state, pass draw info to update function */
if (ctx->dirty)
swr_update_derived(pipe, info);
swr_update_draw_context(ctx);
if (ctx->vs->pipe.stream_output.num_outputs) {
if (!ctx->vs->soFunc[info->mode]) {
STREAMOUT_COMPILE_STATE state = {0};
struct pipe_stream_output_info *so = &ctx->vs->pipe.stream_output;
state.numVertsPerPrim = u_vertices_per_prim(info->mode);
uint32_t offsets[MAX_SO_STREAMS] = {0};
uint32_t num = 0;
for (uint32_t i = 0; i < so->num_outputs; i++) {
assert(so->output[i].stream == 0); // @todo
uint32_t output_buffer = so->output[i].output_buffer;
if (so->output[i].dst_offset != offsets[output_buffer]) {
// hole - need to fill
state.stream.decl[num].bufferIndex = output_buffer;
state.stream.decl[num].hole = true;
state.stream.decl[num].componentMask =
(1 << (so->output[i].dst_offset - offsets[output_buffer]))
- 1;
num++;
offsets[output_buffer] = so->output[i].dst_offset;
}
state.stream.decl[num].bufferIndex = output_buffer;
state.stream.decl[num].attribSlot = so->output[i].register_index - 1;
state.stream.decl[num].componentMask =
((1 << so->output[i].num_components) - 1)
<< so->output[i].start_component;
state.stream.decl[num].hole = false;
num++;
offsets[output_buffer] += so->output[i].num_components;
}
state.stream.numDecls = num;
HANDLE hJitMgr = swr_screen(pipe->screen)->hJitMgr;
ctx->vs->soFunc[info->mode] = JitCompileStreamout(hJitMgr, state);
debug_printf("so shader %p\n", ctx->vs->soFunc[info->mode]);
assert(ctx->vs->soFunc[info->mode] && "Error: SoShader = NULL");
}
SwrSetSoFunc(ctx->swrContext, ctx->vs->soFunc[info->mode], 0);
}
struct swr_vertex_element_state *velems = ctx->velems;
if (!velems->fsFunc
|| (velems->fsState.cutIndex != info->restart_index)
|| (velems->fsState.bEnableCutIndex != info->primitive_restart)) {
velems->fsState.cutIndex = info->restart_index;
velems->fsState.bEnableCutIndex = info->primitive_restart;
/* Create Fetch Shader */
HANDLE hJitMgr = swr_screen(ctx->pipe.screen)->hJitMgr;
velems->fsFunc = JitCompileFetch(hJitMgr, velems->fsState);
debug_printf("fetch shader %p\n", velems->fsFunc);
assert(velems->fsFunc && "Error: FetchShader = NULL");
}
SwrSetFetchFunc(ctx->swrContext, velems->fsFunc);
/* Set up frontend state
* XXX setup provokingVertex & topologyProvokingVertex */
SWR_FRONTEND_STATE feState = {0};
if (ctx->rasterizer->flatshade_first) {
feState.provokingVertex = {1, 0, 0};
} else {
feState.provokingVertex = {2, 1, 2};
}
switch (info->mode) {
case PIPE_PRIM_TRIANGLE_FAN:
feState.topologyProvokingVertex = feState.provokingVertex.triFan;
break;
case PIPE_PRIM_TRIANGLE_STRIP:
case PIPE_PRIM_TRIANGLES:
feState.topologyProvokingVertex = feState.provokingVertex.triStripList;
break;
case PIPE_PRIM_QUAD_STRIP:
case PIPE_PRIM_QUADS:
if (ctx->rasterizer->flatshade_first)
feState.topologyProvokingVertex = 0;
else
feState.topologyProvokingVertex = 3;
break;
case PIPE_PRIM_LINES:
case PIPE_PRIM_LINE_LOOP:
case PIPE_PRIM_LINE_STRIP:
feState.topologyProvokingVertex = feState.provokingVertex.lineStripList;
break;
default:
feState.topologyProvokingVertex = 0;
}
feState.bEnableCutIndex = info->primitive_restart;
SwrSetFrontendState(ctx->swrContext, &feState);
if (info->indexed)
SwrDrawIndexedInstanced(ctx->swrContext,
swr_convert_prim_topology(info->mode),
info->count,
info->instance_count,
info->start,
info->index_bias,
info->start_instance);
else
SwrDrawInstanced(ctx->swrContext,
swr_convert_prim_topology(info->mode),
info->count,
info->instance_count,
info->start,
info->start_instance);
}
struct pipe_context *
swr_create_context(struct pipe_screen *p_screen, void *priv, unsigned flags)
{
struct swr_context *ctx = (struct swr_context *)
AlignedMalloc(sizeof(struct swr_context), KNOB_SIMD_BYTES);
memset(ctx, 0, sizeof(struct swr_context));
swr_screen(p_screen)->pfnSwrGetInterface(ctx->api);
ctx->swrDC.pAPI = &ctx->api;
ctx->blendJIT =
new std::unordered_map<BLEND_COMPILE_STATE, PFN_BLEND_JIT_FUNC>;
SWR_CREATECONTEXT_INFO createInfo;
memset(&createInfo, 0, sizeof(createInfo));
createInfo.privateStateSize = sizeof(swr_draw_context);
createInfo.pfnLoadTile = swr_LoadHotTile;
createInfo.pfnStoreTile = swr_StoreHotTile;
createInfo.pfnClearTile = swr_StoreHotTileClear;
createInfo.pfnUpdateStats = swr_UpdateStats;
createInfo.pfnUpdateStatsFE = swr_UpdateStatsFE;
ctx->swrContext = ctx->api.pfnSwrCreateContext(&createInfo);
ctx->api.pfnSwrInit();
if (ctx->swrContext == NULL)
goto fail;
ctx->pipe.screen = p_screen;
ctx->pipe.destroy = swr_destroy;
ctx->pipe.priv = priv;
ctx->pipe.create_surface = swr_create_surface;
ctx->pipe.surface_destroy = swr_surface_destroy;
ctx->pipe.transfer_map = swr_transfer_map;
ctx->pipe.transfer_unmap = swr_transfer_unmap;
ctx->pipe.transfer_flush_region = swr_transfer_flush_region;
ctx->pipe.buffer_subdata = u_default_buffer_subdata;
ctx->pipe.texture_subdata = u_default_texture_subdata;
ctx->pipe.clear_texture = util_clear_texture;
ctx->pipe.resource_copy_region = swr_resource_copy;
ctx->pipe.render_condition = swr_render_condition;
swr_state_init(&ctx->pipe);
swr_clear_init(&ctx->pipe);
swr_draw_init(&ctx->pipe);
swr_query_init(&ctx->pipe);
ctx->pipe.stream_uploader = u_upload_create_default(&ctx->pipe);
if (!ctx->pipe.stream_uploader)
goto fail;
ctx->pipe.const_uploader = ctx->pipe.stream_uploader;
ctx->pipe.blit = swr_blit;
ctx->blitter = util_blitter_create(&ctx->pipe);
if (!ctx->blitter)
goto fail;
swr_init_scratch_buffers(ctx);
return &ctx->pipe;
fail:
/* Should really validate the init steps and fail gracefully */
swr_destroy(&ctx->pipe);
return NULL;
}
