#include "util/u_memory.h"
#include "util/u_inlines.h"
-#include "util/u_format.h"
+#include "util/format/u_format.h"
#include "util/u_atomic.h"
#include "util/u_upload_mgr.h"
#include "util/u_transfer.h"
#include "api.h"
#include "backend.h"
+#include "knobs.h"
static struct pipe_surface *
swr_create_surface(struct pipe_context *pipe,
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++)
- spr->swr.pBaseAddress[zbase + 4 * x + 3] =
- spr->secondary.pBaseAddress[sbase + 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++)
- spr->swr.pBaseAddress[zbase + 8 * x + 4] =
- spr->secondary.pBaseAddress[sbase + 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;
*transfer = pt;
- return spr->swr.pBaseAddress + offset + spr->mip_offsets[level];
+ return (void*)(spr->swr.xpBaseAddress + offset + spr->mip_offsets[level]);
}
static void
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++)
- spr->secondary.pBaseAddress[sbase + x] =
- spr->swr.pBaseAddress[zbase + 4 * x + 3];
+ ((uint8_t*)(spr->secondary.xpBaseAddress))[sbase + x] =
+ ((uint8_t*)(spr->swr.xpBaseAddress))[zbase + 4 * x + 3];
} else if (spr->base.format == PIPE_FORMAT_Z32_FLOAT_S8X24_UINT) {
for (int x = box.x; x < box.x + box.width; x++)
- spr->secondary.pBaseAddress[sbase + x] =
- spr->swr.pBaseAddress[zbase + 8 * x + 4];
+ ((uint8_t*)(spr->secondary.xpBaseAddress))[sbase + x] =
+ ((uint8_t*)(spr->swr.xpBaseAddress))[zbase + 8 * x + 4];
}
zbase += spr->swr.pitch;
sbase += spr->secondary.pitch;
}
if (ctx->active_queries) {
- SwrEnableStatsFE(ctx->swrContext, FALSE);
- SwrEnableStatsBE(ctx->swrContext, FALSE);
+ ctx->api.pfnSwrEnableStatsFE(ctx->swrContext, FALSE);
+ ctx->api.pfnSwrEnableStatsBE(ctx->swrContext, FALSE);
}
util_blitter_save_vertex_buffer_slot(ctx->blitter, ctx->vertex_buffer);
util_blitter_save_vertex_elements(ctx->blitter, (void *)ctx->velems);
util_blitter_save_vertex_shader(ctx->blitter, (void *)ctx->vs);
util_blitter_save_geometry_shader(ctx->blitter, (void*)ctx->gs);
+ util_blitter_save_tessctrl_shader(ctx->blitter, (void*)ctx->tcs);
+ util_blitter_save_tesseval_shader(ctx->blitter, (void*)ctx->tes);
util_blitter_save_so_targets(
ctx->blitter,
ctx->num_so_targets,
(struct pipe_stream_output_target **)ctx->so_targets);
util_blitter_save_rasterizer(ctx->blitter, (void *)ctx->rasterizer);
- util_blitter_save_viewport(ctx->blitter, &ctx->viewport);
- util_blitter_save_scissor(ctx->blitter, &ctx->scissor);
+ util_blitter_save_viewport(ctx->blitter, &ctx->viewports[0]);
+ util_blitter_save_scissor(ctx->blitter, &ctx->scissors[0]);
util_blitter_save_fragment_shader(ctx->blitter, ctx->fs);
util_blitter_save_blend(ctx->blitter, (void *)ctx->blend);
util_blitter_save_depth_stencil_alpha(ctx->blitter,
util_blitter_blit(ctx->blitter, &info);
if (ctx->active_queries) {
- SwrEnableStatsFE(ctx->swrContext, TRUE);
- SwrEnableStatsBE(ctx->swrContext, TRUE);
+ ctx->api.pfnSwrEnableStatsFE(ctx->swrContext, TRUE);
+ ctx->api.pfnSwrEnableStatsBE(ctx->swrContext, TRUE);
}
}
util_blitter_destroy(ctx->blitter);
for (unsigned i = 0; i < PIPE_MAX_COLOR_BUFS; i++) {
- pipe_surface_reference(&ctx->framebuffer.cbufs[i], NULL);
+ 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);
+ }
}
- pipe_surface_reference(&ctx->framebuffer.zsbuf, 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);
/* Idle core after destroying buffer resources, but before deleting
* context. Destroying resources has potentially called StoreTiles.*/
- SwrWaitForIdle(ctx->swrContext);
+ ctx->api.pfnSwrWaitForIdle(ctx->swrContext);
if (ctx->swrContext)
- SwrDestroyContext(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)
static void
swr_render_condition(struct pipe_context *pipe,
struct pipe_query *query,
- boolean condition,
+ bool condition,
enum pipe_render_cond_flag mode)
{
struct swr_context *ctx = swr_context(pipe);
if (!pDC)
return;
- struct swr_query_result *pqr = (struct swr_query_result *)pDC->pStats;
+ struct swr_query_result *pqr = pDC->pStats;
SWR_STATS *pSwrStats = &pqr->core;
if (!pDC)
return;
- struct swr_query_result *pqr = (struct swr_query_result *)pDC->pStats;
+ struct swr_query_result *pqr = pDC->pStats;
SWR_STATS_FE *pSwrStats = &pqr->coreFE;
p_atomic_add(&pSwrStats->IaVertices, pStats->IaVertices);
}
}
+static void
+swr_UpdateStreamOut(HANDLE hPrivateContext, uint64_t numPrims)
+{
+ swr_draw_context *pDC = (swr_draw_context*)hPrivateContext;
+
+ if (!pDC)
+ return;
+
+ if (pDC->soPrims)
+ *pDC->soPrims += numPrims;
+}
+
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));
+ memset((void*)ctx, 0, sizeof(struct swr_context));
+
+ swr_screen(p_screen)->pfnSwrGetInterface(ctx->api);
+ swr_screen(p_screen)->pfnSwrGetTileInterface(ctx->tileApi);
+ ctx->swrDC.pAPI = &ctx->api;
+ ctx->swrDC.pTileAPI = &ctx->tileApi;
ctx->blendJIT =
new std::unordered_map<BLEND_COMPILE_STATE, PFN_BLEND_JIT_FUNC>;
- SWR_CREATECONTEXT_INFO createInfo;
- memset(&createInfo, 0, sizeof(createInfo));
+ ctx->max_draws_in_flight = KNOB_MAX_DRAWS_IN_FLIGHT;
+
+ SWR_CREATECONTEXT_INFO createInfo {0};
+
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 = SwrCreateContext(&createInfo);
+ createInfo.pfnUpdateStreamOut = swr_UpdateStreamOut;
+ createInfo.pfnMakeGfxPtr = swr_MakeGfxPtr;
+
+ 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);
- SwrInit();
+ ctx->api.pfnSwrInit();
if (ctx->swrContext == NULL)
goto fail;
projects / mesa.git / blobdiff