#include "util/u_math.h"
#include "vl/vl_decoder.h"
#include "vl/vl_video_buffer.h"
+#include "radeon/radeon_video.h"
#include "radeon/radeon_uvd.h"
#include "os/os_time.h"
static const struct debug_named_value r600_debug_options[] = {
/* features */
#if defined(R600_USE_LLVM)
- { "nollvm", DBG_NO_LLVM, "Disable the LLVM shader compiler" },
+ { "llvm", DBG_LLVM, "Enable the LLVM shader compiler" },
#endif
{ "nocpdma", DBG_NO_CP_DMA, "Disable CP DMA" },
- { "nodma", DBG_NO_ASYNC_DMA, "Disable asynchronous DMA" },
/* shader backend */
{ "nosb", DBG_NO_SB, "Disable sb backend for graphics shaders" },
* pipe_context
*/
-static void r600_flush(struct pipe_context *ctx, unsigned flags)
-{
- struct r600_context *rctx = (struct r600_context *)ctx;
- struct pipe_query *render_cond = NULL;
- unsigned render_cond_mode = 0;
- boolean render_cond_cond = FALSE;
-
- if (rctx->b.rings.gfx.cs->cdw == rctx->initial_gfx_cs_size)
- return;
-
- rctx->b.rings.gfx.flushing = true;
- /* Disable render condition. */
- if (rctx->b.current_render_cond) {
- render_cond = rctx->b.current_render_cond;
- render_cond_cond = rctx->b.current_render_cond_cond;
- render_cond_mode = rctx->b.current_render_cond_mode;
- ctx->render_condition(ctx, NULL, FALSE, 0);
- }
-
- r600_context_flush(rctx, flags);
- rctx->b.rings.gfx.flushing = false;
- r600_begin_new_cs(rctx);
-
- /* Re-enable render condition. */
- if (render_cond) {
- ctx->render_condition(ctx, render_cond, render_cond_cond, render_cond_mode);
- }
-
- rctx->initial_gfx_cs_size = rctx->b.rings.gfx.cs->cdw;
-}
-
-static void r600_flush_from_st(struct pipe_context *ctx,
- struct pipe_fence_handle **fence,
- unsigned flags)
-{
- struct r600_context *rctx = (struct r600_context *)ctx;
- unsigned fflags;
-
- fflags = flags & PIPE_FLUSH_END_OF_FRAME ? RADEON_FLUSH_END_OF_FRAME : 0;
- if (fence) {
- *fence = rctx->b.ws->cs_create_fence(rctx->b.rings.gfx.cs);
- }
- /* flush gfx & dma ring, order does not matter as only one can be live */
- if (rctx->b.rings.dma.cs) {
- rctx->b.rings.dma.flush(rctx, fflags);
- }
- rctx->b.rings.gfx.flush(rctx, fflags);
-}
-
-static void r600_flush_gfx_ring(void *ctx, unsigned flags)
-{
- r600_flush((struct pipe_context*)ctx, flags);
-}
-
-static void r600_flush_dma_ring(void *ctx, unsigned flags)
-{
- struct r600_context *rctx = (struct r600_context *)ctx;
- struct radeon_winsys_cs *cs = rctx->b.rings.dma.cs;
-
- if (!cs->cdw) {
- return;
- }
-
- rctx->b.rings.dma.flushing = true;
- rctx->b.ws->cs_flush(cs, flags, 0);
- rctx->b.rings.dma.flushing = false;
-}
-
-static void r600_flush_from_winsys(void *ctx, unsigned flags)
-{
- struct r600_context *rctx = (struct r600_context *)ctx;
-
- rctx->b.rings.gfx.flush(rctx, flags);
-}
-
-static void r600_flush_dma_from_winsys(void *ctx, unsigned flags)
-{
- struct r600_context *rctx = (struct r600_context *)ctx;
-
- rctx->b.rings.dma.flush(rctx, flags);
-}
-
static void r600_destroy_context(struct pipe_context *context)
{
struct r600_context *rctx = (struct r600_context *)context;
{
struct r600_context *rctx = CALLOC_STRUCT(r600_context);
struct r600_screen* rscreen = (struct r600_screen *)screen;
+ struct radeon_winsys *ws = rscreen->b.ws;
if (rctx == NULL)
return NULL;
rctx->b.b.screen = screen;
rctx->b.b.priv = priv;
rctx->b.b.destroy = r600_destroy_context;
- rctx->b.b.flush = r600_flush_from_st;
if (!r600_common_context_init(&rctx->b, &rscreen->b))
goto fail;
goto fail;
}
- if (rscreen->b.trace_bo) {
- rctx->b.rings.gfx.cs = rctx->b.ws->cs_create(rctx->b.ws, RING_GFX, rscreen->b.trace_bo->cs_buf);
- } else {
- rctx->b.rings.gfx.cs = rctx->b.ws->cs_create(rctx->b.ws, RING_GFX, NULL);
- }
- rctx->b.rings.gfx.flush = r600_flush_gfx_ring;
- rctx->b.ws->cs_set_flush_callback(rctx->b.rings.gfx.cs, r600_flush_from_winsys, rctx);
- rctx->b.rings.gfx.flushing = false;
-
- rctx->b.rings.dma.cs = NULL;
- if (rscreen->b.info.r600_has_dma && !(rscreen->b.debug_flags & DBG_NO_ASYNC_DMA)) {
- rctx->b.rings.dma.cs = rctx->b.ws->cs_create(rctx->b.ws, RING_DMA, NULL);
- rctx->b.rings.dma.flush = r600_flush_dma_ring;
- rctx->b.ws->cs_set_flush_callback(rctx->b.rings.dma.cs, r600_flush_dma_from_winsys, rctx);
- rctx->b.rings.dma.flushing = false;
- }
+ rctx->b.rings.gfx.cs = ws->cs_create(ws, RING_GFX,
+ r600_context_gfx_flush, rctx,
+ rscreen->b.trace_bo ?
+ rscreen->b.trace_bo->cs_buf : NULL);
+ rctx->b.rings.gfx.flush = r600_context_gfx_flush;
rctx->allocator_fetch_shader = u_suballocator_create(&rctx->b.b, 64 * 1024, 256,
- 0, PIPE_USAGE_STATIC, FALSE);
+ 0, PIPE_USAGE_DEFAULT, FALSE);
if (!rctx->allocator_fetch_shader)
goto fail;
case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
case PIPE_CAP_USER_INDEX_BUFFERS:
case PIPE_CAP_USER_CONSTANT_BUFFERS:
- case PIPE_CAP_COMPUTE:
case PIPE_CAP_START_INSTANCE:
case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS:
case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
case PIPE_CAP_TEXTURE_MULTISAMPLE:
+ case PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT:
+ case PIPE_CAP_TGSI_VS_WINDOW_SPACE_POSITION:
return 1;
+ case PIPE_CAP_COMPUTE:
+ return rscreen->b.chip_class > R700;
+
case PIPE_CAP_TGSI_TEXCOORD:
return 0;
+ case PIPE_CAP_FAKE_SW_MSAA:
+ return 0;
+
case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE:
return MIN2(rscreen->b.info.vram_size, 0xFFFFFFFF);
return 1;
case PIPE_CAP_GLSL_FEATURE_LEVEL:
+ if (family >= CHIP_CEDAR)
+ return 330;
+ /* pre-evergreen geom shaders need newer kernel */
+ if (rscreen->b.info.drm_minor >= 37)
+ return 330;
return 140;
/* Supported except the original R600. */
/* Supported on Evergreen. */
case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
case PIPE_CAP_CUBE_MAP_ARRAY:
+ case PIPE_CAP_TGSI_VS_LAYER_VIEWPORT:
+ case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS:
return family >= CHIP_CEDAR ? 1 : 0;
/* Unsupported features. */
case PIPE_CAP_FRAGMENT_COLOR_CLAMPED:
case PIPE_CAP_VERTEX_COLOR_CLAMPED:
case PIPE_CAP_USER_VERTEX_BUFFERS:
- case PIPE_CAP_TGSI_VS_LAYER:
+ case PIPE_CAP_TEXTURE_GATHER_SM5:
+ case PIPE_CAP_TEXTURE_QUERY_LOD:
+ case PIPE_CAP_SAMPLE_SHADING:
+ case PIPE_CAP_TEXTURE_GATHER_OFFSETS:
+ case PIPE_CAP_DRAW_INDIRECT:
return 0;
/* Stream output. */
case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS:
return 32*4;
+ /* Geometry shader output. */
+ case PIPE_CAP_MAX_GEOMETRY_OUTPUT_VERTICES:
+ return 1024;
+ case PIPE_CAP_MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS:
+ return 16384;
+ case PIPE_CAP_MAX_VERTEX_STREAMS:
+ return 1;
+
/* Texturing. */
case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
- case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
if (family >= CHIP_CEDAR)
return 15;
else
return 14;
+ case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
+ /* textures support 8192, but layered rendering supports 2048 */
+ return 12;
case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS:
- return rscreen->b.info.drm_minor >= 9 ?
- (family >= CHIP_CEDAR ? 16384 : 8192) : 0;
- case PIPE_CAP_MAX_COMBINED_SAMPLERS:
- return 32;
+ /* textures support 8192, but layered rendering supports 2048 */
+ return rscreen->b.info.drm_minor >= 9 ? 2048 : 0;
/* Render targets. */
case PIPE_CAP_MAX_RENDER_TARGETS:
return 8;
case PIPE_CAP_MAX_VIEWPORTS:
- return 1;
+ return 16;
/* Timer queries, present when the clock frequency is non zero. */
case PIPE_CAP_QUERY_TIME_ELAPSED:
return rscreen->b.info.drm_minor >= 20 &&
rscreen->b.info.r600_clock_crystal_freq != 0;
+ case PIPE_CAP_MIN_TEXTURE_GATHER_OFFSET:
case PIPE_CAP_MIN_TEXEL_OFFSET:
return -8;
+ case PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET:
case PIPE_CAP_MAX_TEXEL_OFFSET:
return 7;
static int r600_get_shader_param(struct pipe_screen* pscreen, unsigned shader, enum pipe_shader_cap param)
{
+ struct r600_screen *rscreen = (struct r600_screen *)pscreen;
+
switch(shader)
{
case PIPE_SHADER_FRAGMENT:
case PIPE_SHADER_VERTEX:
- case PIPE_SHADER_COMPUTE:
+ case PIPE_SHADER_COMPUTE:
break;
case PIPE_SHADER_GEOMETRY:
- /* XXX: support and enable geometry programs */
+ if (rscreen->b.family >= CHIP_CEDAR)
+ break;
+ /* pre-evergreen geom shaders need newer kernel */
+ if (rscreen->b.info.drm_minor >= 37)
+ break;
return 0;
default:
/* XXX: support tessellation on Evergreen */
case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
return 32;
case PIPE_SHADER_CAP_MAX_INPUTS:
- return 32;
+ return shader == PIPE_SHADER_VERTEX ? 16 : 32;
case PIPE_SHADER_CAP_MAX_TEMPS:
return 256; /* Max native temporaries. */
case PIPE_SHADER_CAP_MAX_ADDRS:
} else {
return PIPE_SHADER_IR_TGSI;
}
- }
- return 0;
-}
-
-const char * r600_llvm_gpu_string(enum radeon_family family)
-{
- const char * gpu_family;
-
- switch (family) {
- case CHIP_R600:
- case CHIP_RV630:
- case CHIP_RV635:
- case CHIP_RV670:
- gpu_family = "r600";
- break;
- case CHIP_RV610:
- case CHIP_RV620:
- case CHIP_RS780:
- case CHIP_RS880:
- gpu_family = "rs880";
- break;
- case CHIP_RV710:
- gpu_family = "rv710";
- break;
- case CHIP_RV730:
- gpu_family = "rv730";
- break;
- case CHIP_RV740:
- case CHIP_RV770:
- gpu_family = "rv770";
- break;
- case CHIP_PALM:
- case CHIP_CEDAR:
- gpu_family = "cedar";
- break;
- case CHIP_SUMO:
- case CHIP_SUMO2:
- gpu_family = "sumo";
- break;
- case CHIP_REDWOOD:
- gpu_family = "redwood";
- break;
- case CHIP_JUNIPER:
- gpu_family = "juniper";
- break;
- case CHIP_HEMLOCK:
- case CHIP_CYPRESS:
- gpu_family = "cypress";
- break;
- case CHIP_BARTS:
- gpu_family = "barts";
- break;
- case CHIP_TURKS:
- gpu_family = "turks";
- break;
- case CHIP_CAICOS:
- gpu_family = "caicos";
- break;
- case CHIP_CAYMAN:
- case CHIP_ARUBA:
- gpu_family = "cayman";
- break;
- default:
- gpu_family = "";
- fprintf(stderr, "Chip not supported by r600 llvm "
- "backend, please file a bug at " PACKAGE_BUGREPORT "\n");
- break;
- }
- return gpu_family;
-}
-
-
-static int r600_get_compute_param(struct pipe_screen *screen,
- enum pipe_compute_cap param,
- void *ret)
-{
- struct r600_screen *rscreen = (struct r600_screen *)screen;
- //TODO: select these params by asic
- switch (param) {
- case PIPE_COMPUTE_CAP_IR_TARGET: {
- const char *gpu = r600_llvm_gpu_string(rscreen->b.family);
- if (ret) {
- sprintf(ret, "%s-r600--", gpu);
- }
- return (8 + strlen(gpu)) * sizeof(char);
- }
- case PIPE_COMPUTE_CAP_GRID_DIMENSION:
- if (ret) {
- uint64_t * grid_dimension = ret;
- grid_dimension[0] = 3;
- }
- return 1 * sizeof(uint64_t);
-
- case PIPE_COMPUTE_CAP_MAX_GRID_SIZE:
- if (ret) {
- uint64_t * grid_size = ret;
- grid_size[0] = 65535;
- grid_size[1] = 65535;
- grid_size[2] = 1;
- }
- return 3 * sizeof(uint64_t) ;
-
- case PIPE_COMPUTE_CAP_MAX_BLOCK_SIZE:
- if (ret) {
- uint64_t * block_size = ret;
- block_size[0] = 256;
- block_size[1] = 256;
- block_size[2] = 256;
- }
- return 3 * sizeof(uint64_t);
-
- case PIPE_COMPUTE_CAP_MAX_THREADS_PER_BLOCK:
- if (ret) {
- uint64_t * max_threads_per_block = ret;
- *max_threads_per_block = 256;
- }
- return sizeof(uint64_t);
-
- case PIPE_COMPUTE_CAP_MAX_GLOBAL_SIZE:
- if (ret) {
- uint64_t * max_global_size = ret;
- /* XXX: This is what the proprietary driver reports, we
- * may want to use a different value. */
- *max_global_size = 201326592;
- }
- return sizeof(uint64_t);
-
- case PIPE_COMPUTE_CAP_MAX_INPUT_SIZE:
- if (ret) {
- uint64_t * max_input_size = ret;
- *max_input_size = 1024;
- }
- return sizeof(uint64_t);
-
- case PIPE_COMPUTE_CAP_MAX_LOCAL_SIZE:
- if (ret) {
- uint64_t * max_local_size = ret;
- /* XXX: This is what the proprietary driver reports, we
- * may want to use a different value. */
- *max_local_size = 32768;
- }
- return sizeof(uint64_t);
-
- case PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE:
- if (ret) {
- uint64_t max_global_size;
- uint64_t * max_mem_alloc_size = ret;
- r600_get_compute_param(screen,
- PIPE_COMPUTE_CAP_MAX_GLOBAL_SIZE,
- &max_global_size);
- /* OpenCL requres this value be at least
- * max(MAX_GLOBAL_SIZE / 4, 128 * 1024 *1024)
- * I'm really not sure what value to report here, but
- * MAX_GLOBAL_SIZE / 4 seems resonable.
- */
- *max_mem_alloc_size = max_global_size / 4;
- }
- return sizeof(uint64_t);
-
- default:
- fprintf(stderr, "unknown PIPE_COMPUTE_CAP %d\n", param);
+ case PIPE_SHADER_CAP_DOUBLES:
return 0;
}
+ return 0;
}
static void r600_destroy_screen(struct pipe_screen* pscreen)
if (rscreen == NULL)
return;
- if (!radeon_winsys_unref(rscreen->b.ws))
+ if (!rscreen->b.ws->unref(rscreen->b.ws))
return;
- r600_common_screen_cleanup(&rscreen->b);
-
if (rscreen->global_pool) {
compute_memory_pool_delete(rscreen->global_pool);
}
- if (rscreen->b.trace_bo) {
- rscreen->b.ws->buffer_unmap(rscreen->b.trace_bo->cs_buf);
- pipe_resource_reference((struct pipe_resource**)&rscreen->b.trace_bo, NULL);
- }
-
- rscreen->b.ws->destroy(rscreen->b.ws);
- FREE(rscreen);
+ r600_destroy_common_screen(&rscreen->b);
}
static struct pipe_resource *r600_resource_create(struct pipe_screen *screen,
return NULL;
}
- ws->query_info(ws, &rscreen->b.info);
-
/* Set functions first. */
rscreen->b.b.context_create = r600_create_context;
rscreen->b.b.destroy = r600_destroy_screen;
rscreen->b.b.get_param = r600_get_param;
rscreen->b.b.get_shader_param = r600_get_shader_param;
- rscreen->b.b.get_compute_param = r600_get_compute_param;
- if (rscreen->b.info.chip_class >= EVERGREEN) {
- rscreen->b.b.is_format_supported = evergreen_is_format_supported;
- } else {
- rscreen->b.b.is_format_supported = r600_is_format_supported;
- }
rscreen->b.b.resource_create = r600_resource_create;
if (!r600_common_screen_init(&rscreen->b, ws)) {
return NULL;
}
+ if (rscreen->b.info.chip_class >= EVERGREEN) {
+ rscreen->b.b.is_format_supported = evergreen_is_format_supported;
+ } else {
+ rscreen->b.b.is_format_supported = r600_is_format_supported;
+ }
+
rscreen->b.debug_flags |= debug_get_flags_option("R600_DEBUG", r600_debug_options, 0);
if (debug_get_bool_option("R600_DEBUG_COMPUTE", FALSE))
rscreen->b.debug_flags |= DBG_COMPUTE;
if (debug_get_bool_option("R600_DUMP_SHADERS", FALSE))
rscreen->b.debug_flags |= DBG_FS | DBG_VS | DBG_GS | DBG_PS | DBG_CS;
- if (!debug_get_bool_option("R600_HYPERZ", TRUE))
- rscreen->b.debug_flags |= DBG_NO_HYPERZ;
- if (!debug_get_bool_option("R600_LLVM", TRUE))
- rscreen->b.debug_flags |= DBG_NO_LLVM;
+ if (debug_get_bool_option("R600_HYPERZ", FALSE))
+ rscreen->b.debug_flags |= DBG_HYPERZ;
+ if (debug_get_bool_option("R600_LLVM", FALSE))
+ rscreen->b.debug_flags |= DBG_LLVM;
if (rscreen->b.family == CHIP_UNKNOWN) {
fprintf(stderr, "r600: Unknown chipset 0x%04X\n", rscreen->b.info.pci_id);
rscreen->global_pool = compute_memory_pool_new(rscreen);
- rscreen->b.cs_count = 0;
- if (rscreen->b.info.drm_minor >= 28 && (rscreen->b.debug_flags & DBG_TRACE_CS)) {
- rscreen->b.trace_bo = (struct r600_resource*)pipe_buffer_create(&rscreen->b.b,
- PIPE_BIND_CUSTOM,
- PIPE_USAGE_STAGING,
- 4096);
- if (rscreen->b.trace_bo) {
- rscreen->b.trace_ptr = rscreen->b.ws->buffer_map(rscreen->b.trace_bo->cs_buf, NULL,
- PIPE_TRANSFER_UNSYNCHRONIZED);
- }
- }
-
/* Create the auxiliary context. This must be done last. */
rscreen->b.aux_context = rscreen->b.b.context_create(&rscreen->b.b, NULL);
templ.array_size = 1;
templ.target = PIPE_TEXTURE_2D;
templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
- templ.usage = PIPE_USAGE_STATIC;
+ templ.usage = PIPE_USAGE_DEFAULT;
struct r600_resource *res = r600_resource(rscreen->screen.resource_create(&rscreen->screen, &templ));
unsigned char *map = ws->buffer_map(res->cs_buf, NULL, PIPE_TRANSFER_WRITE);
OSDN Git Service
