#include "r600_pipe_common.h"
#include "r600_cs.h"
#include "tgsi/tgsi_parse.h"
+#include "util/list.h"
#include "util/u_draw_quad.h"
#include "util/u_memory.h"
#include "util/u_format_s3tc.h"
struct pipe_fence_handle *sdma;
};
+/*
+ * shader binary helpers.
+ */
+void radeon_shader_binary_init(struct radeon_shader_binary *b)
+{
+ memset(b, 0, sizeof(*b));
+}
+
+void radeon_shader_binary_clean(struct radeon_shader_binary *b)
+{
+ if (!b)
+ return;
+ FREE(b->code);
+ FREE(b->config);
+ FREE(b->rodata);
+ FREE(b->global_symbol_offsets);
+ FREE(b->relocs);
+ FREE(b->disasm_string);
+}
+
/*
* pipe_context
*/
/* Upload vertices. The hw rectangle has only 3 vertices,
* I guess the 4th one is derived from the first 3.
* The vertex specification should match u_blitter's vertex element state. */
- u_upload_alloc(rctx->uploader, 0, sizeof(float) * 24, &offset, &buf, (void**)&vb);
+ u_upload_alloc(rctx->uploader, 0, sizeof(float) * 24, 256, &offset, &buf, (void**)&vb);
if (!buf)
return;
void r600_need_dma_space(struct r600_common_context *ctx, unsigned num_dw)
{
+ /* Flush the GFX IB if it's not empty. */
+ if (ctx->gfx.cs->cdw > ctx->initial_gfx_cs_size)
+ ctx->gfx.flush(ctx, RADEON_FLUSH_ASYNC, NULL);
+
/* Flush if there's not enough space. */
- if ((num_dw + ctx->rings.dma.cs->cdw) > ctx->rings.dma.cs->max_dw) {
- ctx->rings.dma.flush(ctx, RADEON_FLUSH_ASYNC, NULL);
- assert((num_dw + ctx->rings.dma.cs->cdw) <= ctx->rings.dma.cs->max_dw);
+ if ((num_dw + ctx->dma.cs->cdw) > ctx->dma.cs->max_dw) {
+ ctx->dma.flush(ctx, RADEON_FLUSH_ASYNC, NULL);
+ assert((num_dw + ctx->dma.cs->cdw) <= ctx->dma.cs->max_dw);
}
}
void r600_preflush_suspend_features(struct r600_common_context *ctx)
{
- /* Disable render condition. */
- ctx->saved_render_cond = NULL;
- ctx->saved_render_cond_cond = FALSE;
- ctx->saved_render_cond_mode = 0;
- if (ctx->current_render_cond) {
- ctx->saved_render_cond = ctx->current_render_cond;
- ctx->saved_render_cond_cond = ctx->current_render_cond_cond;
- ctx->saved_render_cond_mode = ctx->current_render_cond_mode;
- ctx->b.render_condition(&ctx->b, NULL, FALSE, 0);
- }
-
/* suspend queries */
- ctx->queries_suspended_for_flush = false;
if (ctx->num_cs_dw_nontimer_queries_suspend) {
+ /* Since non-timer queries are suspended during blits,
+ * we have to guard against double-suspends. */
r600_suspend_nontimer_queries(ctx);
- r600_suspend_timer_queries(ctx);
- ctx->queries_suspended_for_flush = true;
+ ctx->nontimer_queries_suspended_by_flush = true;
}
+ if (!LIST_IS_EMPTY(&ctx->active_timer_queries))
+ r600_suspend_timer_queries(ctx);
ctx->streamout.suspended = false;
if (ctx->streamout.begin_emitted) {
}
/* resume queries */
- if (ctx->queries_suspended_for_flush) {
- r600_resume_nontimer_queries(ctx);
+ if (!LIST_IS_EMPTY(&ctx->active_timer_queries))
r600_resume_timer_queries(ctx);
- }
-
- /* Re-enable render condition. */
- if (ctx->saved_render_cond) {
- ctx->b.render_condition(&ctx->b, ctx->saved_render_cond,
- ctx->saved_render_cond_cond,
- ctx->saved_render_cond_mode);
+ if (ctx->nontimer_queries_suspended_by_flush) {
+ ctx->nontimer_queries_suspended_by_flush = false;
+ r600_resume_nontimer_queries(ctx);
}
}
if (flags & PIPE_FLUSH_END_OF_FRAME)
rflags |= RADEON_FLUSH_END_OF_FRAME;
- if (rctx->rings.dma.cs) {
- rctx->rings.dma.flush(rctx, rflags, fence ? &sdma_fence : NULL);
+ if (rctx->dma.cs) {
+ rctx->dma.flush(rctx, rflags, fence ? &sdma_fence : NULL);
}
- rctx->rings.gfx.flush(rctx, rflags, fence ? &gfx_fence : NULL);
+ rctx->gfx.flush(rctx, rflags, fence ? &gfx_fence : NULL);
/* Both engines can signal out of order, so we need to keep both fences. */
if (gfx_fence || sdma_fence) {
struct pipe_fence_handle **fence)
{
struct r600_common_context *rctx = (struct r600_common_context *)ctx;
- struct radeon_winsys_cs *cs = rctx->rings.dma.cs;
+ struct radeon_winsys_cs *cs = rctx->dma.cs;
- if (!cs->cdw)
- goto done;
-
- rctx->rings.dma.flushing = true;
- rctx->ws->cs_flush(cs, flags, &rctx->last_sdma_fence, 0);
- rctx->rings.dma.flushing = false;
-done:
+ if (cs->cdw)
+ rctx->ws->cs_flush(cs, flags, &rctx->last_sdma_fence, 0);
if (fence)
rctx->ws->fence_reference(fence, rctx->last_sdma_fence);
}
return PIPE_UNKNOWN_CONTEXT_RESET;
}
+static void r600_set_debug_callback(struct pipe_context *ctx,
+ const struct pipe_debug_callback *cb)
+{
+ struct r600_common_context *rctx = (struct r600_common_context *)ctx;
+
+ if (cb)
+ rctx->debug = *cb;
+ else
+ memset(&rctx->debug, 0, sizeof(rctx->debug));
+}
+
bool r600_common_context_init(struct r600_common_context *rctx,
struct r600_common_screen *rscreen)
{
rctx->family = rscreen->family;
rctx->chip_class = rscreen->chip_class;
- if (rscreen->family == CHIP_HAWAII)
- rctx->max_db = 16;
+ if (rscreen->chip_class >= CIK)
+ rctx->max_db = MAX2(8, rscreen->info.num_render_backends);
else if (rscreen->chip_class >= EVERGREEN)
rctx->max_db = 8;
else
rctx->max_db = 4;
+ rctx->b.invalidate_resource = r600_invalidate_resource;
rctx->b.transfer_map = u_transfer_map_vtbl;
rctx->b.transfer_flush_region = u_transfer_flush_region_vtbl;
rctx->b.transfer_unmap = u_transfer_unmap_vtbl;
rctx->b.transfer_inline_write = u_default_transfer_inline_write;
rctx->b.memory_barrier = r600_memory_barrier;
rctx->b.flush = r600_flush_from_st;
+ rctx->b.set_debug_callback = r600_set_debug_callback;
if (rscreen->info.drm_major == 2 && rscreen->info.drm_minor >= 43) {
rctx->b.get_device_reset_status = r600_get_reset_status;
if (!rctx->allocator_so_filled_size)
return false;
- rctx->uploader = u_upload_create(&rctx->b, 1024 * 1024, 256,
+ rctx->uploader = u_upload_create(&rctx->b, 1024 * 1024,
PIPE_BIND_INDEX_BUFFER |
- PIPE_BIND_CONSTANT_BUFFER);
+ PIPE_BIND_CONSTANT_BUFFER, PIPE_USAGE_STREAM);
if (!rctx->uploader)
return false;
if (!rctx->ctx)
return false;
- if (rscreen->info.r600_has_dma && !(rscreen->debug_flags & DBG_NO_ASYNC_DMA)) {
- rctx->rings.dma.cs = rctx->ws->cs_create(rctx->ctx, RING_DMA,
- r600_flush_dma_ring,
- rctx, NULL);
- rctx->rings.dma.flush = r600_flush_dma_ring;
+ if (rscreen->info.has_sdma && !(rscreen->debug_flags & DBG_NO_ASYNC_DMA)) {
+ rctx->dma.cs = rctx->ws->cs_create(rctx->ctx, RING_DMA,
+ r600_flush_dma_ring,
+ rctx, NULL);
+ rctx->dma.flush = r600_flush_dma_ring;
}
return true;
void r600_common_context_cleanup(struct r600_common_context *rctx)
{
- if (rctx->rings.gfx.cs)
- rctx->ws->cs_destroy(rctx->rings.gfx.cs);
- if (rctx->rings.dma.cs)
- rctx->ws->cs_destroy(rctx->rings.dma.cs);
+ if (rctx->gfx.cs)
+ rctx->ws->cs_destroy(rctx->gfx.cs);
+ if (rctx->dma.cs)
+ rctx->ws->cs_destroy(rctx->dma.cs);
if (rctx->ctx)
rctx->ws->ctx_destroy(rctx->ctx);
struct r600_common_context *rctx = (struct r600_common_context *)ctx;
struct r600_resource *rr = (struct r600_resource *)r;
- if (r == NULL) {
+ if (!r) {
return;
}
static const struct debug_named_value common_debug_options[] = {
/* logging */
{ "tex", DBG_TEX, "Print texture info" },
- { "texmip", DBG_TEXMIP, "Print texture info (mipmapped only)" },
{ "compute", DBG_COMPUTE, "Print compute info" },
{ "vm", DBG_VM, "Print virtual addresses when creating resources" },
{ "trace_cs", DBG_TRACE_CS, "Trace cs and write rlockup_<csid>.c file with faulty cs" },
{ "noir", DBG_NO_IR, "Don't print the LLVM IR"},
{ "notgsi", DBG_NO_TGSI, "Don't print the TGSI"},
{ "noasm", DBG_NO_ASM, "Don't print disassembled shaders"},
+ { "preoptir", DBG_PREOPT_IR, "Print the LLVM IR before initial optimizations" },
/* features */
{ "nodma", DBG_NO_ASYNC_DMA, "Disable asynchronous DMA" },
{ "check_vm", DBG_CHECK_VM, "Check VM faults and dump debug info." },
{ "nodcc", DBG_NO_DCC, "Disable DCC." },
{ "nodccclear", DBG_NO_DCC_CLEAR, "Disable DCC fast clear." },
+ { "norbplus", DBG_NO_RB_PLUS, "Disable RB+ on Stoney." },
+ { "sisched", DBG_SI_SCHED, "Enable LLVM SI Machine Instruction Scheduler." },
DEBUG_NAMED_VALUE_END /* must be last */
};
case CHIP_TONGA: return "tonga";
case CHIP_ICELAND: return "iceland";
case CHIP_CARRIZO: return "carrizo";
- case CHIP_FIJI: return "fiji";
#if HAVE_LLVM <= 0x0307
+ case CHIP_FIJI: return "tonga";
case CHIP_STONEY: return "carrizo";
#else
+ case CHIP_FIJI: return "fiji";
case CHIP_STONEY: return "stoney";
#endif
default: return "";
case PIPE_COMPUTE_CAP_MAX_CLOCK_FREQUENCY:
if (ret) {
uint32_t *max_clock_frequency = ret;
- *max_clock_frequency = rscreen->info.max_sclk;
+ *max_clock_frequency = rscreen->info.max_shader_clock;
}
return sizeof(uint32_t);
case PIPE_COMPUTE_CAP_MAX_COMPUTE_UNITS:
if (ret) {
uint32_t *max_compute_units = ret;
- *max_compute_units = rscreen->info.max_compute_units;
+ *max_compute_units = rscreen->info.num_good_compute_units;
}
return sizeof(uint32_t);
struct r600_common_screen *rscreen = (struct r600_common_screen*)screen;
return 1000000 * rscreen->ws->query_value(rscreen->ws, RADEON_TIMESTAMP) /
- rscreen->info.r600_clock_crystal_freq;
-}
-
-static int r600_get_driver_query_info(struct pipe_screen *screen,
- unsigned index,
- struct pipe_driver_query_info *info)
-{
- struct r600_common_screen *rscreen = (struct r600_common_screen*)screen;
- struct pipe_driver_query_info list[] = {
- {"num-compilations", R600_QUERY_NUM_COMPILATIONS, {0}, PIPE_DRIVER_QUERY_TYPE_UINT64,
- PIPE_DRIVER_QUERY_RESULT_TYPE_CUMULATIVE},
- {"num-shaders-created", R600_QUERY_NUM_SHADERS_CREATED, {0}, PIPE_DRIVER_QUERY_TYPE_UINT64,
- PIPE_DRIVER_QUERY_RESULT_TYPE_CUMULATIVE},
- {"draw-calls", R600_QUERY_DRAW_CALLS, {0}},
- {"requested-VRAM", R600_QUERY_REQUESTED_VRAM, {rscreen->info.vram_size}, PIPE_DRIVER_QUERY_TYPE_BYTES},
- {"requested-GTT", R600_QUERY_REQUESTED_GTT, {rscreen->info.gart_size}, PIPE_DRIVER_QUERY_TYPE_BYTES},
- {"buffer-wait-time", R600_QUERY_BUFFER_WAIT_TIME, {0}, PIPE_DRIVER_QUERY_TYPE_MICROSECONDS,
- PIPE_DRIVER_QUERY_RESULT_TYPE_CUMULATIVE},
- {"num-cs-flushes", R600_QUERY_NUM_CS_FLUSHES, {0}},
- {"num-bytes-moved", R600_QUERY_NUM_BYTES_MOVED, {0}, PIPE_DRIVER_QUERY_TYPE_BYTES,
- PIPE_DRIVER_QUERY_RESULT_TYPE_CUMULATIVE},
- {"VRAM-usage", R600_QUERY_VRAM_USAGE, {rscreen->info.vram_size}, PIPE_DRIVER_QUERY_TYPE_BYTES},
- {"GTT-usage", R600_QUERY_GTT_USAGE, {rscreen->info.gart_size}, PIPE_DRIVER_QUERY_TYPE_BYTES},
- {"GPU-load", R600_QUERY_GPU_LOAD, {100}},
- {"temperature", R600_QUERY_GPU_TEMPERATURE, {125}},
- {"shader-clock", R600_QUERY_CURRENT_GPU_SCLK, {0}, PIPE_DRIVER_QUERY_TYPE_HZ},
- {"memory-clock", R600_QUERY_CURRENT_GPU_MCLK, {0}, PIPE_DRIVER_QUERY_TYPE_HZ},
- };
- unsigned num_queries;
-
- if (rscreen->info.drm_major == 2 && rscreen->info.drm_minor >= 42)
- num_queries = Elements(list);
- else if (rscreen->info.drm_major == 3)
- num_queries = Elements(list) - 3;
- else
- num_queries = Elements(list) - 4;
-
- if (!info)
- return num_queries;
-
- if (index >= num_queries)
- return 0;
-
- *info = list[index];
- return 1;
+ rscreen->info.clock_crystal_freq;
}
static void r600_fence_reference(struct pipe_screen *screen,
static bool r600_interpret_tiling(struct r600_common_screen *rscreen,
uint32_t tiling_config)
{
- switch ((tiling_config & 0xe) >> 1) {
- case 0:
- rscreen->tiling_info.num_channels = 1;
- break;
- case 1:
- rscreen->tiling_info.num_channels = 2;
- break;
- case 2:
- rscreen->tiling_info.num_channels = 4;
- break;
- case 3:
- rscreen->tiling_info.num_channels = 8;
- break;
- default:
- return false;
- }
-
switch ((tiling_config & 0x30) >> 4) {
case 0:
rscreen->tiling_info.num_banks = 4;
static bool evergreen_interpret_tiling(struct r600_common_screen *rscreen,
uint32_t tiling_config)
{
- switch (tiling_config & 0xf) {
- case 0:
- rscreen->tiling_info.num_channels = 1;
- break;
- case 1:
- rscreen->tiling_info.num_channels = 2;
- break;
- case 2:
- rscreen->tiling_info.num_channels = 4;
- break;
- case 3:
- rscreen->tiling_info.num_channels = 8;
- break;
- default:
- return false;
- }
-
switch ((tiling_config & 0xf0) >> 4) {
case 0:
rscreen->tiling_info.num_banks = 4;
rscreen->b.get_device_vendor = r600_get_device_vendor;
rscreen->b.get_compute_param = r600_get_compute_param;
rscreen->b.get_paramf = r600_get_paramf;
- rscreen->b.get_driver_query_info = r600_get_driver_query_info;
rscreen->b.get_timestamp = r600_get_timestamp;
rscreen->b.fence_finish = r600_fence_finish;
rscreen->b.fence_reference = r600_fence_reference;
}
r600_init_screen_texture_functions(rscreen);
+ r600_init_screen_query_functions(rscreen);
rscreen->ws = ws;
rscreen->family = rscreen->info.family;
PIPE_USAGE_STAGING,
4096);
if (rscreen->trace_bo) {
- rscreen->trace_ptr = rscreen->ws->buffer_map(rscreen->trace_bo->cs_buf, NULL,
+ rscreen->trace_ptr = rscreen->ws->buffer_map(rscreen->trace_bo->buf, NULL,
PIPE_TRANSFER_UNSYNCHRONIZED);
}
}
if (rscreen->debug_flags & DBG_INFO) {
printf("pci_id = 0x%x\n", rscreen->info.pci_id);
- printf("family = %i\n", rscreen->info.family);
+ printf("family = %i (%s)\n", rscreen->info.family,
+ r600_get_chip_name(rscreen));
printf("chip_class = %i\n", rscreen->info.chip_class);
- printf("gart_size = %i MB\n", (int)(rscreen->info.gart_size >> 20));
- printf("vram_size = %i MB\n", (int)(rscreen->info.vram_size >> 20));
- printf("max_sclk = %i\n", rscreen->info.max_sclk);
- printf("max_compute_units = %i\n", rscreen->info.max_compute_units);
- printf("max_se = %i\n", rscreen->info.max_se);
- printf("max_sh_per_se = %i\n", rscreen->info.max_sh_per_se);
- printf("drm = %i.%i.%i\n", rscreen->info.drm_major,
- rscreen->info.drm_minor, rscreen->info.drm_patchlevel);
+ printf("gart_size = %i MB\n", (int)DIV_ROUND_UP(rscreen->info.gart_size, 1024*1024));
+ printf("vram_size = %i MB\n", (int)DIV_ROUND_UP(rscreen->info.vram_size, 1024*1024));
+ printf("has_virtual_memory = %i\n", rscreen->info.has_virtual_memory);
+ printf("gfx_ib_pad_with_type2 = %i\n", rscreen->info.gfx_ib_pad_with_type2);
+ printf("has_sdma = %i\n", rscreen->info.has_sdma);
printf("has_uvd = %i\n", rscreen->info.has_uvd);
printf("vce_fw_version = %i\n", rscreen->info.vce_fw_version);
- printf("r600_num_backends = %i\n", rscreen->info.r600_num_backends);
- printf("r600_clock_crystal_freq = %i\n", rscreen->info.r600_clock_crystal_freq);
+ printf("vce_harvest_config = %i\n", rscreen->info.vce_harvest_config);
+ printf("clock_crystal_freq = %i\n", rscreen->info.clock_crystal_freq);
+ printf("drm = %i.%i.%i\n", rscreen->info.drm_major,
+ rscreen->info.drm_minor, rscreen->info.drm_patchlevel);
+ printf("has_userptr = %i\n", rscreen->info.has_userptr);
+
+ printf("r600_max_quad_pipes = %i\n", rscreen->info.r600_max_quad_pipes);
+ printf("max_shader_clock = %i\n", rscreen->info.max_shader_clock);
+ printf("num_good_compute_units = %i\n", rscreen->info.num_good_compute_units);
+ printf("max_se = %i\n", rscreen->info.max_se);
+ printf("max_sh_per_se = %i\n", rscreen->info.max_sh_per_se);
+
+ printf("r600_gb_backend_map = %i\n", rscreen->info.r600_gb_backend_map);
+ printf("r600_gb_backend_map_valid = %i\n", rscreen->info.r600_gb_backend_map_valid);
printf("r600_tiling_config = 0x%x\n", rscreen->info.r600_tiling_config);
- printf("r600_num_tile_pipes = %i\n", rscreen->info.r600_num_tile_pipes);
- printf("r600_max_pipes = %i\n", rscreen->info.r600_max_pipes);
- printf("r600_virtual_address = %i\n", rscreen->info.r600_virtual_address);
- printf("r600_has_dma = %i\n", rscreen->info.r600_has_dma);
- printf("r600_backend_map = %i\n", rscreen->info.r600_backend_map);
- printf("r600_backend_map_valid = %i\n", rscreen->info.r600_backend_map_valid);
+ printf("num_render_backends = %i\n", rscreen->info.num_render_backends);
+ printf("num_tile_pipes = %i\n", rscreen->info.num_tile_pipes);
printf("si_tile_mode_array_valid = %i\n", rscreen->info.si_tile_mode_array_valid);
printf("cik_macrotile_mode_array_valid = %i\n", rscreen->info.cik_macrotile_mode_array_valid);
}
void r600_destroy_common_screen(struct r600_common_screen *rscreen)
{
+ r600_perfcounters_destroy(rscreen);
r600_gpu_load_kill_thread(rscreen);
pipe_mutex_destroy(rscreen->gpu_load_mutex);
}
bool r600_can_dump_shader(struct r600_common_screen *rscreen,
- const struct tgsi_token *tokens)
+ unsigned processor)
{
- /* Compute shader don't have tgsi_tokens */
- if (!tokens)
- return (rscreen->debug_flags & DBG_CS) != 0;
-
- switch (tgsi_get_processor_type(tokens)) {
+ switch (processor) {
case TGSI_PROCESSOR_VERTEX:
return (rscreen->debug_flags & DBG_VS) != 0;
case TGSI_PROCESSOR_TESS_CTRL:
projects / mesa.git / blobdiff