/*
Copyright (C) Intel Corp. 2006. All Rights Reserved.
- Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ Intel funded Tungsten Graphics to
develop this 3D driver.
-
+
Permission is hereby granted, free of charge, to any person obtaining
a copy of this software and associated documentation files (the
"Software"), to deal in the Software without restriction, including
distribute, sublicense, and/or sell copies of the Software, and to
permit persons to whom the Software is furnished to do so, subject to
the following conditions:
-
+
The above copyright notice and this permission notice (including the
next paragraph) shall be included in all copies or substantial
portions of the Software.
-
+
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
-
+
**********************************************************************/
/*
* Authors:
- * Keith Whitwell <keith@tungstengraphics.com>
+ * Keith Whitwell <keithw@vmware.com>
*/
-
+
+#include <pthread.h>
#include "main/imports.h"
#include "main/enums.h"
#include "main/shaderobj.h"
#include "program/prog_parameter.h"
+#include "program/prog_print.h"
#include "program/program.h"
#include "program/programopt.h"
#include "tnl/tnl.h"
-#include "talloc.h"
+#include "util/ralloc.h"
+#include "glsl/ir.h"
#include "brw_context.h"
+#include "brw_shader.h"
+#include "brw_nir.h"
#include "brw_wm.h"
+#include "intel_batchbuffer.h"
-static void brwBindProgram( struct gl_context *ctx,
- GLenum target,
- struct gl_program *prog )
+static unsigned
+get_new_program_id(struct intel_screen *screen)
{
- struct brw_context *brw = brw_context(ctx);
-
- switch (target) {
- case GL_VERTEX_PROGRAM_ARB:
- brw->state.dirty.brw |= BRW_NEW_VERTEX_PROGRAM;
- break;
- case GL_FRAGMENT_PROGRAM_ARB:
- brw->state.dirty.brw |= BRW_NEW_FRAGMENT_PROGRAM;
- break;
- }
+ static pthread_mutex_t m = PTHREAD_MUTEX_INITIALIZER;
+ pthread_mutex_lock(&m);
+ unsigned id = screen->program_id++;
+ pthread_mutex_unlock(&m);
+ return id;
}
static struct gl_program *brwNewProgram( struct gl_context *ctx,
- GLenum target,
+ GLenum target,
GLuint id )
{
struct brw_context *brw = brw_context(ctx);
case GL_VERTEX_PROGRAM_ARB: {
struct brw_vertex_program *prog = CALLOC_STRUCT(brw_vertex_program);
if (prog) {
- prog->id = brw->program_id++;
+ prog->id = get_new_program_id(brw->intelScreen);
- return _mesa_init_vertex_program( ctx, &prog->program,
- target, id );
+ return _mesa_init_gl_program(&prog->program.Base, target, id);
}
else
return NULL;
case GL_FRAGMENT_PROGRAM_ARB: {
struct brw_fragment_program *prog = CALLOC_STRUCT(brw_fragment_program);
if (prog) {
- prog->id = brw->program_id++;
+ prog->id = get_new_program_id(brw->intelScreen);
- return _mesa_init_fragment_program( ctx, &prog->program,
- target, id );
+ return _mesa_init_gl_program(&prog->program.Base, target, id);
}
else
return NULL;
}
+ case GL_GEOMETRY_PROGRAM_NV: {
+ struct brw_geometry_program *prog = CALLOC_STRUCT(brw_geometry_program);
+ if (prog) {
+ prog->id = get_new_program_id(brw->intelScreen);
+
+ return _mesa_init_gl_program(&prog->program, target, id);
+ } else {
+ return NULL;
+ }
+ }
+
+ case GL_COMPUTE_PROGRAM_NV: {
+ struct brw_compute_program *prog = CALLOC_STRUCT(brw_compute_program);
+ if (prog) {
+ prog->id = get_new_program_id(brw->intelScreen);
+
+ return _mesa_init_gl_program(&prog->program.Base, target, id);
+ } else {
+ return NULL;
+ }
+ }
+
default:
- return _mesa_new_program(ctx, target, id);
+ unreachable("Unsupported target in brwNewProgram()");
}
}
}
-static GLboolean brwIsProgramNative( struct gl_context *ctx,
- GLenum target,
- struct gl_program *prog )
-{
- return GL_TRUE;
-}
-
-static void
-shader_error(struct gl_context *ctx, struct gl_program *prog, const char *msg)
-{
- struct gl_shader_program *shader;
-
- shader = _mesa_lookup_shader_program(ctx, prog->Id);
-
- if (shader) {
- shader->InfoLog = talloc_strdup_append(shader->InfoLog, msg);
- shader->LinkStatus = GL_FALSE;
- }
-}
-
-static GLboolean brwProgramStringNotify( struct gl_context *ctx,
- GLenum target,
- struct gl_program *prog )
+static GLboolean
+brwProgramStringNotify(struct gl_context *ctx,
+ GLenum target,
+ struct gl_program *prog)
{
struct brw_context *brw = brw_context(ctx);
- int i;
- if (target == GL_FRAGMENT_PROGRAM_ARB) {
+ switch (target) {
+ case GL_FRAGMENT_PROGRAM_ARB: {
struct gl_fragment_program *fprog = (struct gl_fragment_program *) prog;
struct brw_fragment_program *newFP = brw_fragment_program(fprog);
const struct brw_fragment_program *curFP =
brw_fragment_program_const(brw->fragment_program);
- struct gl_shader_program *shader_program;
-
- if (fprog->FogOption) {
- _mesa_append_fog_code(ctx, fprog);
- fprog->FogOption = GL_NONE;
- }
if (newFP == curFP)
- brw->state.dirty.brw |= BRW_NEW_FRAGMENT_PROGRAM;
- newFP->id = brw->program_id++;
- newFP->isGLSL = brw_wm_is_glsl(fprog);
+ brw->ctx.NewDriverState |= BRW_NEW_FRAGMENT_PROGRAM;
+ newFP->id = get_new_program_id(brw->intelScreen);
- /* Don't reject fragment shaders for their Mesa IR state when we're
- * using the new FS backend.
- */
- shader_program = _mesa_lookup_shader_program(ctx, prog->Id);
- if (shader_program
- && shader_program->_LinkedShaders[MESA_SHADER_FRAGMENT]) {
- return GL_TRUE;
- }
+ brw_add_texrect_params(prog);
+
+ prog->nir = brw_create_nir(brw, NULL, prog, MESA_SHADER_FRAGMENT, true);
+
+ brw_fs_precompile(ctx, NULL, prog);
+ break;
}
- else if (target == GL_VERTEX_PROGRAM_ARB) {
+ case GL_VERTEX_PROGRAM_ARB: {
struct gl_vertex_program *vprog = (struct gl_vertex_program *) prog;
struct brw_vertex_program *newVP = brw_vertex_program(vprog);
const struct brw_vertex_program *curVP =
brw_vertex_program_const(brw->vertex_program);
if (newVP == curVP)
- brw->state.dirty.brw |= BRW_NEW_VERTEX_PROGRAM;
+ brw->ctx.NewDriverState |= BRW_NEW_VERTEX_PROGRAM;
if (newVP->program.IsPositionInvariant) {
_mesa_insert_mvp_code(ctx, &newVP->program);
}
- newVP->id = brw->program_id++;
+ newVP->id = get_new_program_id(brw->intelScreen);
/* Also tell tnl about it:
*/
_tnl_program_string(ctx, target, prog);
+
+ brw_add_texrect_params(prog);
+
+ prog->nir = brw_create_nir(brw, NULL, prog, MESA_SHADER_VERTEX,
+ brw->intelScreen->compiler->scalar_vs);
+
+ brw_vs_precompile(ctx, NULL, prog);
+ break;
}
+ default:
+ /*
+ * driver->ProgramStringNotify is only called for ARB programs, fixed
+ * function vertex programs, and ir_to_mesa (which isn't used by the
+ * i965 back-end). Therefore, even after geometry shaders are added,
+ * this function should only ever be called with a target of
+ * GL_VERTEX_PROGRAM_ARB or GL_FRAGMENT_PROGRAM_ARB.
+ */
+ unreachable("Unexpected target in brwProgramStringNotify");
+ }
+
+ return true;
+}
+
+static void
+brw_memory_barrier(struct gl_context *ctx, GLbitfield barriers)
+{
+ struct brw_context *brw = brw_context(ctx);
+ unsigned bits = (PIPE_CONTROL_DATA_CACHE_INVALIDATE |
+ PIPE_CONTROL_NO_WRITE |
+ PIPE_CONTROL_CS_STALL);
+ assert(brw->gen >= 7 && brw->gen <= 9);
- /* Reject programs with subroutines, which are totally broken at the moment
- * (all program flows return when any program flow returns, and
- * the VS also hangs if a function call calls a function.
- *
- * See piglit glsl-{vs,fs}-functions-[23] tests.
+ if (barriers & (GL_VERTEX_ATTRIB_ARRAY_BARRIER_BIT |
+ GL_ELEMENT_ARRAY_BARRIER_BIT |
+ GL_COMMAND_BARRIER_BIT))
+ bits |= PIPE_CONTROL_VF_CACHE_INVALIDATE;
+
+ if (barriers & GL_UNIFORM_BARRIER_BIT)
+ bits |= (PIPE_CONTROL_TEXTURE_CACHE_INVALIDATE |
+ PIPE_CONTROL_CONST_CACHE_INVALIDATE);
+
+ if (barriers & GL_TEXTURE_FETCH_BARRIER_BIT)
+ bits |= PIPE_CONTROL_TEXTURE_CACHE_INVALIDATE;
+
+ if (barriers & GL_TEXTURE_UPDATE_BARRIER_BIT)
+ bits |= PIPE_CONTROL_RENDER_TARGET_FLUSH;
+
+ if (barriers & GL_FRAMEBUFFER_BARRIER_BIT)
+ bits |= (PIPE_CONTROL_DEPTH_CACHE_FLUSH |
+ PIPE_CONTROL_RENDER_TARGET_FLUSH);
+
+ /* Typed surface messages are handled by the render cache on IVB, so we
+ * need to flush it too.
*/
- for (i = 0; i < prog->NumInstructions; i++) {
- struct prog_instruction *inst = prog->Instructions + i;
- int r;
-
- if (prog->Instructions[i].Opcode == OPCODE_CAL) {
- shader_error(ctx, prog,
- "i965 driver doesn't yet support uninlined function "
- "calls. Move to using a single return statement at "
- "the end of the function to work around it.\n");
- return GL_FALSE;
- }
+ if (brw->gen == 7 && !brw->is_haswell)
+ bits |= PIPE_CONTROL_RENDER_TARGET_FLUSH;
- if (prog->Instructions[i].Opcode == OPCODE_RET) {
- shader_error(ctx, prog,
- "i965 driver doesn't yet support \"return\" "
- "from main().\n");
- return GL_FALSE;
- }
+ brw_emit_pipe_control_flush(brw, bits);
+}
- for (r = 0; r < _mesa_num_inst_src_regs(inst->Opcode); r++) {
- if (prog->Instructions[i].SrcReg[r].RelAddr &&
- prog->Instructions[i].SrcReg[r].File == PROGRAM_INPUT) {
- shader_error(ctx, prog,
- "Variable indexing of shader inputs unsupported\n");
- return GL_FALSE;
- }
- }
+void
+brw_add_texrect_params(struct gl_program *prog)
+{
+ for (int texunit = 0; texunit < BRW_MAX_TEX_UNIT; texunit++) {
+ if (!(prog->TexturesUsed[texunit] & (1 << TEXTURE_RECT_INDEX)))
+ continue;
+
+ int tokens[STATE_LENGTH] = {
+ STATE_INTERNAL,
+ STATE_TEXRECT_SCALE,
+ texunit,
+ 0,
+ 0
+ };
+
+ _mesa_add_state_reference(prog->Parameters, (gl_state_index *)tokens);
+ }
+}
- if (target == GL_FRAGMENT_PROGRAM_ARB &&
- prog->Instructions[i].DstReg.RelAddr &&
- prog->Instructions[i].DstReg.File == PROGRAM_OUTPUT) {
- shader_error(ctx, prog,
- "Variable indexing of FS outputs unsupported\n");
- return GL_FALSE;
- }
- if (target == GL_FRAGMENT_PROGRAM_ARB) {
- if ((prog->Instructions[i].DstReg.RelAddr &&
- prog->Instructions[i].DstReg.File == PROGRAM_TEMPORARY) ||
- (prog->Instructions[i].SrcReg[0].RelAddr &&
- prog->Instructions[i].SrcReg[0].File == PROGRAM_TEMPORARY) ||
- (prog->Instructions[i].SrcReg[1].RelAddr &&
- prog->Instructions[i].SrcReg[1].File == PROGRAM_TEMPORARY) ||
- (prog->Instructions[i].SrcReg[2].RelAddr &&
- prog->Instructions[i].SrcReg[2].File == PROGRAM_TEMPORARY)) {
- shader_error(ctx, prog,
- "Variable indexing of variable arrays in the FS "
- "unsupported\n");
- return GL_FALSE;
- }
- }
+void
+brw_get_scratch_bo(struct brw_context *brw,
+ drm_intel_bo **scratch_bo, int size)
+{
+ drm_intel_bo *old_bo = *scratch_bo;
+
+ if (old_bo && old_bo->size < size) {
+ drm_intel_bo_unreference(old_bo);
+ old_bo = NULL;
}
- return GL_TRUE;
+ if (!old_bo) {
+ *scratch_bo = drm_intel_bo_alloc(brw->bufmgr, "scratch bo", size, 4096);
+ }
}
void brwInitFragProgFuncs( struct dd_function_table *functions )
{
- assert(functions->ProgramStringNotify == _tnl_program_string);
+ /* assert(functions->ProgramStringNotify == _tnl_program_string); */
- functions->BindProgram = brwBindProgram;
functions->NewProgram = brwNewProgram;
functions->DeleteProgram = brwDeleteProgram;
- functions->IsProgramNative = brwIsProgramNative;
functions->ProgramStringNotify = brwProgramStringNotify;
functions->NewShader = brw_new_shader;
- functions->NewShaderProgram = brw_new_shader_program;
- functions->CompileShader = brw_compile_shader;
functions->LinkShader = brw_link_shader;
+
+ functions->MemoryBarrier = brw_memory_barrier;
+}
+
+struct shader_times {
+ uint64_t time;
+ uint64_t written;
+ uint64_t reset;
+};
+
+void
+brw_init_shader_time(struct brw_context *brw)
+{
+ const int max_entries = 2048;
+ brw->shader_time.bo =
+ drm_intel_bo_alloc(brw->bufmgr, "shader time",
+ max_entries * SHADER_TIME_STRIDE * 3, 4096);
+ brw->shader_time.names = rzalloc_array(brw, const char *, max_entries);
+ brw->shader_time.ids = rzalloc_array(brw, int, max_entries);
+ brw->shader_time.types = rzalloc_array(brw, enum shader_time_shader_type,
+ max_entries);
+ brw->shader_time.cumulative = rzalloc_array(brw, struct shader_times,
+ max_entries);
+ brw->shader_time.max_entries = max_entries;
+}
+
+static int
+compare_time(const void *a, const void *b)
+{
+ uint64_t * const *a_val = a;
+ uint64_t * const *b_val = b;
+
+ /* We don't just subtract because we're turning the value to an int. */
+ if (**a_val < **b_val)
+ return -1;
+ else if (**a_val == **b_val)
+ return 0;
+ else
+ return 1;
+}
+
+static void
+print_shader_time_line(const char *stage, const char *name,
+ int shader_num, uint64_t time, uint64_t total)
+{
+ fprintf(stderr, "%-6s%-18s", stage, name);
+
+ if (shader_num != 0)
+ fprintf(stderr, "%4d: ", shader_num);
+ else
+ fprintf(stderr, " : ");
+
+ fprintf(stderr, "%16lld (%7.2f Gcycles) %4.1f%%\n",
+ (long long)time,
+ (double)time / 1000000000.0,
+ (double)time / total * 100.0);
+}
+
+static void
+brw_report_shader_time(struct brw_context *brw)
+{
+ if (!brw->shader_time.bo || !brw->shader_time.num_entries)
+ return;
+
+ uint64_t scaled[brw->shader_time.num_entries];
+ uint64_t *sorted[brw->shader_time.num_entries];
+ uint64_t total_by_type[ST_CS + 1];
+ memset(total_by_type, 0, sizeof(total_by_type));
+ double total = 0;
+ for (int i = 0; i < brw->shader_time.num_entries; i++) {
+ uint64_t written = 0, reset = 0;
+ enum shader_time_shader_type type = brw->shader_time.types[i];
+
+ sorted[i] = &scaled[i];
+
+ switch (type) {
+ case ST_VS:
+ case ST_GS:
+ case ST_FS8:
+ case ST_FS16:
+ case ST_CS:
+ written = brw->shader_time.cumulative[i].written;
+ reset = brw->shader_time.cumulative[i].reset;
+ break;
+
+ default:
+ /* I sometimes want to print things that aren't the 3 shader times.
+ * Just print the sum in that case.
+ */
+ written = 1;
+ reset = 0;
+ break;
+ }
+
+ uint64_t time = brw->shader_time.cumulative[i].time;
+ if (written) {
+ scaled[i] = time / written * (written + reset);
+ } else {
+ scaled[i] = time;
+ }
+
+ switch (type) {
+ case ST_VS:
+ case ST_GS:
+ case ST_FS8:
+ case ST_FS16:
+ case ST_CS:
+ total_by_type[type] += scaled[i];
+ break;
+ default:
+ break;
+ }
+
+ total += scaled[i];
+ }
+
+ if (total == 0) {
+ fprintf(stderr, "No shader time collected yet\n");
+ return;
+ }
+
+ qsort(sorted, brw->shader_time.num_entries, sizeof(sorted[0]), compare_time);
+
+ fprintf(stderr, "\n");
+ fprintf(stderr, "type ID cycles spent %% of total\n");
+ for (int s = 0; s < brw->shader_time.num_entries; s++) {
+ const char *stage;
+ /* Work back from the sorted pointers times to a time to print. */
+ int i = sorted[s] - scaled;
+
+ if (scaled[i] == 0)
+ continue;
+
+ int shader_num = brw->shader_time.ids[i];
+ const char *shader_name = brw->shader_time.names[i];
+
+ switch (brw->shader_time.types[i]) {
+ case ST_VS:
+ stage = "vs";
+ break;
+ case ST_GS:
+ stage = "gs";
+ break;
+ case ST_FS8:
+ stage = "fs8";
+ break;
+ case ST_FS16:
+ stage = "fs16";
+ break;
+ case ST_CS:
+ stage = "cs";
+ break;
+ default:
+ stage = "other";
+ break;
+ }
+
+ print_shader_time_line(stage, shader_name, shader_num,
+ scaled[i], total);
+ }
+
+ fprintf(stderr, "\n");
+ print_shader_time_line("total", "vs", 0, total_by_type[ST_VS], total);
+ print_shader_time_line("total", "gs", 0, total_by_type[ST_GS], total);
+ print_shader_time_line("total", "fs8", 0, total_by_type[ST_FS8], total);
+ print_shader_time_line("total", "fs16", 0, total_by_type[ST_FS16], total);
+ print_shader_time_line("total", "cs", 0, total_by_type[ST_CS], total);
+}
+
+static void
+brw_collect_shader_time(struct brw_context *brw)
+{
+ if (!brw->shader_time.bo)
+ return;
+
+ /* This probably stalls on the last rendering. We could fix that by
+ * delaying reading the reports, but it doesn't look like it's a big
+ * overhead compared to the cost of tracking the time in the first place.
+ */
+ drm_intel_bo_map(brw->shader_time.bo, true);
+ void *bo_map = brw->shader_time.bo->virtual;
+
+ for (int i = 0; i < brw->shader_time.num_entries; i++) {
+ uint32_t *times = bo_map + i * 3 * SHADER_TIME_STRIDE;
+
+ brw->shader_time.cumulative[i].time += times[SHADER_TIME_STRIDE * 0 / 4];
+ brw->shader_time.cumulative[i].written += times[SHADER_TIME_STRIDE * 1 / 4];
+ brw->shader_time.cumulative[i].reset += times[SHADER_TIME_STRIDE * 2 / 4];
+ }
+
+ /* Zero the BO out to clear it out for our next collection.
+ */
+ memset(bo_map, 0, brw->shader_time.bo->size);
+ drm_intel_bo_unmap(brw->shader_time.bo);
+}
+
+void
+brw_collect_and_report_shader_time(struct brw_context *brw)
+{
+ brw_collect_shader_time(brw);
+
+ if (brw->shader_time.report_time == 0 ||
+ get_time() - brw->shader_time.report_time >= 1.0) {
+ brw_report_shader_time(brw);
+ brw->shader_time.report_time = get_time();
+ }
}
+/**
+ * Chooses an index in the shader_time buffer and sets up tracking information
+ * for our printouts.
+ *
+ * Note that this holds on to references to the underlying programs, which may
+ * change their lifetimes compared to normal operation.
+ */
+int
+brw_get_shader_time_index(struct brw_context *brw,
+ struct gl_shader_program *shader_prog,
+ struct gl_program *prog,
+ enum shader_time_shader_type type)
+{
+ int shader_time_index = brw->shader_time.num_entries++;
+ assert(shader_time_index < brw->shader_time.max_entries);
+ brw->shader_time.types[shader_time_index] = type;
+
+ int id = shader_prog ? shader_prog->Name : prog->Id;
+ const char *name;
+ if (id == 0) {
+ name = "ff";
+ } else if (!shader_prog) {
+ name = "prog";
+ } else if (shader_prog->Label) {
+ name = ralloc_strdup(brw->shader_time.names, shader_prog->Label);
+ } else {
+ name = "glsl";
+ }
+
+ brw->shader_time.names[shader_time_index] = name;
+ brw->shader_time.ids[shader_time_index] = id;
+
+ return shader_time_index;
+}
+
+void
+brw_destroy_shader_time(struct brw_context *brw)
+{
+ drm_intel_bo_unreference(brw->shader_time.bo);
+ brw->shader_time.bo = NULL;
+}
+
+void
+brw_stage_prog_data_free(const void *p)
+{
+ struct brw_stage_prog_data *prog_data = (struct brw_stage_prog_data *)p;
+
+ ralloc_free(prog_data->param);
+ ralloc_free(prog_data->pull_param);
+ ralloc_free(prog_data->image_param);
+}
+
+void
+brw_dump_ir(const char *stage, struct gl_shader_program *shader_prog,
+ struct gl_shader *shader, struct gl_program *prog)
+{
+ if (shader_prog) {
+ if (shader->ir) {
+ fprintf(stderr,
+ "GLSL IR for native %s shader %d:\n",
+ stage, shader_prog->Name);
+ _mesa_print_ir(stderr, shader->ir, NULL);
+ fprintf(stderr, "\n\n");
+ }
+ } else {
+ fprintf(stderr, "ARB_%s_program %d ir for native %s shader\n",
+ stage, prog->Id, stage);
+ _mesa_print_program(prog);
+ }
+}
+
+void
+brw_setup_tex_for_precompile(struct brw_context *brw,
+ struct brw_sampler_prog_key_data *tex,
+ struct gl_program *prog)
+{
+ const bool has_shader_channel_select = brw->is_haswell || brw->gen >= 8;
+ unsigned sampler_count = _mesa_fls(prog->SamplersUsed);
+ for (unsigned i = 0; i < sampler_count; i++) {
+ if (!has_shader_channel_select && (prog->ShadowSamplers & (1 << i))) {
+ /* Assume DEPTH_TEXTURE_MODE is the default: X, X, X, 1 */
+ tex->swizzles[i] =
+ MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_ONE);
+ } else {
+ /* Color sampler: assume no swizzling. */
+ tex->swizzles[i] = SWIZZLE_XYZW;
+ }
+ }
+}
projects / mesa.git / blobdiff