#include "glsl/ralloc.h"
-/** Return number of src args for given instruction */
-GLuint brw_wm_nr_args( GLuint opcode )
-{
- switch (opcode) {
- case WM_FRONTFACING:
- case WM_PIXELXY:
- return 0;
- case WM_CINTERP:
- case WM_WPOSXY:
- case WM_DELTAXY:
- return 1;
- case WM_LINTERP:
- case WM_PIXELW:
- return 2;
- case WM_FB_WRITE:
- case WM_PINTERP:
- return 3;
- default:
- assert(opcode < MAX_OPCODE);
- return _mesa_num_inst_src_regs(opcode);
- }
-}
-
-
-GLuint brw_wm_is_scalar_result( GLuint opcode )
-{
- switch (opcode) {
- case OPCODE_COS:
- case OPCODE_EX2:
- case OPCODE_LG2:
- case OPCODE_POW:
- case OPCODE_RCP:
- case OPCODE_RSQ:
- case OPCODE_SIN:
- case OPCODE_DP2:
- case OPCODE_DP3:
- case OPCODE_DP4:
- case OPCODE_DPH:
- case OPCODE_DST:
- return 1;
-
- default:
- return 0;
- }
-}
-
-
-/**
- * Do GPU code generation for non-GLSL shader. non-GLSL shaders have
- * no flow control instructions so we can more readily do SSA-style
- * optimizations.
- */
-static void
-brw_wm_non_glsl_emit(struct brw_context *brw, struct brw_wm_compile *c)
-{
- /* Augment fragment program. Add instructions for pre- and
- * post-fragment-program tasks such as interpolation and fogging.
- */
- brw_wm_pass_fp(c);
-
- /* Translate to intermediate representation. Build register usage
- * chains.
- */
- brw_wm_pass0(c);
-
- /* Dead code removal.
- */
- brw_wm_pass1(c);
-
- /* Register allocation.
- * Divide by two because we operate on 16 pixels at a time and require
- * two GRF entries for each logical shader register.
- */
- c->grf_limit = BRW_WM_MAX_GRF / 2;
-
- brw_wm_pass2(c);
-
- /* how many general-purpose registers are used */
- c->prog_data.reg_blocks = brw_register_blocks(c->max_wm_grf);
-
- /* Emit GEN4 code.
- */
- brw_wm_emit(c);
-}
-
-
/**
* Return a bitfield where bit n is set if barycentric interpolation mode n
* (see enum brw_wm_barycentric_interp_mode) is needed by the fragment shader.
* modes are in use, and set the appropriate bits in
* barycentric_interp_modes.
*/
- for (attr = 0; attr < FRAG_ATTRIB_MAX; ++attr) {
+ for (attr = 0; attr < VARYING_SLOT_MAX; ++attr) {
enum glsl_interp_qualifier interp_qualifier =
fprog->InterpQualifier[attr];
bool is_centroid = fprog->IsCentroid & BITFIELD64_BIT(attr);
- bool is_gl_Color = attr == FRAG_ATTRIB_COL0 || attr == FRAG_ATTRIB_COL1;
+ bool is_gl_Color = attr == VARYING_SLOT_COL0 || attr == VARYING_SLOT_COL1;
/* Ignore unused inputs. */
if (!(fprog->Base.InputsRead & BITFIELD64_BIT(attr)))
continue;
/* Ignore WPOS and FACE, because they don't require interpolation. */
- if (attr == FRAG_ATTRIB_WPOS || attr == FRAG_ATTRIB_FACE)
+ if (attr == VARYING_SLOT_POS || attr == VARYING_SLOT_FACE)
continue;
/* Determine the set (or sets) of barycentric coordinates needed to
return barycentric_interp_modes;
}
+bool
+brw_wm_prog_data_compare(const void *in_a, const void *in_b,
+ int aux_size, const void *in_key)
+{
+ const struct brw_wm_prog_data *a = in_a;
+ const struct brw_wm_prog_data *b = in_b;
+
+ /* Compare all the struct up to the pointers. */
+ if (memcmp(a, b, offsetof(struct brw_wm_prog_data, param)))
+ return false;
+
+ if (memcmp(a->param, b->param, a->nr_params * sizeof(void *)))
+ return false;
+
+ if (memcmp(a->pull_param, b->pull_param, a->nr_pull_params * sizeof(void *)))
+ return false;
+
+ return true;
+}
void
-brw_wm_payload_setup(struct brw_context *brw,
- struct brw_wm_compile *c)
+brw_wm_prog_data_free(const void *in_prog_data)
{
- struct intel_context *intel = &brw->intel;
- bool uses_depth = (c->fp->program.Base.InputsRead &
- (1 << FRAG_ATTRIB_WPOS)) != 0;
- unsigned barycentric_interp_modes = c->prog_data.barycentric_interp_modes;
- int i;
-
- if (intel->gen >= 6) {
- /* R0-1: masks, pixel X/Y coordinates. */
- c->nr_payload_regs = 2;
- /* R2: only for 32-pixel dispatch.*/
-
- /* R3-26: barycentric interpolation coordinates. These appear in the
- * same order that they appear in the brw_wm_barycentric_interp_mode
- * enum. Each set of coordinates occupies 2 registers if dispatch width
- * == 8 and 4 registers if dispatch width == 16. Coordinates only
- * appear if they were enabled using the "Barycentric Interpolation
- * Mode" bits in WM_STATE.
- */
- for (i = 0; i < BRW_WM_BARYCENTRIC_INTERP_MODE_COUNT; ++i) {
- if (barycentric_interp_modes & (1 << i)) {
- c->barycentric_coord_reg[i] = c->nr_payload_regs;
- c->nr_payload_regs += 2;
- if (c->dispatch_width == 16) {
- c->nr_payload_regs += 2;
- }
- }
- }
+ const struct brw_wm_prog_data *prog_data = in_prog_data;
- /* R27: interpolated depth if uses source depth */
- if (uses_depth) {
- c->source_depth_reg = c->nr_payload_regs;
- c->nr_payload_regs++;
- if (c->dispatch_width == 16) {
- /* R28: interpolated depth if not 8-wide. */
- c->nr_payload_regs++;
- }
- }
- /* R29: interpolated W set if GEN6_WM_USES_SOURCE_W.
- */
- if (uses_depth) {
- c->source_w_reg = c->nr_payload_regs;
- c->nr_payload_regs++;
- if (c->dispatch_width == 16) {
- /* R30: interpolated W if not 8-wide. */
- c->nr_payload_regs++;
- }
- }
- /* R31: MSAA position offsets. */
- /* R32-: bary for 32-pixel. */
- /* R58-59: interp W for 32-pixel. */
-
- if (c->fp->program.Base.OutputsWritten &
- BITFIELD64_BIT(FRAG_RESULT_DEPTH)) {
- c->source_depth_to_render_target = true;
- c->computes_depth = true;
- }
- } else {
- brw_wm_lookup_iz(intel, c);
- }
+ ralloc_free((void *)prog_data->param);
+ ralloc_free((void *)prog_data->pull_param);
}
/**
struct brw_fragment_program *fp,
struct brw_wm_prog_key *key)
{
- struct intel_context *intel = &brw->intel;
struct brw_wm_compile *c;
const GLuint *program;
+ struct gl_shader *fs = NULL;
GLuint program_size;
- c = brw->wm.compile_data;
- if (c == NULL) {
- brw->wm.compile_data = rzalloc(NULL, struct brw_wm_compile);
- c = brw->wm.compile_data;
- if (c == NULL) {
- /* Ouch - big out of memory problem. Can't continue
- * without triggering a segfault, no way to signal,
- * so just return.
- */
- return false;
- }
+ if (prog)
+ fs = prog->_LinkedShaders[MESA_SHADER_FRAGMENT];
+
+ c = rzalloc(NULL, struct brw_wm_compile);
+
+ /* Allocate the references to the uniforms that will end up in the
+ * prog_data associated with the compiled program, and which will be freed
+ * by the state cache.
+ */
+ int param_count;
+ if (fs) {
+ param_count = fs->num_uniform_components;
} else {
- void *instruction = c->instruction;
- void *prog_instructions = c->prog_instructions;
- void *vreg = c->vreg;
- void *refs = c->refs;
- memset(c, 0, sizeof(*brw->wm.compile_data));
- c->instruction = instruction;
- c->prog_instructions = prog_instructions;
- c->vreg = vreg;
- c->refs = refs;
+ param_count = fp->program.Base.Parameters->NumParameters * 4;
}
- memcpy(&c->key, key, sizeof(*key));
-
- c->fp = fp;
- c->env_param = brw->intel.ctx.FragmentProgram.Parameters;
+ /* The backend also sometimes adds params for texture size. */
+ param_count += 2 * BRW_MAX_TEX_UNIT;
+ c->prog_data.param = rzalloc_array(NULL, const float *, param_count);
+ c->prog_data.pull_param = rzalloc_array(NULL, const float *, param_count);
- brw_init_compile(brw, &c->func, c);
+ memcpy(&c->key, key, sizeof(*key));
c->prog_data.barycentric_interp_modes =
brw_compute_barycentric_interp_modes(brw, c->key.flat_shade,
&fp->program);
- if (prog && prog->_LinkedShaders[MESA_SHADER_FRAGMENT]) {
- if (!brw_wm_fs_emit(brw, c, prog))
- return false;
- } else {
- if (!c->instruction) {
- c->instruction = rzalloc_array(c, struct brw_wm_instruction, BRW_WM_MAX_INSN);
- c->prog_instructions = rzalloc_array(c, struct prog_instruction, BRW_WM_MAX_INSN);
- c->vreg = rzalloc_array(c, struct brw_wm_value, BRW_WM_MAX_VREG);
- c->refs = rzalloc_array(c, struct brw_wm_ref, BRW_WM_MAX_REF);
- }
-
- /* Fallback for fixed function and ARB_fp shaders. */
- c->dispatch_width = 16;
- brw_wm_payload_setup(brw, c);
- brw_wm_non_glsl_emit(brw, c);
- c->prog_data.dispatch_width = 16;
- }
+ program = brw_wm_fs_emit(brw, c, &fp->program, prog, &program_size);
+ if (program == NULL)
+ return false;
/* Scratch space is used for register spilling */
if (c->last_scratch) {
+ perf_debug("Fragment shader triggered register spilling. "
+ "Try reducing the number of live scalar values to "
+ "improve performance.\n");
+
c->prog_data.total_scratch = brw_get_scratch_size(c->last_scratch);
- brw_get_scratch_bo(intel, &brw->wm.scratch_bo,
+ brw_get_scratch_bo(brw, &brw->wm.scratch_bo,
c->prog_data.total_scratch * brw->max_wm_threads);
}
if (unlikely(INTEL_DEBUG & DEBUG_WM))
fprintf(stderr, "\n");
- /* get the program
- */
- program = brw_get_program(&c->func, &program_size);
-
brw_upload_cache(&brw->cache, BRW_WM_PROG,
&c->key, sizeof(c->key),
program, program_size,
&c->prog_data, sizeof(c->prog_data),
&brw->wm.prog_offset, &brw->wm.prog_data);
+ ralloc_free(c);
+
return true;
}
+static bool
+key_debug(struct brw_context *brw, const char *name, int a, int b)
+{
+ if (a != b) {
+ perf_debug(" %s %d->%d\n", name, a, b);
+ return true;
+ } else {
+ return false;
+ }
+}
+
+bool
+brw_debug_recompile_sampler_key(struct brw_context *brw,
+ const struct brw_sampler_prog_key_data *old_key,
+ const struct brw_sampler_prog_key_data *key)
+{
+ bool found = false;
+
+ for (unsigned int i = 0; i < MAX_SAMPLERS; i++) {
+ found |= key_debug(brw, "EXT_texture_swizzle or DEPTH_TEXTURE_MODE",
+ old_key->swizzles[i], key->swizzles[i]);
+ }
+ found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 1st coordinate",
+ old_key->gl_clamp_mask[0], key->gl_clamp_mask[0]);
+ found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 2nd coordinate",
+ old_key->gl_clamp_mask[1], key->gl_clamp_mask[1]);
+ found |= key_debug(brw, "GL_CLAMP enabled on any texture unit's 3rd coordinate",
+ old_key->gl_clamp_mask[2], key->gl_clamp_mask[2]);
+ found |= key_debug(brw, "GL_MESA_ycbcr texturing\n",
+ old_key->yuvtex_mask, key->yuvtex_mask);
+ found |= key_debug(brw, "GL_MESA_ycbcr UV swapping\n",
+ old_key->yuvtex_swap_mask, key->yuvtex_swap_mask);
+
+ return found;
+}
+
+void
+brw_wm_debug_recompile(struct brw_context *brw,
+ struct gl_shader_program *prog,
+ const struct brw_wm_prog_key *key)
+{
+ struct brw_cache_item *c = NULL;
+ const struct brw_wm_prog_key *old_key = NULL;
+ bool found = false;
+
+ perf_debug("Recompiling fragment shader for program %d\n", prog->Name);
+
+ for (unsigned int i = 0; i < brw->cache.size; i++) {
+ for (c = brw->cache.items[i]; c; c = c->next) {
+ if (c->cache_id == BRW_WM_PROG) {
+ old_key = c->key;
+
+ if (old_key->program_string_id == key->program_string_id)
+ break;
+ }
+ }
+ if (c)
+ break;
+ }
+
+ if (!c) {
+ perf_debug(" Didn't find previous compile in the shader cache for debug\n");
+ return;
+ }
+
+ found |= key_debug(brw, "alphatest, computed depth, depth test, or "
+ "depth write",
+ old_key->iz_lookup, key->iz_lookup);
+ found |= key_debug(brw, "depth statistics",
+ old_key->stats_wm, key->stats_wm);
+ found |= key_debug(brw, "flat shading",
+ old_key->flat_shade, key->flat_shade);
+ found |= key_debug(brw, "number of color buffers",
+ old_key->nr_color_regions, key->nr_color_regions);
+ found |= key_debug(brw, "MRT alpha test or alpha-to-coverage",
+ old_key->replicate_alpha, key->replicate_alpha);
+ found |= key_debug(brw, "rendering to FBO",
+ old_key->render_to_fbo, key->render_to_fbo);
+ found |= key_debug(brw, "fragment color clamping",
+ old_key->clamp_fragment_color, key->clamp_fragment_color);
+ found |= key_debug(brw, "line smoothing",
+ old_key->line_aa, key->line_aa);
+ found |= key_debug(brw, "renderbuffer height",
+ old_key->drawable_height, key->drawable_height);
+ found |= key_debug(brw, "input slots valid",
+ old_key->input_slots_valid, key->input_slots_valid);
+
+ found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex);
+
+ if (!found) {
+ perf_debug(" Something else\n");
+ }
+}
+
void
brw_populate_sampler_prog_key_data(struct gl_context *ctx,
const struct gl_program *prog,
+ unsigned sampler_count,
struct brw_sampler_prog_key_data *key)
{
- for (int i = 0; i < BRW_MAX_TEX_UNIT; i++) {
- if (!prog->TexturesUsed[i])
+ struct brw_context *brw = brw_context(ctx);
+
+ for (int s = 0; s < sampler_count; s++) {
+ key->swizzles[s] = SWIZZLE_NOOP;
+
+ if (!(prog->SamplersUsed & (1 << s)))
continue;
- const struct gl_texture_unit *unit = &ctx->Texture.Unit[i];
+ int unit_id = prog->SamplerUnits[s];
+ const struct gl_texture_unit *unit = &ctx->Texture.Unit[unit_id];
if (unit->_ReallyEnabled && unit->_Current->Target != GL_TEXTURE_BUFFER) {
const struct gl_texture_object *t = unit->_Current;
const struct gl_texture_image *img = t->Image[0][t->BaseLevel];
- struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, i);
- int swizzles[SWIZZLE_NIL + 1] = {
- SWIZZLE_X,
- SWIZZLE_Y,
- SWIZZLE_Z,
- SWIZZLE_W,
- SWIZZLE_ZERO,
- SWIZZLE_ONE,
- SWIZZLE_NIL
- };
-
- if (img->_BaseFormat == GL_DEPTH_COMPONENT ||
- img->_BaseFormat == GL_DEPTH_STENCIL) {
- /* We handle GL_DEPTH_TEXTURE_MODE here instead of as surface
- * format overrides because shadow comparison always returns the
- * result of the comparison in all channels anyway.
- */
- switch (t->DepthMode) {
- case GL_ALPHA:
- swizzles[0] = SWIZZLE_ZERO;
- swizzles[1] = SWIZZLE_ZERO;
- swizzles[2] = SWIZZLE_ZERO;
- swizzles[3] = SWIZZLE_X;
- break;
- case GL_LUMINANCE:
- swizzles[0] = SWIZZLE_X;
- swizzles[1] = SWIZZLE_X;
- swizzles[2] = SWIZZLE_X;
- swizzles[3] = SWIZZLE_ONE;
- break;
- case GL_INTENSITY:
- swizzles[0] = SWIZZLE_X;
- swizzles[1] = SWIZZLE_X;
- swizzles[2] = SWIZZLE_X;
- swizzles[3] = SWIZZLE_X;
- break;
- case GL_RED:
- swizzles[0] = SWIZZLE_X;
- swizzles[1] = SWIZZLE_ZERO;
- swizzles[2] = SWIZZLE_ZERO;
- swizzles[3] = SWIZZLE_ONE;
- break;
- }
- }
+ struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit_id);
+
+ const bool alpha_depth = t->DepthMode == GL_ALPHA &&
+ (img->_BaseFormat == GL_DEPTH_COMPONENT ||
+ img->_BaseFormat == GL_DEPTH_STENCIL);
+
+ /* Haswell handles texture swizzling as surface format overrides
+ * (except for GL_ALPHA); all other platforms need MOVs in the shader.
+ */
+ if (!brw->is_haswell || alpha_depth)
+ key->swizzles[s] = brw_get_texture_swizzle(ctx, t);
if (img->InternalFormat == GL_YCBCR_MESA) {
- key->yuvtex_mask |= 1 << i;
+ key->yuvtex_mask |= 1 << s;
if (img->TexFormat == MESA_FORMAT_YCBCR)
- key->yuvtex_swap_mask |= 1 << i;
+ key->yuvtex_swap_mask |= 1 << s;
}
- key->swizzles[i] =
- MAKE_SWIZZLE4(swizzles[GET_SWZ(t->_Swizzle, 0)],
- swizzles[GET_SWZ(t->_Swizzle, 1)],
- swizzles[GET_SWZ(t->_Swizzle, 2)],
- swizzles[GET_SWZ(t->_Swizzle, 3)]);
-
if (sampler->MinFilter != GL_NEAREST &&
sampler->MagFilter != GL_NEAREST) {
if (sampler->WrapS == GL_CLAMP)
- key->gl_clamp_mask[0] |= 1 << i;
+ key->gl_clamp_mask[0] |= 1 << s;
if (sampler->WrapT == GL_CLAMP)
- key->gl_clamp_mask[1] |= 1 << i;
+ key->gl_clamp_mask[1] |= 1 << s;
if (sampler->WrapR == GL_CLAMP)
- key->gl_clamp_mask[2] |= 1 << i;
+ key->gl_clamp_mask[2] |= 1 << s;
}
}
- else {
- key->swizzles[i] = SWIZZLE_NOOP;
- }
}
}
static void brw_wm_populate_key( struct brw_context *brw,
struct brw_wm_prog_key *key )
{
- struct gl_context *ctx = &brw->intel.ctx;
- struct intel_context *intel = &brw->intel;
+ struct gl_context *ctx = &brw->ctx;
/* BRW_NEW_FRAGMENT_PROGRAM */
const struct brw_fragment_program *fp =
(struct brw_fragment_program *)brw->fragment_program;
/* Build the index for table lookup
*/
- if (intel->gen < 6) {
+ if (brw->gen < 6) {
/* _NEW_COLOR */
if (fp->program.UsesKill || ctx->Color.AlphaEnabled)
lookup |= IZ_PS_KILL_ALPHATEST_BIT;
if (ctx->Depth.Test && ctx->Depth.Mask) /* ?? */
lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
- /* _NEW_STENCIL */
+ /* _NEW_STENCIL | _NEW_BUFFERS */
if (ctx->Stencil._Enabled) {
lookup |= IZ_STENCIL_TEST_ENABLE_BIT;
/* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
if (ctx->Line.SmoothFlag) {
- if (brw->intel.reduced_primitive == GL_LINES) {
+ if (brw->reduced_primitive == GL_LINES) {
line_aa = AA_ALWAYS;
}
- else if (brw->intel.reduced_primitive == GL_TRIANGLES) {
+ else if (brw->reduced_primitive == GL_TRIANGLES) {
if (ctx->Polygon.FrontMode == GL_LINE) {
line_aa = AA_SOMETIMES;
}
key->line_aa = line_aa;
- key->stats_wm = brw->intel.stats_wm;
- /* BRW_NEW_WM_INPUT_DIMENSIONS */
- key->proj_attrib_mask = brw->wm.input_size_masks[4-1];
+ if (brw->gen < 6)
+ key->stats_wm = brw->stats_wm;
/* _NEW_LIGHT */
key->flat_shade = (ctx->Light.ShadeModel == GL_FLAT);
key->clamp_fragment_color = ctx->Color._ClampFragmentColor;
/* _NEW_TEXTURE */
- brw_populate_sampler_prog_key_data(ctx, prog, &key->tex);
+ brw_populate_sampler_prog_key_data(ctx, prog, brw->wm.sampler_count,
+ &key->tex);
/* _NEW_BUFFERS */
/*
* For DRI2 the origin_x/y will always be (0,0) but we still need the
* drawable height in order to invert the Y axis.
*/
- if (fp->program.Base.InputsRead & FRAG_BIT_WPOS) {
+ if (fp->program.Base.InputsRead & VARYING_BIT_POS) {
key->drawable_height = ctx->DrawBuffer->Height;
}
- if ((fp->program.Base.InputsRead & FRAG_BIT_WPOS) || program_uses_dfdy) {
+ if ((fp->program.Base.InputsRead & VARYING_BIT_POS) || program_uses_dfdy) {
key->render_to_fbo = _mesa_is_user_fbo(ctx->DrawBuffer);
}
/* _NEW_BUFFERS */
key->nr_color_regions = ctx->DrawBuffer->_NumColorDrawBuffers;
- /* CACHE_NEW_VS_PROG */
- key->vp_outputs_written = brw->vs.prog_data->outputs_written;
+ /* _NEW_MULTISAMPLE, _NEW_COLOR, _NEW_BUFFERS */
+ key->replicate_alpha = ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
+ (ctx->Multisample.SampleAlphaToCoverage || ctx->Color.AlphaEnabled);
+
+ /* BRW_NEW_VUE_MAP_GEOM_OUT */
+ if (brw->gen < 6)
+ key->input_slots_valid = brw->vue_map_geom_out.slots_valid;
/* The unique fragment program ID */
key->program_string_id = fp->id;
static void
brw_upload_wm_prog(struct brw_context *brw)
{
- struct intel_context *intel = &brw->intel;
- struct gl_context *ctx = &intel->ctx;
+ struct gl_context *ctx = &brw->ctx;
struct brw_wm_prog_key key;
struct brw_fragment_program *fp = (struct brw_fragment_program *)
brw->fragment_program;
_NEW_LIGHT |
_NEW_FRAG_CLAMP |
_NEW_BUFFERS |
- _NEW_TEXTURE),
+ _NEW_TEXTURE |
+ _NEW_MULTISAMPLE),
.brw = (BRW_NEW_FRAGMENT_PROGRAM |
- BRW_NEW_WM_INPUT_DIMENSIONS |
- BRW_NEW_REDUCED_PRIMITIVE),
- .cache = CACHE_NEW_VS_PROG,
+ BRW_NEW_REDUCED_PRIMITIVE |
+ BRW_NEW_VUE_MAP_GEOM_OUT |
+ BRW_NEW_STATS_WM)
},
.emit = brw_upload_wm_prog
};