* and the size of the body. So long-ish loops with very simple bodies
* can be unrolled, as well as short loops with larger bodies.
*/
-const GLuint MAX_FOR_LOOP_UNROLL_COMPLEXITY = 200;
+const GLuint MAX_FOR_LOOP_UNROLL_COMPLEXITY = 256;
break;
case SLANG_SPEC_STRUCT:
sz = _slang_field_offset(spec, 0); /* special use */
- if (sz > 4) {
+ if (sz == 1) {
+ /* 1-float structs are actually troublesome to deal with since they
+ * might get placed at R.x, R.y, R.z or R.z. Return size=2 to
+ * ensure the object is placed at R.x
+ */
+ sz = 2;
+ }
+ else if (sz > 4) {
sz = (sz + 3) & ~0x3; /* round up to multiple of four */
}
break;
}
-
-/**
- * Establish the binding between a slang_ir_node and a slang_variable.
- * Then, allocate/attach a slang_ir_storage object to the IR node if needed.
- * The IR node must be a IR_VAR or IR_VAR_DECL node.
- * \param n the IR node
- * \param var the variable to associate with the IR node
- */
-static void
-_slang_attach_storage(slang_ir_node *n, slang_variable *var)
-{
- assert(n);
- assert(var);
- assert(n->Opcode == IR_VAR || n->Opcode == IR_VAR_DECL);
- assert(!n->Var || n->Var == var);
-
- n->Var = var;
-
- if (!n->Store) {
- /* need to setup storage */
- if (n->Var && n->Var->store) {
- /* node storage info = var storage info */
- n->Store = n->Var->store;
- }
- else {
- /* alloc new storage info */
- n->Store = _slang_new_ir_storage(PROGRAM_UNDEFINED, -7, -5);
-#if 0
- printf("%s var=%s Store=%p Size=%d\n", __FUNCTION__,
- (char*) var->a_name,
- (void*) n->Store, n->Store->Size);
-#endif
- if (n->Var)
- n->Var->store = n->Store;
- assert(n->Var->store);
- }
- }
-}
-
-
/**
* Return the TEXTURE_*_INDEX value that corresponds to a sampler type,
* or -1 if the type is not a sampler.
{ NULL, 0 }
};
static const struct output_info fragOutputs[] = {
- { "gl_FragColor", FRAG_RESULT_COLR },
- { "gl_FragDepth", FRAG_RESULT_DEPR },
+ { "gl_FragColor", FRAG_RESULT_COLOR },
+ { "gl_FragDepth", FRAG_RESULT_DEPTH },
{ "gl_FragData", FRAG_RESULT_DATA0 },
{ NULL, 0 }
};
/* float binary op */
{ "float_power", IR_POW, 1, 2 },
/* texture / sampler */
- { "vec4_tex1d", IR_TEX, 1, 2 },
- { "vec4_texb1d", IR_TEXB, 1, 2 }, /* 1d w/ bias */
- { "vec4_texp1d", IR_TEXP, 1, 2 }, /* 1d w/ projection */
- { "vec4_tex2d", IR_TEX, 1, 2 },
- { "vec4_texb2d", IR_TEXB, 1, 2 }, /* 2d w/ bias */
- { "vec4_texp2d", IR_TEXP, 1, 2 }, /* 2d w/ projection */
- { "vec4_tex3d", IR_TEX, 1, 2 },
- { "vec4_texb3d", IR_TEXB, 1, 2 }, /* 3d w/ bias */
- { "vec4_texp3d", IR_TEXP, 1, 2 }, /* 3d w/ projection */
- { "vec4_texcube", IR_TEX, 1, 2 }, /* cubemap */
- { "vec4_tex_rect", IR_TEX, 1, 2 }, /* rectangle */
- { "vec4_texp_rect", IR_TEX, 1, 2 },/* rectangle w/ projection */
+ { "vec4_tex_1d", IR_TEX, 1, 2 },
+ { "vec4_tex_1d_bias", IR_TEXB, 1, 2 }, /* 1d w/ bias */
+ { "vec4_tex_1d_proj", IR_TEXP, 1, 2 }, /* 1d w/ projection */
+ { "vec4_tex_2d", IR_TEX, 1, 2 },
+ { "vec4_tex_2d_bias", IR_TEXB, 1, 2 }, /* 2d w/ bias */
+ { "vec4_tex_2d_proj", IR_TEXP, 1, 2 }, /* 2d w/ projection */
+ { "vec4_tex_3d", IR_TEX, 1, 2 },
+ { "vec4_tex_3d_bias", IR_TEXB, 1, 2 }, /* 3d w/ bias */
+ { "vec4_tex_3d_proj", IR_TEXP, 1, 2 }, /* 3d w/ projection */
+ { "vec4_tex_cube", IR_TEX, 1, 2 }, /* cubemap */
+ { "vec4_tex_rect", IR_TEX, 1, 2 }, /* rectangle */
+ { "vec4_tex_rect_bias", IR_TEX, 1, 2 }, /* rectangle w/ projection */
+
+ /* texture / sampler but with shadow comparison */
+ { "vec4_tex_1d_shadow", IR_TEX_SH, 1, 2 },
+ { "vec4_tex_1d_bias_shadow", IR_TEXB_SH, 1, 2 },
+ { "vec4_tex_1d_proj_shadow", IR_TEXP_SH, 1, 2 },
+ { "vec4_tex_2d_shadow", IR_TEX_SH, 1, 2 },
+ { "vec4_tex_2d_bias_shadow", IR_TEXB_SH, 1, 2 },
+ { "vec4_tex_2d_proj_shadow", IR_TEXP_SH, 1, 2 },
+ { "vec4_tex_rect_shadow", IR_TEX_SH, 1, 2 },
+ { "vec4_tex_rect_proj_shadow", IR_TEXP_SH, 1, 2 },
/* unary op */
{ "ivec4_to_vec4", IR_I_TO_F, 1, 1 }, /* int[4] to float[4] */
{
slang_ir_node *n = new_node0(IR_VAR);
if (n) {
- _slang_attach_storage(n, var);
+ ASSERT(var);
+ ASSERT(var->store);
+ ASSERT(!n->Store);
+ ASSERT(!n->Var);
+
+ /* Set IR node's Var and Store pointers */
+ n->Var = var;
+ n->Store = var->store;
}
return n;
}
/*_slang_label_delete(A->curFuncEndLabel);*/
A->curFuncEndLabel = prevFuncEndLabel;
+ if (A->pragmas->Debug) {
+ char s[1000];
+ snprintf(s, sizeof(s), "Call/inline %s()", (char *) fun->header.a_name);
+ n->Comment = _slang_strdup(s);
+ }
+
return n;
}
*/
slang_variable *p = slang_variable_scope_grow(fun->parameters);
char name[10];
- snprintf(name, sizeof(name), "p%d", i);
+ _mesa_snprintf(name, sizeof(name), "p%d", i);
p->a_name = slang_atom_pool_atom(A->atoms, name);
p->type.qualifier = SLANG_QUAL_CONST;
p->type.specifier.type = baseType;
}
+static void
+_unroll_loop_inc(slang_assemble_ctx * A)
+{
+ A->UnrollLoop++;
+}
+
+
+static void
+_unroll_loop_dec(slang_assemble_ctx * A)
+{
+ A->UnrollLoop--;
+}
+
+
/**
* Unroll a for-loop.
* First we determine the number of iterations to unroll.
slang_ir_node *n, *root = NULL;
slang_atom varId;
+ /* Set flag so code generator knows we're unrolling loops */
+ _unroll_loop_inc( A );
+
if (oper->children[0].type == SLANG_OPER_BLOCK_NO_NEW_SCOPE) {
/* for (int i=0; ... */
slang_variable *var;
/* make a copy of the loop body */
body = slang_operation_new(1);
- if (!body)
+ if (!body) {
+ _unroll_loop_dec( A );
return NULL;
+ }
- if (!slang_operation_copy(body, &oper->children[3]))
+ if (!slang_operation_copy(body, &oper->children[3])) {
+ _unroll_loop_dec( A );
return NULL;
+ }
/* in body, replace instances of 'varId' with literal 'iter' */
{
if (!oldVar) {
/* undeclared loop variable */
slang_operation_delete(body);
+ _unroll_loop_dec( A );
return NULL;
}
/* do IR codegen for body */
n = _slang_gen_operation(A, body);
+ if (!n) {
+ _unroll_loop_dec( A );
+ return NULL;
+ }
+
root = new_seq(root, n);
slang_operation_delete(body);
}
+ _unroll_loop_dec( A );
+
return root;
}
if (is_operation_type(&oper->children[1], SLANG_OPER_BREAK)
&& !haveElseClause) {
/* Special case: generate a conditional break */
+ if (!A->CurLoop && A->UnrollLoop) /* trying to unroll */
+ return NULL;
ifBody = new_break_if_true(A->CurLoop, cond);
return ifBody;
}
else if (is_operation_type(&oper->children[1], SLANG_OPER_CONTINUE)
&& !haveElseClause) {
- /* Special case: generate a conditional break */
+ /* Special case: generate a conditional continue */
+ if (!A->CurLoop && A->UnrollLoop) /* trying to unroll */
+ return NULL;
ifBody = new_cont_if_true(A->CurLoop, cond);
return ifBody;
}
else {
/* general case */
ifBody = _slang_gen_operation(A, &oper->children[1]);
+ if (!ifBody)
+ return NULL;
if (haveElseClause)
elseBody = _slang_gen_operation(A, &oper->children[2]);
else
slang_ir_node *varDecl, *n;
slang_ir_storage *store;
GLint arrayLen, size, totalSize; /* if array then totalSize > size */
- enum register_file file;
+ gl_register_file file;
/*assert(!var->declared);*/
var->declared = GL_TRUE;
file = PROGRAM_TEMPORARY;
}
- totalSize = size = _slang_sizeof_type_specifier(&var->type.specifier);
+ size = _slang_sizeof_type_specifier(&var->type.specifier);
if (size <= 0) {
slang_info_log_error(A->log, "invalid declaration for '%s'", varName);
return NULL;
if (!varDecl)
return NULL;
- _slang_attach_storage(varDecl, var); /* undefined storage at first */
- assert(var->store);
- assert(varDecl->Store == var->store);
- assert(varDecl->Store);
- assert(varDecl->Store->Index < 0);
- store = var->store;
+ /* Allocate slang_ir_storage for this variable if needed.
+ * Note that we may not actually allocate a constant or temporary register
+ * until later.
+ */
+ if (!var->store) {
+ GLint index = -7; /* TBD / unknown */
+ var->store = _slang_new_ir_storage(file, index, totalSize);
+ if (!var->store)
+ return NULL; /* out of memory */
+ }
- assert(store == varDecl->Store);
+ /* set the IR node's Var and Store pointers */
+ varDecl->Var = var;
+ varDecl->Store = var->store;
- /* Fill in storage fields which we now know. store->Index/Swizzle may be
- * set for some cases below. Otherwise, store->Index/Swizzle will be set
- * during code emit.
- */
- store->File = file;
- store->Size = totalSize;
+ store = var->store;
/* if there's an initializer, generate IR for the expression */
if (initializer) {
if (lhs && rhs) {
/* convert lhs swizzle into writemask */
const GLuint swizzle = root_swizzle(lhs->Store);
- GLuint writemask, newSwizzle;
+ GLuint writemask, newSwizzle = 0x0;
if (!swizzle_to_writemask(A, swizzle, &writemask, &newSwizzle)) {
/* Non-simple writemask, need to swizzle right hand side in
* order to put components into the right place.
return _slang_gen_while(A, oper);
case SLANG_OPER_BREAK:
if (!A->CurLoop) {
- slang_info_log_error(A->log, "'break' not in loop");
+ if (!A->UnrollLoop)
+ slang_info_log_error(A->log, "'break' not in loop");
return NULL;
}
return new_break(A->CurLoop);
case SLANG_OPER_CONTINUE:
if (!A->CurLoop) {
- slang_info_log_error(A->log, "'continue' not in loop");
+ if (!A->UnrollLoop)
+ slang_info_log_error(A->log, "'continue' not in loop");
return NULL;
}
return _slang_gen_continue(A, oper);
}
+/**
+ * Check if the given type specifier is a rectangular texture sampler.
+ */
+static GLboolean
+is_rect_sampler_spec(const slang_type_specifier *spec)
+{
+ while (spec->_array) {
+ spec = spec->_array;
+ }
+ return spec->type == SLANG_SPEC_SAMPLER2DRECT ||
+ spec->type == SLANG_SPEC_SAMPLER2DRECTSHADOW;
+}
+
+
+
/**
* Called by compiler when a global variable has been parsed/compiled.
* Here we examine the variable's type to determine what kind of register
slang_ir_storage *store = NULL;
int dbg = 0;
const GLenum datatype = _slang_gltype_from_specifier(&var->type.specifier);
- const GLint texIndex = sampler_to_texture_index(var->type.specifier.type);
const GLint size = _slang_sizeof_type_specifier(&var->type.specifier);
const GLint arrayLen = _slang_array_length(var);
const GLint totalSize = _slang_array_size(size, arrayLen);
+ GLint texIndex = sampler_to_texture_index(var->type.specifier.type);
+
+ /* check for sampler2D arrays */
+ if (texIndex == -1 && var->type.specifier._array)
+ texIndex = sampler_to_texture_index(var->type.specifier._array->type);
if (texIndex != -1) {
/* This is a texture sampler variable...
}
#if FEATURE_es2_glsl /* XXX should use FEATURE_texture_rect */
/* disallow rect samplers */
- if (var->type.specifier.type == SLANG_SPEC_SAMPLER2DRECT ||
- var->type.specifier.type == SLANG_SPEC_SAMPLER2DRECTSHADOW) {
+ if (is_rect_sampler_spec(&var->type.specifier)) {
slang_info_log_error(A->log, "invalid sampler type for '%s'", varName);
return GL_FALSE;
}
+#else
+ (void) is_rect_sampler_spec; /* silence warning */
#endif
{
GLint sampNum = _mesa_add_sampler(prog->Parameters, varName, datatype);
- store = _slang_new_ir_storage(PROGRAM_SAMPLER, sampNum, texIndex);
+ store = _slang_new_ir_storage_sampler(sampNum, texIndex, totalSize);
+
+ /* If we have a sampler array, then we need to allocate the
+ * additional samplers to ensure we don't allocate them elsewhere.
+ * We can't directly use _mesa_add_sampler() as that checks the
+ * varName and gets a match, so we call _mesa_add_parameter()
+ * directly and use the last sampler number from the call above.
+ */
+ if (arrayLen > 0) {
+ GLint a = arrayLen - 1;
+ GLint i;
+ for (i = 0; i < a; i++) {
+ GLfloat value = (GLfloat)(i + sampNum + 1);
+ (void) _mesa_add_parameter(prog->Parameters, PROGRAM_SAMPLER,
+ varName, 1, datatype, &value, NULL, 0x0);
+ }
+ }
}
if (dbg) printf("SAMPLER ");
}
if (prog) {
/* user-defined uniform */
if (datatype == GL_NONE) {
- if (var->type.specifier.type == SLANG_SPEC_STRUCT) {
+ if ((var->type.specifier.type == SLANG_SPEC_ARRAY &&
+ var->type.specifier._array->type == SLANG_SPEC_STRUCT) ||
+ (var->type.specifier.type == SLANG_SPEC_STRUCT)) {
/* temporary work-around */
GLenum datatype = GL_FLOAT;
GLint uniformLoc = _mesa_add_uniform(prog->Parameters, varName,
totalSize, datatype, NULL);
store = _slang_new_ir_storage_swz(PROGRAM_UNIFORM, uniformLoc,
totalSize, swizzle);
+
+ if (arrayLen > 0) {
+ GLint a = arrayLen - 1;
+ GLint i;
+ for (i = 0; i < a; i++) {
+ GLfloat value = (GLfloat)(i + uniformLoc + 1);
+ (void) _mesa_add_parameter(prog->Parameters, PROGRAM_UNIFORM,
+ varName, 1, datatype, &value, NULL, 0x0);
+ }
+ }
/* XXX what we need to do is unroll the struct into its
* basic types, creating a uniform variable for each.
n = _slang_gen_var_decl(A, var, var->initializer);
/* emit GPU instructions */
- success = _slang_emit_code(n, A->vartable, A->program, GL_FALSE, A->log);
+ success = _slang_emit_code(n, A->vartable, A->program, A->pragmas, GL_FALSE, A->log);
_slang_free_ir_tree(n);
}
#endif
/* Emit program instructions */
- success = _slang_emit_code(n, A->vartable, A->program, GL_TRUE, A->log);
+ success = _slang_emit_code(n, A->vartable, A->program, A->pragmas, GL_TRUE, A->log);
_slang_free_ir_tree(n);
/* free codegen context */