#include "program/prog_optimize.h"
#include "program/prog_print.h"
#include "program/program.h"
-#include "program/prog_uniform.h"
#include "program/prog_parameter.h"
#include "program/sampler.h"
(1 << PROGRAM_CONSTANT) | \
(1 << PROGRAM_UNIFORM))
+/**
+ * Maximum number of temporary registers.
+ *
+ * It is too big for stack allocated arrays -- it will cause stack overflow on
+ * Windows and likely Mac OS X.
+ */
#define MAX_TEMPS 4096
/* will be 4 for GLSL 4.00 */
/** List of immediate_storage */
exec_list immediates;
- int num_immediates;
+ unsigned num_immediates;
/** List of function_entry */
exec_list function_signatures;
bool process_move_condition(ir_rvalue *ir);
- void remove_output_reads(gl_register_file type);
void simplify_cmp(void);
void rename_temp_register(int index, int new_index);
case3(SLT, ISLT, USLT);
case2iu(ISHR, USHR);
+
+ case2fi(SSG, ISSG);
+ case3(ABS, IABS, IABS);
default: break;
}
fp->OriginUpperLeft = ir->origin_upper_left;
fp->PixelCenterInteger = ir->pixel_center_integer;
-
- } else if (strcmp(ir->name, "gl_FragDepth") == 0) {
- struct gl_fragment_program *fp = (struct gl_fragment_program *)this->prog;
- switch (ir->depth_layout) {
- case ir_depth_layout_none:
- fp->FragDepthLayout = FRAG_DEPTH_LAYOUT_NONE;
- break;
- case ir_depth_layout_any:
- fp->FragDepthLayout = FRAG_DEPTH_LAYOUT_ANY;
- break;
- case ir_depth_layout_greater:
- fp->FragDepthLayout = FRAG_DEPTH_LAYOUT_GREATER;
- break;
- case ir_depth_layout_less:
- fp->FragDepthLayout = FRAG_DEPTH_LAYOUT_LESS;
- break;
- case ir_depth_layout_unchanged:
- fp->FragDepthLayout = FRAG_DEPTH_LAYOUT_UNCHANGED;
- break;
- default:
- assert(0);
- break;
- }
}
if (ir->mode == ir_var_uniform && strncmp(ir->name, "gl_", 3) == 0) {
}
break;
case ir_unop_neg:
- assert(result_dst.type == GLSL_TYPE_FLOAT || result_dst.type == GLSL_TYPE_INT);
- if (result_dst.type == GLSL_TYPE_INT)
+ if (result_dst.type == GLSL_TYPE_INT || result_dst.type == GLSL_TYPE_UINT)
emit(ir, TGSI_OPCODE_INEG, result_dst, op[0]);
else {
op[0].negate = ~op[0].negate;
}
break;
case ir_unop_abs:
- assert(result_dst.type == GLSL_TYPE_FLOAT);
emit(ir, TGSI_OPCODE_ABS, result_dst, op[0]);
break;
case ir_unop_sign:
emit(ir, TGSI_OPCODE_DDX, result_dst, op[0]);
break;
case ir_unop_dFdy:
- op[0].negate = ~op[0].negate;
- emit(ir, TGSI_OPCODE_DDY, result_dst, op[0]);
+ {
+ /* The X component contains 1 or -1 depending on whether the framebuffer
+ * is a FBO or the window system buffer, respectively.
+ * It is then multiplied with the source operand of DDY.
+ */
+ static const gl_state_index transform_y_state[STATE_LENGTH]
+ = { STATE_INTERNAL, STATE_FB_WPOS_Y_TRANSFORM };
+
+ unsigned transform_y_index =
+ _mesa_add_state_reference(this->prog->Parameters,
+ transform_y_state);
+
+ st_src_reg transform_y = st_src_reg(PROGRAM_STATE_VAR,
+ transform_y_index,
+ glsl_type::vec4_type);
+ transform_y.swizzle = SWIZZLE_XXXX;
+
+ st_src_reg temp = get_temp(glsl_type::vec4_type);
+
+ emit(ir, TGSI_OPCODE_MUL, st_dst_reg(temp), transform_y, op[0]);
+ emit(ir, TGSI_OPCODE_DDY, result_dst, temp);
break;
+ }
case ir_unop_noise: {
/* At some point, a motivated person could add a better
else
emit(ir, TGSI_OPCODE_TRUNC, result_dst, op[0]);
break;
+ case ir_unop_f2u:
+ if (native_integers)
+ emit(ir, TGSI_OPCODE_F2U, result_dst, op[0]);
+ else
+ emit(ir, TGSI_OPCODE_TRUNC, result_dst, op[0]);
+ break;
+ case ir_unop_bitcast_f2i:
+ case ir_unop_bitcast_f2u:
+ case ir_unop_bitcast_i2f:
+ case ir_unop_bitcast_u2f:
+ result_src = op[0];
+ break;
case ir_unop_f2b:
emit(ir, TGSI_OPCODE_SNE, result_dst, op[0], st_src_reg_for_float(0.0));
break;
emit(ir, TGSI_OPCODE_TRUNC, result_dst, op[0]);
break;
case ir_unop_ceil:
- op[0].negate = ~op[0].negate;
- emit(ir, TGSI_OPCODE_FLR, result_dst, op[0]);
- result_src.negate = ~result_src.negate;
+ emit(ir, TGSI_OPCODE_CEIL, result_dst, op[0]);
break;
case ir_unop_floor:
emit(ir, TGSI_OPCODE_FLR, result_dst, op[0]);
break;
+ case ir_unop_round_even:
+ emit(ir, TGSI_OPCODE_ROUND, result_dst, op[0]);
+ break;
case ir_unop_fract:
emit(ir, TGSI_OPCODE_FRC, result_dst, op[0]);
break;
}
case ir_binop_lshift:
if (native_integers) {
- emit(ir, TGSI_OPCODE_SHL, result_dst, op[0]);
+ emit(ir, TGSI_OPCODE_SHL, result_dst, op[0], op[1]);
break;
}
case ir_binop_rshift:
if (native_integers) {
- emit(ir, TGSI_OPCODE_ISHR, result_dst, op[0]);
+ emit(ir, TGSI_OPCODE_ISHR, result_dst, op[0], op[1]);
break;
}
case ir_binop_bit_and:
if (native_integers) {
- emit(ir, TGSI_OPCODE_AND, result_dst, op[0]);
+ emit(ir, TGSI_OPCODE_AND, result_dst, op[0], op[1]);
break;
}
case ir_binop_bit_xor:
if (native_integers) {
- emit(ir, TGSI_OPCODE_XOR, result_dst, op[0]);
+ emit(ir, TGSI_OPCODE_XOR, result_dst, op[0], op[1]);
break;
}
case ir_binop_bit_or:
if (native_integers) {
- emit(ir, TGSI_OPCODE_OR, result_dst, op[0]);
+ emit(ir, TGSI_OPCODE_OR, result_dst, op[0], op[1]);
break;
}
- case ir_unop_round_even:
+
assert(!"GLSL 1.30 features unsupported");
break;
+ case ir_binop_ubo_load:
+ assert(!"not yet supported");
+ break;
+
case ir_quadop_vector:
/* This operation should have already been handled.
*/
assert(var->location != -1);
entry = new(mem_ctx) variable_storage(var,
PROGRAM_OUTPUT,
- var->location);
+ var->location + var->index);
break;
case ir_var_system_value:
entry = new(mem_ctx) variable_storage(var,
gl_type = native_integers ? GL_BOOL : GL_FLOAT;
for (i = 0; i < ir->type->vector_elements; i++) {
if (native_integers)
- values[i].b = ir->value.b[i];
+ values[i].u = ir->value.b[i] ? ~0 : 0;
else
values[i].f = ir->value.b[i];
}
glsl_to_tgsi_visitor::visit(ir_call *ir)
{
glsl_to_tgsi_instruction *call_inst;
- ir_function_signature *sig = ir->get_callee();
+ ir_function_signature *sig = ir->callee;
function_entry *entry = get_function_signature(sig);
int i;
ir->shadow_comparitor->accept(this);
/* XXX This will need to be updated for cubemap array samplers. */
- if (sampler_type->sampler_dimensionality == GLSL_SAMPLER_DIM_2D &&
- sampler_type->sampler_array) {
+ if ((sampler_type->sampler_dimensionality == GLSL_SAMPLER_DIM_2D &&
+ sampler_type->sampler_array) ||
+ sampler_type->sampler_dimensionality == GLSL_SAMPLER_DIM_CUBE) {
coord_dst.writemask = WRITEMASK_W;
} else {
coord_dst.writemask = WRITEMASK_Z;
case GLSL_SAMPLER_DIM_BUF:
assert(!"FINISHME: Implement ARB_texture_buffer_object");
break;
+ case GLSL_SAMPLER_DIM_EXTERNAL:
+ inst->tex_target = TEXTURE_EXTERNAL_INDEX;
+ break;
default:
assert(!"Should not get here.");
}
void
glsl_to_tgsi_visitor::visit(ir_discard *ir)
{
- struct gl_fragment_program *fp = (struct gl_fragment_program *)this->prog;
-
if (ir->condition) {
ir->condition->accept(this);
this->result.negate = ~this->result.negate;
} else {
emit(ir, TGSI_OPCODE_KILP);
}
-
- fp->UsesKill = GL_TRUE;
}
void
num_immediates = 0;
current_function = NULL;
num_address_regs = 0;
+ samplers_used = 0;
indirect_addr_temps = false;
indirect_addr_consts = false;
+ glsl_version = 0;
+ native_integers = false;
mem_ctx = ralloc_context(NULL);
+ ctx = NULL;
+ prog = NULL;
+ shader_program = NULL;
+ options = NULL;
}
glsl_to_tgsi_visitor::~glsl_to_tgsi_visitor()
if (is_tex_instruction(inst->op)) {
v->samplers_used |= 1 << inst->sampler;
- prog->SamplerTargets[inst->sampler] =
- (gl_texture_index)inst->tex_target;
if (inst->tex_shadow) {
prog->ShadowSamplers |= 1 << inst->sampler;
}
}
prog->SamplersUsed = v->samplers_used;
- _mesa_update_shader_textures_used(prog);
-}
-
-/**
- * Check if the given vertex/fragment/shader program is within the
- * resource limits of the context (number of texture units, etc).
- * If any of those checks fail, record a linker error.
- *
- * XXX more checks are needed...
- */
-static void
-check_resources(const struct gl_context *ctx,
- struct gl_shader_program *shader_program,
- glsl_to_tgsi_visitor *prog,
- struct gl_program *proginfo)
-{
- switch (proginfo->Target) {
- case GL_VERTEX_PROGRAM_ARB:
- if (_mesa_bitcount(prog->samplers_used) >
- ctx->Const.MaxVertexTextureImageUnits) {
- fail_link(shader_program, "Too many vertex shader texture samplers");
- }
- if (proginfo->Parameters->NumParameters > MAX_UNIFORMS) {
- fail_link(shader_program, "Too many vertex shader constants");
- }
- break;
- case MESA_GEOMETRY_PROGRAM:
- if (_mesa_bitcount(prog->samplers_used) >
- ctx->Const.MaxGeometryTextureImageUnits) {
- fail_link(shader_program, "Too many geometry shader texture samplers");
- }
- if (proginfo->Parameters->NumParameters >
- MAX_GEOMETRY_UNIFORM_COMPONENTS / 4) {
- fail_link(shader_program, "Too many geometry shader constants");
- }
- break;
- case GL_FRAGMENT_PROGRAM_ARB:
- if (_mesa_bitcount(prog->samplers_used) >
- ctx->Const.MaxTextureImageUnits) {
- fail_link(shader_program, "Too many fragment shader texture samplers");
- }
- if (proginfo->Parameters->NumParameters > MAX_UNIFORMS) {
- fail_link(shader_program, "Too many fragment shader constants");
- }
- break;
- default:
- _mesa_problem(ctx, "unexpected program type in check_resources()");
- }
+ if (v->shader_program != NULL)
+ _mesa_update_shader_textures_used(v->shader_program, prog);
}
-
static void
set_uniform_initializer(struct gl_context *ctx, void *mem_ctx,
struct gl_shader_program *shader_program,
return;
}
- int loc = _mesa_get_uniform_location(ctx, shader_program, name);
-
- if (loc == -1) {
+ unsigned offset;
+ unsigned index = _mesa_get_uniform_location(ctx, shader_program, name,
+ &offset);
+ if (offset == GL_INVALID_INDEX) {
fail_link(shader_program,
"Couldn't find uniform for initializer %s\n", name);
return;
}
+ int loc = _mesa_uniform_merge_location_offset(index, offset);
for (unsigned int i = 0; i < (type->is_array() ? type->length : 1); i++) {
ir_constant *element;
element_type->matrix_columns,
element_type->vector_elements,
loc, 1, GL_FALSE, (GLfloat *)values);
- loc += element_type->matrix_columns;
} else {
_mesa_uniform(ctx, shader_program, loc, element_type->matrix_columns,
values, element_type->gl_type);
- loc += type_size(element_type);
}
- }
-}
-
-/*
- * Scan/rewrite program to remove reads of custom (output) registers.
- * The passed type has to be either PROGRAM_OUTPUT or PROGRAM_VARYING
- * (for vertex shaders).
- * In GLSL shaders, varying vars can be read and written.
- * On some hardware, trying to read an output register causes trouble.
- * So, rewrite the program to use a temporary register in this case.
- *
- * Based on _mesa_remove_output_reads from programopt.c.
- */
-void
-glsl_to_tgsi_visitor::remove_output_reads(gl_register_file type)
-{
- GLuint i;
- GLint outputMap[VERT_RESULT_MAX];
- GLint outputTypes[VERT_RESULT_MAX];
- GLuint numVaryingReads = 0;
- GLboolean usedTemps[MAX_TEMPS];
- GLuint firstTemp = 0;
-
- _mesa_find_used_registers(prog, PROGRAM_TEMPORARY,
- usedTemps, MAX_TEMPS);
-
- assert(type == PROGRAM_VARYING || type == PROGRAM_OUTPUT);
- assert(prog->Target == GL_VERTEX_PROGRAM_ARB || type != PROGRAM_VARYING);
-
- for (i = 0; i < VERT_RESULT_MAX; i++)
- outputMap[i] = -1;
- /* look for instructions which read from varying vars */
- foreach_iter(exec_list_iterator, iter, this->instructions) {
- glsl_to_tgsi_instruction *inst = (glsl_to_tgsi_instruction *)iter.get();
- const GLuint numSrc = num_inst_src_regs(inst->op);
- GLuint j;
- for (j = 0; j < numSrc; j++) {
- if (inst->src[j].file == type) {
- /* replace the read with a temp reg */
- const GLuint var = inst->src[j].index;
- if (outputMap[var] == -1) {
- numVaryingReads++;
- outputMap[var] = _mesa_find_free_register(usedTemps,
- MAX_TEMPS,
- firstTemp);
- outputTypes[var] = inst->src[j].type;
- firstTemp = outputMap[var] + 1;
- }
- inst->src[j].file = PROGRAM_TEMPORARY;
- inst->src[j].index = outputMap[var];
- }
- }
- }
-
- if (numVaryingReads == 0)
- return; /* nothing to be done */
-
- /* look for instructions which write to the varying vars identified above */
- foreach_iter(exec_list_iterator, iter, this->instructions) {
- glsl_to_tgsi_instruction *inst = (glsl_to_tgsi_instruction *)iter.get();
- if (inst->dst.file == type && outputMap[inst->dst.index] >= 0) {
- /* change inst to write to the temp reg, instead of the varying */
- inst->dst.file = PROGRAM_TEMPORARY;
- inst->dst.index = outputMap[inst->dst.index];
- }
- }
-
- /* insert new MOV instructions at the end */
- for (i = 0; i < VERT_RESULT_MAX; i++) {
- if (outputMap[i] >= 0) {
- /* MOV VAR[i], TEMP[tmp]; */
- st_src_reg src = st_src_reg(PROGRAM_TEMPORARY, outputMap[i], outputTypes[i]);
- st_dst_reg dst = st_dst_reg(type, WRITEMASK_XYZW, outputTypes[i]);
- dst.index = i;
- this->emit(NULL, TGSI_OPCODE_MOV, dst, src);
- }
+ loc++;
}
}
void
glsl_to_tgsi_visitor::simplify_cmp(void)
{
- unsigned tempWrites[MAX_TEMPS];
+ unsigned *tempWrites;
unsigned outputWrites[MAX_PROGRAM_OUTPUTS];
- memset(tempWrites, 0, sizeof(tempWrites));
+ tempWrites = new unsigned[MAX_TEMPS];
+ if (!tempWrites) {
+ return;
+ }
+ memset(tempWrites, 0, sizeof(unsigned) * MAX_TEMPS);
memset(outputWrites, 0, sizeof(outputWrites));
foreach_iter(exec_list_iterator, iter, this->instructions) {
inst->op == TGSI_OPCODE_END ||
inst->op == TGSI_OPCODE_ENDSUB ||
inst->op == TGSI_OPCODE_RET) {
- return;
+ break;
}
if (inst->dst.file == PROGRAM_OUTPUT) {
inst->src[0] = inst->src[1];
}
}
+
+ delete [] tempWrites;
}
/* Replaces all references to a temporary register index with another index. */
switch (inst->op) {
case TGSI_OPCODE_BGNLOOP:
case TGSI_OPCODE_ENDLOOP:
+ case TGSI_OPCODE_CONT:
+ case TGSI_OPCODE_BRK:
/* End of a basic block, clear the write array entirely.
- * FIXME: This keeps us from killing dead code when the writes are
+ *
+ * This keeps us from killing dead code when the writes are
* on either side of a loop, even when the register isn't touched
- * inside the loop.
+ * inside the loop. However, glsl_to_tgsi_visitor doesn't seem to emit
+ * dead code of this type, so it shouldn't make a difference as long as
+ * the dead code elimination pass in the GLSL compiler does its job.
*/
memset(writes, 0, sizeof(*writes) * this->next_temp * 4);
break;
case TGSI_OPCODE_ENDIF:
- --level;
- break;
-
case TGSI_OPCODE_ELSE:
- /* Clear all channels written inside the preceding if block from the
- * write array, but leave those that were not touched.
- *
- * FIXME: This destroys opportunities to remove dead code inside of
- * IF blocks that are followed by an ELSE block.
+ /* Promote the recorded level of all channels written inside the
+ * preceding if or else block to the level above the if/else block.
*/
for (int r = 0; r < this->next_temp; r++) {
for (int c = 0; c < 4; c++) {
if (!writes[4 * r + c])
continue;
- if (write_level[4 * r + c] >= level)
- writes[4 * r + c] = NULL;
+ if (write_level[4 * r + c] == level)
+ write_level[4 * r + c] = level-1;
}
}
+
+ if(inst->op == TGSI_OPCODE_ENDIF)
+ --level;
+
break;
case TGSI_OPCODE_IF:
/* Copy attributes of the glsl_to_tgsi_visitor in the original shader. */
v->ctx = original->ctx;
v->prog = prog;
+ v->shader_program = NULL;
v->glsl_version = original->glsl_version;
v->native_integers = original->native_integers;
v->options = original->options;
v->indirect_addr_temps = original->indirect_addr_temps;
v->indirect_addr_consts = original->indirect_addr_consts;
memcpy(&v->immediates, &original->immediates, sizeof(v->immediates));
+ v->num_immediates = original->num_immediates;
/*
* Get initial pixel color from the texture.
inst->sampler = 0;
inst->tex_target = TEXTURE_2D_INDEX;
- prog->InputsRead |= (1 << FRAG_ATTRIB_TEX0);
+ prog->InputsRead |= FRAG_BIT_TEX0;
prog->SamplersUsed |= (1 << 0); /* mark sampler 0 as used */
v->samplers_used |= (1 << 0);
* new visitor. */
foreach_iter(exec_list_iterator, iter, original->instructions) {
glsl_to_tgsi_instruction *inst = (glsl_to_tgsi_instruction *)iter.get();
+ glsl_to_tgsi_instruction *newinst;
st_src_reg src_regs[3];
if (inst->dst.file == PROGRAM_OUTPUT)
src_regs[i].index = src0.index;
}
else if (src_regs[i].file == PROGRAM_INPUT)
- prog->InputsRead |= (1 << src_regs[i].index);
+ prog->InputsRead |= BITFIELD64_BIT(src_regs[i].index);
}
- v->emit(NULL, inst->op, inst->dst, src_regs[0], src_regs[1], src_regs[2]);
+ newinst = v->emit(NULL, inst->op, inst->dst, src_regs[0], src_regs[1], src_regs[2]);
+ newinst->tex_target = inst->tex_target;
}
/* Make modifications to fragment program info. */
/* Copy attributes of the glsl_to_tgsi_visitor in the original shader. */
v->ctx = original->ctx;
v->prog = prog;
+ v->shader_program = NULL;
v->glsl_version = original->glsl_version;
v->native_integers = original->native_integers;
v->options = original->options;
v->indirect_addr_temps = original->indirect_addr_temps;
v->indirect_addr_consts = original->indirect_addr_consts;
memcpy(&v->immediates, &original->immediates, sizeof(v->immediates));
+ v->num_immediates = original->num_immediates;
/* TEX tmp0, fragment.texcoord[0], texture[0], 2D; */
coord = st_src_reg(PROGRAM_INPUT, FRAG_ATTRIB_TEX0, glsl_type::vec2_type);
inst->sampler = samplerIndex;
inst->tex_target = TEXTURE_2D_INDEX;
- prog->InputsRead |= (1 << FRAG_ATTRIB_TEX0);
+ prog->InputsRead |= FRAG_BIT_TEX0;
prog->SamplersUsed |= (1 << samplerIndex); /* mark sampler as used */
v->samplers_used |= (1 << samplerIndex);
* new visitor. */
foreach_iter(exec_list_iterator, iter, original->instructions) {
glsl_to_tgsi_instruction *inst = (glsl_to_tgsi_instruction *)iter.get();
+ glsl_to_tgsi_instruction *newinst;
st_src_reg src_regs[3];
if (inst->dst.file == PROGRAM_OUTPUT)
for (int i=0; i<3; i++) {
src_regs[i] = inst->src[i];
if (src_regs[i].file == PROGRAM_INPUT)
- prog->InputsRead |= (1 << src_regs[i].index);
+ prog->InputsRead |= BITFIELD64_BIT(src_regs[i].index);
}
- v->emit(NULL, inst->op, inst->dst, src_regs[0], src_regs[1], src_regs[2]);
+ newinst = v->emit(NULL, inst->op, inst->dst, src_regs[0], src_regs[1], src_regs[2]);
+ newinst->tex_target = inst->tex_target;
}
/* Make modifications to fragment program info. */
struct ureg_src samplers[PIPE_MAX_SAMPLERS];
struct ureg_src systemValues[SYSTEM_VALUE_MAX];
- /* Extra info for handling point size clamping in vertex shader */
- struct ureg_dst pointSizeResult; /**< Actual point size output register */
- struct ureg_src pointSizeConst; /**< Point size range constant register */
- GLint pointSizeOutIndex; /**< Temp point size output register */
- GLboolean prevInstWrotePointSize;
-
const GLuint *inputMapping;
const GLuint *outputMapping;
/** Map Mesa's SYSTEM_VALUE_x to TGSI_SEMANTIC_x */
static unsigned mesa_sysval_to_semantic[SYSTEM_VALUE_MAX] = {
TGSI_SEMANTIC_FACE,
+ TGSI_SEMANTIC_VERTEXID,
TGSI_SEMANTIC_INSTANCEID
};
case PROGRAM_TEMPORARY:
if (ureg_dst_is_undef(t->temps[index]))
- t->temps[index] = ureg_DECL_temporary(t->ureg);
+ t->temps[index] = ureg_DECL_local_temporary(t->ureg);
return t->temps[index];
case PROGRAM_OUTPUT:
- if (t->procType == TGSI_PROCESSOR_VERTEX && index == VERT_RESULT_PSIZ)
- t->prevInstWrotePointSize = GL_TRUE;
-
if (t->procType == TGSI_PROCESSOR_VERTEX)
assert(index < VERT_RESULT_MAX);
else if (t->procType == TGSI_PROCESSOR_FRAGMENT)
static struct ureg_src
src_register(struct st_translate *t,
gl_register_file file,
- GLuint index)
+ GLint index)
{
switch(file) {
case PROGRAM_UNDEFINED:
case PROGRAM_TEMPORARY:
assert(index >= 0);
- assert(index < Elements(t->temps));
+ assert(index < (int) Elements(t->temps));
if (ureg_dst_is_undef(t->temps[index]))
- t->temps[index] = ureg_DECL_temporary(t->ureg);
+ t->temps[index] = ureg_DECL_local_temporary(t->ureg);
return ureg_src(t->temps[index]);
case PROGRAM_NAMED_PARAM:
return ureg_src(t->address[index]);
case PROGRAM_SYSTEM_VALUE:
- assert(index < Elements(t->systemValues));
+ assert(index < (int) Elements(t->systemValues));
return t->systemValues[index];
default:
static struct ureg_dst
translate_dst(struct st_translate *t,
const st_dst_reg *dst_reg,
- bool saturate)
+ bool saturate, bool clamp_color)
{
struct ureg_dst dst = dst_register(t,
dst_reg->file,
if (saturate)
dst = ureg_saturate(dst);
+ else if (clamp_color && dst_reg->file == PROGRAM_OUTPUT) {
+ /* Clamp colors for ARB_color_buffer_float. */
+ switch (t->procType) {
+ case TGSI_PROCESSOR_VERTEX:
+ /* XXX if the geometry shader is present, this must be done there
+ * instead of here. */
+ if (dst_reg->index == VERT_RESULT_COL0 ||
+ dst_reg->index == VERT_RESULT_COL1 ||
+ dst_reg->index == VERT_RESULT_BFC0 ||
+ dst_reg->index == VERT_RESULT_BFC1) {
+ dst = ureg_saturate(dst);
+ }
+ break;
+
+ case TGSI_PROCESSOR_FRAGMENT:
+ if (dst_reg->index >= FRAG_RESULT_COLOR) {
+ dst = ureg_saturate(dst);
+ }
+ break;
+ }
+ }
if (dst_reg->reladdr != NULL)
dst = ureg_dst_indirect(dst, ureg_src(t->address[0]));
offset.SwizzleX = in_offset->SwizzleX;
offset.SwizzleY = in_offset->SwizzleY;
offset.SwizzleZ = in_offset->SwizzleZ;
+ offset.Padding = 0;
return offset;
}
static void
compile_tgsi_instruction(struct st_translate *t,
- const glsl_to_tgsi_instruction *inst)
+ const glsl_to_tgsi_instruction *inst,
+ bool clamp_dst_color_output)
{
struct ureg_program *ureg = t->ureg;
GLuint i;
if (num_dst)
dst[0] = translate_dst(t,
&inst->dst,
- inst->saturate);
+ inst->saturate,
+ clamp_dst_color_output);
for (i = 0; i < num_src; i++)
src[i] = translate_src(t, &inst->src[i]);
ureg_tex_insn(ureg,
inst->op,
dst, num_dst,
- translate_texture_target(inst->tex_target, inst->tex_shadow),
+ st_translate_texture_target(inst->tex_target, inst->tex_shadow),
texoffsets, inst->tex_offset_num_offset,
src, num_src);
return;
* or not, which is determined by testing against the inversion
* state variable used below, which will be either +1 or -1.
*/
- struct ureg_dst adj_temp = ureg_DECL_temporary(ureg);
+ struct ureg_dst adj_temp = ureg_DECL_local_temporary(ureg);
ureg_CMP(ureg, adj_temp,
ureg_scalar(wpostrans, invert ? 2 : 0),
const ubyte inputSemanticName[],
const ubyte inputSemanticIndex[],
const GLuint interpMode[],
+ const GLboolean is_centroid[],
GLuint numOutputs,
const GLuint outputMapping[],
const ubyte outputSemanticName[],
const ubyte outputSemanticIndex[],
- boolean passthrough_edgeflags)
+ boolean passthrough_edgeflags,
+ boolean clamp_color)
{
- struct st_translate translate, *t;
+ struct st_translate *t;
unsigned i;
enum pipe_error ret = PIPE_OK;
assert(numInputs <= Elements(t->inputs));
assert(numOutputs <= Elements(t->outputs));
- t = &translate;
+ t = CALLOC_STRUCT(st_translate);
+ if (!t) {
+ ret = PIPE_ERROR_OUT_OF_MEMORY;
+ goto out;
+ }
+
memset(t, 0, sizeof *t);
t->procType = procType;
t->inputMapping = inputMapping;
t->outputMapping = outputMapping;
t->ureg = ureg;
- t->pointSizeOutIndex = -1;
- t->prevInstWrotePointSize = GL_FALSE;
+
+ if (program->shader_program) {
+ for (i = 0; i < program->shader_program->NumUserUniformStorage; i++) {
+ struct gl_uniform_storage *const storage =
+ &program->shader_program->UniformStorage[i];
+
+ _mesa_uniform_detach_all_driver_storage(storage);
+ }
+ }
/*
* Declare input attributes.
*/
if (procType == TGSI_PROCESSOR_FRAGMENT) {
for (i = 0; i < numInputs; i++) {
- t->inputs[i] = ureg_DECL_fs_input(ureg,
- inputSemanticName[i],
- inputSemanticIndex[i],
- interpMode[i]);
+ t->inputs[i] = ureg_DECL_fs_input_cyl_centroid(ureg,
+ inputSemanticName[i],
+ inputSemanticIndex[i],
+ interpMode[i], 0,
+ is_centroid[i]);
}
if (proginfo->InputsRead & FRAG_BIT_WPOS) {
break;
default:
assert(!"fragment shader outputs must be POSITION/STENCIL/COLOR");
- return PIPE_ERROR_BAD_INPUT;
+ ret = PIPE_ERROR_BAD_INPUT;
+ goto out;
}
}
}
t->outputs[i] = ureg_DECL_output(ureg,
outputSemanticName[i],
outputSemanticIndex[i]);
- if ((outputSemanticName[i] == TGSI_SEMANTIC_PSIZE) && proginfo->Id) {
- /* Writing to the point size result register requires special
- * handling to implement clamping.
- */
- static const gl_state_index pointSizeClampState[STATE_LENGTH]
- = { STATE_INTERNAL, STATE_POINT_SIZE_IMPL_CLAMP, (gl_state_index)0, (gl_state_index)0, (gl_state_index)0 };
- /* XXX: note we are modifying the incoming shader here! Need to
- * do this before emitting the constant decls below, or this
- * will be missed.
- */
- unsigned pointSizeClampConst =
- _mesa_add_state_reference(proginfo->Parameters,
- pointSizeClampState);
- struct ureg_dst psizregtemp = ureg_DECL_temporary(ureg);
- t->pointSizeConst = ureg_DECL_constant(ureg, pointSizeClampConst);
- t->pointSizeResult = t->outputs[i];
- t->pointSizeOutIndex = i;
- t->outputs[i] = psizregtemp;
- }
}
if (passthrough_edgeflags)
emit_edgeflags(t);
if (sysInputs & (1 << i)) {
unsigned semName = mesa_sysval_to_semantic[i];
t->systemValues[i] = ureg_DECL_system_value(ureg, numSys, semName, 0);
+ if (semName == TGSI_SEMANTIC_INSTANCEID ||
+ semName == TGSI_SEMANTIC_VERTEXID) {
+ /* From Gallium perspective, these system values are always
+ * integer, and require native integer support. However, if
+ * native integer is supported on the vertex stage but not the
+ * pixel stage (e.g, i915g + draw), Mesa will generate IR that
+ * assumes these system values are floats. To resolve the
+ * inconsistency, we insert a U2F.
+ */
+ struct st_context *st = st_context(ctx);
+ struct pipe_screen *pscreen = st->pipe->screen;
+ assert(procType == TGSI_PROCESSOR_VERTEX);
+ assert(pscreen->get_shader_param(pscreen, PIPE_SHADER_VERTEX, PIPE_SHADER_CAP_INTEGERS));
+ if (!ctx->Const.NativeIntegers) {
+ struct ureg_dst temp = ureg_DECL_local_temporary(t->ureg);
+ ureg_U2F( t->ureg, ureg_writemask(temp, TGSI_WRITEMASK_X), t->systemValues[i]);
+ t->systemValues[i] = ureg_scalar(ureg_src(temp), 0);
+ }
+ }
numSys++;
sysInputs &= ~(1 << i);
}
*/
for (i = 0; i < (unsigned)program->next_temp; i++) {
/* XXX use TGSI_FILE_TEMPORARY_ARRAY when it's supported by ureg */
- t->temps[i] = ureg_DECL_temporary(t->ureg);
+ t->temps[i] = ureg_DECL_local_temporary(t->ureg);
}
}
i = 0;
foreach_iter(exec_list_iterator, iter, program->immediates) {
immediate_storage *imm = (immediate_storage *)iter.get();
+ assert(i < program->num_immediates);
t->immediates[i++] = emit_immediate(t, imm->values, imm->type, imm->size);
}
+ assert(i == program->num_immediates);
/* texture samplers */
for (i = 0; i < ctx->Const.MaxTextureImageUnits; i++) {
*/
foreach_iter(exec_list_iterator, iter, program->instructions) {
set_insn_start(t, ureg_get_instruction_number(ureg));
- compile_tgsi_instruction(t, (glsl_to_tgsi_instruction *)iter.get());
-
- if (t->prevInstWrotePointSize && proginfo->Id) {
- /* The previous instruction wrote to the (fake) vertex point size
- * result register. Now we need to clamp that value to the min/max
- * point size range, putting the result into the real point size
- * register.
- * Note that we can't do this easily at the end of program due to
- * possible early return.
- */
- set_insn_start(t, ureg_get_instruction_number(ureg));
- ureg_MAX(t->ureg,
- ureg_writemask(t->outputs[t->pointSizeOutIndex], WRITEMASK_X),
- ureg_src(t->outputs[t->pointSizeOutIndex]),
- ureg_swizzle(t->pointSizeConst, 1,1,1,1));
- ureg_MIN(t->ureg, ureg_writemask(t->pointSizeResult, WRITEMASK_X),
- ureg_src(t->outputs[t->pointSizeOutIndex]),
- ureg_swizzle(t->pointSizeConst, 2,2,2,2));
- }
- t->prevInstWrotePointSize = GL_FALSE;
+ compile_tgsi_instruction(t, (glsl_to_tgsi_instruction *)iter.get(),
+ clamp_color);
}
/* Fix up all emitted labels:
t->insn[t->labels[i].branch_target]);
}
+ if (program->shader_program) {
+ /* This has to be done last. Any operation the can cause
+ * prog->ParameterValues to get reallocated (e.g., anything that adds a
+ * program constant) has to happen before creating this linkage.
+ */
+ for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) {
+ if (program->shader_program->_LinkedShaders[i] == NULL)
+ continue;
+
+ _mesa_associate_uniform_storage(ctx, program->shader_program,
+ program->shader_program->_LinkedShaders[i]->Program->Parameters);
+ }
+ }
+
out:
- FREE(t->insn);
- FREE(t->labels);
- FREE(t->constants);
- FREE(t->immediates);
+ if (t) {
+ FREE(t->insn);
+ FREE(t->labels);
+ FREE(t->constants);
+ FREE(t->immediates);
+
+ if (t->error) {
+ debug_printf("%s: translate error flag set\n", __FUNCTION__);
+ }
- if (t->error) {
- debug_printf("%s: translate error flag set\n", __FUNCTION__);
+ FREE(t);
}
return ret;
static struct gl_program *
get_mesa_program(struct gl_context *ctx,
struct gl_shader_program *shader_program,
- struct gl_shader *shader)
+ struct gl_shader *shader)
{
- glsl_to_tgsi_visitor* v = new glsl_to_tgsi_visitor();
+ glsl_to_tgsi_visitor* v;
struct gl_program *prog;
GLenum target;
const char *target_string;
if (!prog)
return NULL;
prog->Parameters = _mesa_new_parameter_list();
+ v = new glsl_to_tgsi_visitor();
v->ctx = ctx;
v->prog = prog;
v->shader_program = shader_program;
_mesa_generate_parameters_list_for_uniforms(shader_program, shader,
prog->Parameters);
+ /* Remove reads from output registers. */
+ lower_output_reads(shader->ir);
+
/* Emit intermediate IR for main(). */
visit_exec_list(shader->ir, v);
}
#endif
- /* Remove reads to output registers, and to varyings in vertex shaders. */
- v->remove_output_reads(PROGRAM_OUTPUT);
- if (target == GL_VERTEX_PROGRAM_ARB)
- v->remove_output_reads(PROGRAM_VARYING);
-
/* Perform optimizations on the instructions in the glsl_to_tgsi_visitor. */
v->simplify_cmp();
v->copy_propagate();
_mesa_print_ir(shader->ir, NULL);
printf("\n");
printf("\n");
+ fflush(stdout);
}
prog->Instructions = NULL;
prog->NumInstructions = 0;
- do_set_program_inouts(shader->ir, prog);
+ do_set_program_inouts(shader->ir, prog, shader->Type == GL_FRAGMENT_SHADER);
count_resources(v, prog);
- check_resources(ctx, shader_program, v, prog);
-
_mesa_reference_program(ctx, &shader->Program, prog);
+ /* This has to be done last. Any operation the can cause
+ * prog->ParameterValues to get reallocated (e.g., anything that adds a
+ * program constant) has to happen before creating this linkage.
+ */
+ _mesa_associate_uniform_storage(ctx, shader_program, prog->Parameters);
+ if (!shader_program->LinkStatus) {
+ return NULL;
+ }
+
struct st_vertex_program *stvp;
struct st_fragment_program *stfp;
struct st_geometry_program *stgp;
&ctx->ShaderCompilerOptions[_mesa_shader_type_to_index(prog->_LinkedShaders[i]->Type)];
do {
+ unsigned what_to_lower = MOD_TO_FRACT | DIV_TO_MUL_RCP |
+ EXP_TO_EXP2 | LOG_TO_LOG2;
+ if (options->EmitNoPow)
+ what_to_lower |= POW_TO_EXP2;
+ if (!ctx->Const.NativeIntegers)
+ what_to_lower |= INT_DIV_TO_MUL_RCP;
+
progress = false;
/* Lowering */
do_mat_op_to_vec(ir);
- lower_instructions(ir, (MOD_TO_FRACT | DIV_TO_MUL_RCP | EXP_TO_EXP2
- | LOG_TO_LOG2 | INT_DIV_TO_MUL_RCP
- | ((options->EmitNoPow) ? POW_TO_EXP2 : 0)));
+ lower_instructions(ir, what_to_lower);
progress = do_lower_jumps(ir, true, true, options->EmitNoMainReturn, options->EmitNoCont, options->EmitNoLoops) || progress;
- progress = do_common_optimization(ir, true, options->MaxUnrollIterations) || progress;
+ progress = do_common_optimization(ir, true, true,
+ options->MaxUnrollIterations)
+ || progress;
progress = lower_quadop_vector(ir, false) || progress;
return GL_TRUE;
}
+void
+st_translate_stream_output_info(glsl_to_tgsi_visitor *glsl_to_tgsi,
+ const GLuint outputMapping[],
+ struct pipe_stream_output_info *so)
+{
+ unsigned i;
+ struct gl_transform_feedback_info *info =
+ &glsl_to_tgsi->shader_program->LinkedTransformFeedback;
+
+ for (i = 0; i < info->NumOutputs; i++) {
+ so->output[i].register_index =
+ outputMapping[info->Outputs[i].OutputRegister];
+ so->output[i].start_component = info->Outputs[i].ComponentOffset;
+ so->output[i].num_components = info->Outputs[i].NumComponents;
+ so->output[i].output_buffer = info->Outputs[i].OutputBuffer;
+ so->output[i].dst_offset = info->Outputs[i].DstOffset;
+ }
+
+ for (i = 0; i < PIPE_MAX_SO_BUFFERS; i++) {
+ so->stride[i] = info->BufferStride[i];
+ }
+ so->num_outputs = info->NumOutputs;
+}
+
} /* extern "C" */
projects / mesa.git / blobdiff