*/
#include "pipe/p_compiler.h"
+#include "pipe/p_context.h"
+#include "pipe/p_screen.h"
#include "pipe/p_shader_tokens.h"
#include "pipe/p_state.h"
#include "tgsi/tgsi_ureg.h"
#include "st_mesa_to_tgsi.h"
-#include "shader/prog_instruction.h"
-#include "shader/prog_parameter.h"
-#include "shader/prog_print.h"
+#include "st_context.h"
+#include "program/prog_instruction.h"
+#include "program/prog_parameter.h"
#include "util/u_debug.h"
#include "util/u_math.h"
#include "util/u_memory.h"
+
+#define PROGRAM_ANY_CONST ((1 << PROGRAM_LOCAL_PARAM) | \
+ (1 << PROGRAM_ENV_PARAM) | \
+ (1 << PROGRAM_STATE_VAR) | \
+ (1 << PROGRAM_NAMED_PARAM) | \
+ (1 << PROGRAM_CONSTANT) | \
+ (1 << PROGRAM_UNIFORM))
+
+
struct label {
unsigned branch_target;
unsigned token;
};
+
+/**
+ * Intermediate state used during shader translation.
+ */
struct st_translate {
struct ureg_program *ureg;
struct ureg_src inputs[PIPE_MAX_SHADER_INPUTS];
struct ureg_dst address[1];
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
+};
+
+
+/**
+ * Make note of a branch to a label in the TGSI code.
+ * After we've emitted all instructions, we'll go over the list
+ * of labels built here and patch the TGSI code with the actual
+ * location of each label.
+ */
static unsigned *get_label( struct st_translate *t,
unsigned branch_target )
{
}
+/**
+ * Called prior to emitting the TGSI code for each Mesa instruction.
+ * Allocate additional space for instructions if needed.
+ * Update the insn[] array so the next Mesa instruction points to
+ * the next TGSI instruction.
+ */
static void set_insn_start( struct st_translate *t,
unsigned start )
{
}
-/*
- * Map mesa register file to TGSI register file.
+/**
+ * Map a Mesa dst register to a TGSI ureg_dst register.
*/
static struct ureg_dst
dst_register( struct st_translate *t,
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)
+ assert(index < FRAG_RESULT_MAX);
+ else
+ assert(index < GEOM_RESULT_MAX);
+
+ assert(t->outputMapping[index] < Elements(t->outputs));
+
return t->outputs[t->outputMapping[index]];
case PROGRAM_ADDRESS:
}
+/**
+ * Map a Mesa src register to a TGSI ureg_src register.
+ */
static struct ureg_src
src_register( struct st_translate *t,
gl_register_file file,
- GLuint index )
+ GLint index )
{
switch( file ) {
case PROGRAM_UNDEFINED:
return ureg_src_undef();
case PROGRAM_TEMPORARY:
+ assert(index >= 0);
+ assert(index < Elements(t->temps));
if (ureg_dst_is_undef(t->temps[index]))
t->temps[index] = ureg_DECL_temporary( t->ureg );
return ureg_src(t->temps[index]);
- case PROGRAM_STATE_VAR:
case PROGRAM_NAMED_PARAM:
case PROGRAM_ENV_PARAM:
+ case PROGRAM_LOCAL_PARAM:
case PROGRAM_UNIFORM:
- case PROGRAM_CONSTANT: /* ie, immediate */
+ assert(index >= 0);
return t->constants[index];
+ case PROGRAM_STATE_VAR:
+ case PROGRAM_CONSTANT: /* ie, immediate */
+ if (index < 0)
+ return ureg_DECL_constant( t->ureg, 0 );
+ else
+ return t->constants[index];
case PROGRAM_INPUT:
+ assert(t->inputMapping[index] < Elements(t->inputs));
return t->inputs[t->inputMapping[index]];
case PROGRAM_OUTPUT:
+ assert(t->outputMapping[index] < Elements(t->outputs));
return ureg_src(t->outputs[t->outputMapping[index]]); /* not needed? */
case PROGRAM_ADDRESS:
return ureg_src(t->address[index]);
+ case PROGRAM_SYSTEM_VALUE:
+ assert(index < Elements(t->systemValues));
+ return t->systemValues[index];
+
default:
debug_assert( 0 );
return ureg_src_undef();
/**
* Map mesa texture target to TGSI texture target.
*/
-static unsigned
+unsigned
translate_texture_target( GLuint textarget,
GLboolean shadow )
{
case TEXTURE_1D_INDEX: return TGSI_TEXTURE_SHADOW1D;
case TEXTURE_2D_INDEX: return TGSI_TEXTURE_SHADOW2D;
case TEXTURE_RECT_INDEX: return TGSI_TEXTURE_SHADOWRECT;
+ case TEXTURE_1D_ARRAY_INDEX: return TGSI_TEXTURE_SHADOW1D_ARRAY;
+ case TEXTURE_2D_ARRAY_INDEX: return TGSI_TEXTURE_SHADOW2D_ARRAY;
default: break;
}
}
case TEXTURE_3D_INDEX: return TGSI_TEXTURE_3D;
case TEXTURE_CUBE_INDEX: return TGSI_TEXTURE_CUBE;
case TEXTURE_RECT_INDEX: return TGSI_TEXTURE_RECT;
+ case TEXTURE_1D_ARRAY_INDEX: return TGSI_TEXTURE_1D_ARRAY;
+ case TEXTURE_2D_ARRAY_INDEX: return TGSI_TEXTURE_2D_ARRAY;
+ case TEXTURE_EXTERNAL_INDEX: return TGSI_TEXTURE_2D;
default:
debug_assert( 0 );
return TGSI_TEXTURE_1D;
}
+/**
+ * Create a TGSI ureg_dst register from a Mesa dest register.
+ */
static struct ureg_dst
translate_dst( struct st_translate *t,
const struct prog_dst_register *DstReg,
}
+/**
+ * Create a TGSI ureg_src register from a Mesa src register.
+ */
static struct ureg_src
translate_src( struct st_translate *t,
const struct prog_src_register *SrcReg )
{
struct ureg_src src = src_register( t, SrcReg->File, SrcReg->Index );
+ if (t->procType == TGSI_PROCESSOR_GEOMETRY && SrcReg->HasIndex2) {
+ src = src_register( t, SrcReg->File, SrcReg->Index2 );
+ if (SrcReg->RelAddr2)
+ src = ureg_src_dimension_indirect( src, ureg_src(t->address[0]),
+ SrcReg->Index);
+ else
+ src = ureg_src_dimension( src, SrcReg->Index);
+ }
+
src = ureg_swizzle( src,
GET_SWZ( SrcReg->Swizzle, 0 ) & 0x3,
GET_SWZ( SrcReg->Swizzle, 1 ) & 0x3,
if (SrcReg->Abs)
src = ureg_abs(src);
- if (SrcReg->RelAddr)
+ if (SrcReg->RelAddr) {
src = ureg_src_indirect( src, ureg_src(t->address[0]));
-
+ if (SrcReg->File != PROGRAM_INPUT &&
+ SrcReg->File != PROGRAM_OUTPUT) {
+ /* If SrcReg->Index was negative, it was set to zero in
+ * src_register(). Reassign it now. But don't do this
+ * for input/output regs since they get remapped while
+ * const buffers don't.
+ */
+ src.Index = SrcReg->Index;
+ }
+ }
+
return src;
}
-static unsigned
+unsigned
translate_opcode( unsigned op )
{
switch( op ) {
return TGSI_OPCODE_DST;
case OPCODE_ELSE:
return TGSI_OPCODE_ELSE;
+ case OPCODE_EMIT_VERTEX:
+ return TGSI_OPCODE_EMIT;
+ case OPCODE_END_PRIMITIVE:
+ return TGSI_OPCODE_ENDPRIM;
case OPCODE_ENDIF:
return TGSI_OPCODE_ENDIF;
case OPCODE_ENDLOOP:
dst, num_dst,
translate_texture_target( inst->TexSrcTarget,
inst->TexShadow ),
+ NULL, 0,
src, num_src );
return;
/**
- * Emit the TGSI instructions for inverting the WPOS y coordinate.
+ * Emit the TGSI instructions for inverting and adjusting WPOS.
+ * This code is unavoidable because it also depends on whether
+ * a FBO is bound (STATE_FB_WPOS_Y_TRANSFORM).
*/
static void
-emit_inverted_wpos( struct st_translate *t,
- const struct gl_program *program )
+emit_wpos_adjustment( struct st_translate *t,
+ const struct gl_program *program,
+ boolean invert,
+ GLfloat adjX, GLfloat adjY[2])
{
struct ureg_program *ureg = t->ureg;
* Need to replace instances of INPUT[WPOS] with temp T
* where T = INPUT[WPOS] by y is inverted.
*/
- static const gl_state_index winSizeState[STATE_LENGTH]
- = { STATE_INTERNAL, STATE_FB_SIZE, 0, 0, 0 };
+ static const gl_state_index wposTransformState[STATE_LENGTH]
+ = { STATE_INTERNAL, STATE_FB_WPOS_Y_TRANSFORM, 0, 0, 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 winHeightConst = _mesa_add_state_reference(program->Parameters,
- winSizeState);
+ unsigned wposTransConst = _mesa_add_state_reference(program->Parameters,
+ wposTransformState);
- struct ureg_src winsize = ureg_DECL_constant( ureg, winHeightConst );
+ struct ureg_src wpostrans = ureg_DECL_constant( ureg, wposTransConst );
struct ureg_dst wpos_temp = ureg_DECL_temporary( ureg );
struct ureg_src wpos_input = t->inputs[t->inputMapping[FRAG_ATTRIB_WPOS]];
- /* MOV wpos_temp, input[wpos]
- */
- ureg_MOV( ureg, wpos_temp, wpos_input );
+ /* First, apply the coordinate shift: */
+ if (adjX || adjY[0] || adjY[1]) {
+ if (adjY[0] != adjY[1]) {
+ /* Adjust the y coordinate by adjY[1] or adjY[0] respectively
+ * depending on whether inversion is actually going to be applied
+ * 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);
+
+ ureg_CMP(ureg, adj_temp,
+ ureg_scalar(wpostrans, invert ? 2 : 0),
+ ureg_imm4f(ureg, adjX, adjY[0], 0.0f, 0.0f),
+ ureg_imm4f(ureg, adjX, adjY[1], 0.0f, 0.0f));
+ ureg_ADD(ureg, wpos_temp, wpos_input, ureg_src(adj_temp));
+ } else {
+ ureg_ADD(ureg, wpos_temp, wpos_input,
+ ureg_imm4f(ureg, adjX, adjY[0], 0.0f, 0.0f));
+ }
+ wpos_input = ureg_src(wpos_temp);
+ } else {
+ /* MOV wpos_temp, input[wpos]
+ */
+ ureg_MOV( ureg, wpos_temp, wpos_input );
+ }
- /* SUB wpos_temp.y, winsize_const, wpos_input
+ /* Now the conditional y flip: STATE_FB_WPOS_Y_TRANSFORM.xy/zw will be
+ * inversion/identity, or the other way around if we're drawing to an FBO.
*/
- ureg_SUB( ureg,
- ureg_writemask(wpos_temp, TGSI_WRITEMASK_Y ),
- winsize,
- wpos_input);
+ if (invert) {
+ /* MAD wpos_temp.y, wpos_input, wpostrans.xxxx, wpostrans.yyyy
+ */
+ ureg_MAD( ureg,
+ ureg_writemask(wpos_temp, TGSI_WRITEMASK_Y ),
+ wpos_input,
+ ureg_scalar(wpostrans, 0),
+ ureg_scalar(wpostrans, 1));
+ } else {
+ /* MAD wpos_temp.y, wpos_input, wpostrans.zzzz, wpostrans.wwww
+ */
+ ureg_MAD( ureg,
+ ureg_writemask(wpos_temp, TGSI_WRITEMASK_Y ),
+ wpos_input,
+ ureg_scalar(wpostrans, 2),
+ ureg_scalar(wpostrans, 3));
+ }
/* Use wpos_temp as position input from here on:
*/
}
+/**
+ * Emit fragment position/ooordinate code.
+ */
+static void
+emit_wpos(struct st_context *st,
+ struct st_translate *t,
+ const struct gl_program *program,
+ struct ureg_program *ureg)
+{
+ const struct gl_fragment_program *fp =
+ (const struct gl_fragment_program *) program;
+ struct pipe_screen *pscreen = st->pipe->screen;
+ GLfloat adjX = 0.0f;
+ GLfloat adjY[2] = { 0.0f, 0.0f };
+ boolean invert = FALSE;
+
+ /* Query the pixel center conventions supported by the pipe driver and set
+ * adjX, adjY to help out if it cannot handle the requested one internally.
+ *
+ * The bias of the y-coordinate depends on whether y-inversion takes place
+ * (adjY[1]) or not (adjY[0]), which is in turn dependent on whether we are
+ * drawing to an FBO (causes additional inversion), and whether the the pipe
+ * driver origin and the requested origin differ (the latter condition is
+ * stored in the 'invert' variable).
+ *
+ * For height = 100 (i = integer, h = half-integer, l = lower, u = upper):
+ *
+ * center shift only:
+ * i -> h: +0.5
+ * h -> i: -0.5
+ *
+ * inversion only:
+ * l,i -> u,i: ( 0.0 + 1.0) * -1 + 100 = 99
+ * l,h -> u,h: ( 0.5 + 0.0) * -1 + 100 = 99.5
+ * u,i -> l,i: (99.0 + 1.0) * -1 + 100 = 0
+ * u,h -> l,h: (99.5 + 0.0) * -1 + 100 = 0.5
+ *
+ * inversion and center shift:
+ * l,i -> u,h: ( 0.0 + 0.5) * -1 + 100 = 99.5
+ * l,h -> u,i: ( 0.5 + 0.5) * -1 + 100 = 99
+ * u,i -> l,h: (99.0 + 0.5) * -1 + 100 = 0.5
+ * u,h -> l,i: (99.5 + 0.5) * -1 + 100 = 0
+ */
+ if (fp->OriginUpperLeft) {
+ /* Fragment shader wants origin in upper-left */
+ if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT)) {
+ /* the driver supports upper-left origin */
+ }
+ else if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT)) {
+ /* the driver supports lower-left origin, need to invert Y */
+ ureg_property_fs_coord_origin(ureg, TGSI_FS_COORD_ORIGIN_LOWER_LEFT);
+ invert = TRUE;
+ }
+ else
+ assert(0);
+ }
+ else {
+ /* Fragment shader wants origin in lower-left */
+ if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT))
+ /* the driver supports lower-left origin */
+ ureg_property_fs_coord_origin(ureg, TGSI_FS_COORD_ORIGIN_LOWER_LEFT);
+ else if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT))
+ /* the driver supports upper-left origin, need to invert Y */
+ invert = TRUE;
+ else
+ assert(0);
+ }
+
+ if (fp->PixelCenterInteger) {
+ /* Fragment shader wants pixel center integer */
+ if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER)) {
+ /* the driver supports pixel center integer */
+ adjY[1] = 1.0f;
+ ureg_property_fs_coord_pixel_center(ureg, TGSI_FS_COORD_PIXEL_CENTER_INTEGER);
+ }
+ else if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER)) {
+ /* the driver supports pixel center half integer, need to bias X,Y */
+ adjX = -0.5f;
+ adjY[0] = -0.5f;
+ adjY[1] = 0.5f;
+ }
+ else
+ assert(0);
+ }
+ else {
+ /* Fragment shader wants pixel center half integer */
+ if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER)) {
+ /* the driver supports pixel center half integer */
+ }
+ else if (pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER)) {
+ /* the driver supports pixel center integer, need to bias X,Y */
+ adjX = adjY[0] = adjY[1] = 0.5f;
+ ureg_property_fs_coord_pixel_center(ureg, TGSI_FS_COORD_PIXEL_CENTER_INTEGER);
+ }
+ else
+ assert(0);
+ }
+
+ /* we invert after adjustment so that we avoid the MOV to temporary,
+ * and reuse the adjustment ADD instead */
+ emit_wpos_adjustment(t, program, invert, adjX, adjY);
+}
+
+
/**
* OpenGL's fragment gl_FrontFace input is 1 for front-facing, 0 for back.
* TGSI uses +1 for front, -1 for back.
}
+static void
+emit_edgeflags( struct st_translate *t,
+ const struct gl_program *program )
+{
+ struct ureg_program *ureg = t->ureg;
+ struct ureg_dst edge_dst = t->outputs[t->outputMapping[VERT_RESULT_EDGE]];
+ struct ureg_src edge_src = t->inputs[t->inputMapping[VERT_ATTRIB_EDGEFLAG]];
+
+ ureg_MOV( ureg, edge_dst, edge_src );
+}
+
+
/**
* Translate Mesa program to TGSI format.
* \param program the program to translate
* \param outputSemanticIndex the semantic index (ex: which texcoord) for
* each output
*
- * \return array of translated tokens, caller's responsibility to free
+ * \return PIPE_OK or PIPE_ERROR_OUT_OF_MEMORY
*/
enum pipe_error
st_translate_mesa_program(
- GLcontext *ctx,
+ struct gl_context *ctx,
uint procType,
struct ureg_program *ureg,
const struct gl_program *program,
GLuint numOutputs,
const GLuint outputMapping[],
const ubyte outputSemanticName[],
- const ubyte outputSemanticIndex[] )
+ const ubyte outputSemanticIndex[],
+ boolean passthrough_edgeflags )
{
struct st_translate translate, *t;
unsigned i;
+ enum pipe_error ret = PIPE_OK;
+
+ assert(numInputs <= Elements(t->inputs));
+ assert(numOutputs <= Elements(t->outputs));
t = &translate;
memset(t, 0, sizeof *t);
t->inputMapping = inputMapping;
t->outputMapping = outputMapping;
t->ureg = ureg;
+ t->pointSizeOutIndex = -1;
+ t->prevInstWrotePointSize = GL_FALSE;
/*_mesa_print_program(program);*/
*/
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]);
+ if (program->InputFlags[0] & PROG_PARAM_BIT_CYL_WRAP) {
+ t->inputs[i] = ureg_DECL_fs_input_cyl(ureg,
+ inputSemanticName[i],
+ inputSemanticIndex[i],
+ interpMode[i],
+ TGSI_CYLINDRICAL_WRAP_X);
+ }
+ else {
+ t->inputs[i] = ureg_DECL_fs_input(ureg,
+ inputSemanticName[i],
+ inputSemanticIndex[i],
+ interpMode[i]);
+ }
}
if (program->InputsRead & FRAG_BIT_WPOS) {
/* Must do this after setting up t->inputs, and before
* emitting constant references, below:
*/
- emit_inverted_wpos( t, program );
+ emit_wpos(st_context(ctx), t, program, ureg);
}
if (program->InputsRead & FRAG_BIT_FACE) {
t->outputs[i] = ureg_writemask( t->outputs[i],
TGSI_WRITEMASK_Z );
break;
+ case TGSI_SEMANTIC_STENCIL:
+ t->outputs[i] = ureg_DECL_output( ureg,
+ TGSI_SEMANTIC_STENCIL, /* Stencil */
+ outputSemanticIndex[i] );
+ t->outputs[i] = ureg_writemask( t->outputs[i],
+ TGSI_WRITEMASK_Y );
+ break;
case TGSI_SEMANTIC_COLOR:
t->outputs[i] = ureg_DECL_output( ureg,
TGSI_SEMANTIC_COLOR,
}
}
}
+ else if (procType == TGSI_PROCESSOR_GEOMETRY) {
+ for (i = 0; i < numInputs; i++) {
+ t->inputs[i] = ureg_DECL_gs_input(ureg,
+ i,
+ inputSemanticName[i],
+ inputSemanticIndex[i]);
+ }
+
+ for (i = 0; i < numOutputs; i++) {
+ t->outputs[i] = ureg_DECL_output( ureg,
+ outputSemanticName[i],
+ outputSemanticIndex[i] );
+ }
+ }
else {
+ assert(procType == TGSI_PROCESSOR_VERTEX);
+
for (i = 0; i < numInputs; i++) {
t->inputs[i] = ureg_DECL_vs_input(ureg, i);
}
t->outputs[i] = ureg_DECL_output( ureg,
outputSemanticName[i],
outputSemanticIndex[i] );
+ if ((outputSemanticName[i] == TGSI_SEMANTIC_PSIZE) && program->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, 0, 0, 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(program->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, program );
}
/* Declare address register.
t->address[0] = ureg_DECL_address( ureg );
}
+ /* Declare misc input registers
+ */
+ {
+ GLbitfield sysInputs = program->SystemValuesRead;
+ unsigned numSys = 0;
+ for (i = 0; sysInputs; i++) {
+ if (sysInputs & (1 << i)) {
+ unsigned semName = mesa_sysval_to_semantic[i];
+ t->systemValues[i] = ureg_DECL_system_value(ureg, numSys, semName, 0);
+ numSys++;
+ sysInputs &= ~(1 << i);
+ }
+ }
+ }
+
+ if (program->IndirectRegisterFiles & (1 << PROGRAM_TEMPORARY)) {
+ /* If temps are accessed with indirect addressing, declare temporaries
+ * in sequential order. Else, we declare them on demand elsewhere.
+ */
+ for (i = 0; i < program->NumTemporaries; i++) {
+ /* XXX use TGSI_FILE_TEMPORARY_ARRAY when it's supported by ureg */
+ t->temps[i] = ureg_DECL_temporary( t->ureg );
+ }
+ }
/* Emit constants and immediates. Mesa uses a single index space
* for these, so we put all the translated regs in t->constants.
*/
if (program->Parameters) {
-
t->constants = CALLOC( program->Parameters->NumParameters,
sizeof t->constants[0] );
- if (t->constants == NULL)
+ if (t->constants == NULL) {
+ ret = PIPE_ERROR_OUT_OF_MEMORY;
goto out;
-
+ }
+
for (i = 0; i < program->Parameters->NumParameters; i++) {
switch (program->Parameters->Parameters[i].Type) {
case PROGRAM_ENV_PARAM:
+ case PROGRAM_LOCAL_PARAM:
case PROGRAM_STATE_VAR:
case PROGRAM_NAMED_PARAM:
case PROGRAM_UNIFORM:
t->constants[i] = ureg_DECL_constant( ureg, i );
break;
- /* Emit immediates only when there is no address register
- * in use. FIXME: Be smarter and recognize param arrays:
+ /* Emit immediates only when there's no indirect addressing of
+ * the const buffer.
+ * FIXME: Be smarter and recognize param arrays:
* indirect addressing is only valid within the referenced
* array.
*/
case PROGRAM_CONSTANT:
- if (program->NumAddressRegs > 0)
+ if (program->IndirectRegisterFiles & PROGRAM_ANY_CONST)
t->constants[i] = ureg_DECL_constant( ureg, i );
else
t->constants[i] =
ureg_DECL_immediate( ureg,
- program->Parameters->ParameterValues[i],
+ (const float*) program->Parameters->ParameterValues[i],
4 );
break;
default:
for (i = 0; i < program->NumInstructions; i++) {
set_insn_start( t, ureg_get_instruction_number( ureg ));
compile_instruction( t, &program->Instructions[i] );
+
+ if (t->prevInstWrotePointSize && program->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;
}
/* Fix up all emitted labels:
t->insn[t->labels[i].branch_target] );
}
- return PIPE_OK;
-
out:
FREE(t->insn);
FREE(t->labels);
debug_printf("%s: translate error flag set\n", __FUNCTION__);
}
-/* ???
- if (!tokens) {
- debug_printf("%s: failed to translate Mesa program:\n", __FUNCTION__);
- _mesa_print_program(program);
- debug_assert(0);
- }
-*/
- return PIPE_ERROR_OUT_OF_MEMORY;
+ return ret;
}
/**
- * Tokens cannot be free with _mesa_free otherwise the builtin gallium
+ * Tokens cannot be free with free otherwise the builtin gallium
* malloc debugging will get confused.
*/
void