/**
* This function executes vertex programs
*/
static GLboolean
run_vp( struct gl_context *ctx, struct tnl_pipeline_stage *stage )
{
TNLcontext *tnl = TNL_CONTEXT(ctx);
struct vp_stage_data *store = VP_STAGE_DATA(stage);
struct vertex_buffer *VB = &tnl->vb;
struct gl_program *program = ctx->VertexProgram._Current;
struct gl_program_machine *machine = &store->machine;
GLuint outputs[VARYING_SLOT_MAX], numOutputs;
GLuint i, j;
if (!program)
return GL_TRUE;
/* ARB program or vertex shader */
_mesa_load_state_parameters(ctx, program->Parameters);
/* make list of outputs to save some time below */
numOutputs = 0;
for (i = 0; i < VARYING_SLOT_MAX; i++) {
if (program->info.outputs_written & BITFIELD64_BIT(i)) {
outputs[numOutputs++] = i;
}
}
/* Allocate result vectors. We delay this until now to avoid allocating
* memory that would never be used if we don't run the software tnl pipeline.
*/
if (!store->results[0].storage) {
for (i = 0; i < VARYING_SLOT_MAX; i++) {
assert(!store->results[i].storage);
_mesa_vector4f_alloc( &store->results[i], 0, VB->Size, 32 );
store->results[i].size = 4;
}
}
map_textures(ctx, program);
for (i = 0; i < VB->Count; i++) {
GLuint attr;
init_machine(ctx, machine, tnl->CurInstance);
#if 0
printf("Input %d: %f, %f, %f, %f\n", i,
VB->AttribPtr[0]->data[i][0],
VB->AttribPtr[0]->data[i][1],
VB->AttribPtr[0]->data[i][2],
VB->AttribPtr[0]->data[i][3]);
printf(" color: %f, %f, %f, %f\n",
VB->AttribPtr[3]->data[i][0],
VB->AttribPtr[3]->data[i][1],
VB->AttribPtr[3]->data[i][2],
VB->AttribPtr[3]->data[i][3]);
printf(" normal: %f, %f, %f, %f\n",
VB->AttribPtr[2]->data[i][0],
VB->AttribPtr[2]->data[i][1],
VB->AttribPtr[2]->data[i][2],
VB->AttribPtr[2]->data[i][3]);
#endif
/* the vertex array case */
for (attr = 0; attr < VERT_ATTRIB_MAX; attr++) {
if (program->info.inputs_read & BITFIELD64_BIT(attr)) {
const GLubyte *ptr = (const GLubyte*) VB->AttribPtr[attr]->data;
const GLuint size = VB->AttribPtr[attr]->size;
const GLuint stride = VB->AttribPtr[attr]->stride;
const GLfloat *data = (GLfloat *) (ptr + stride * i);
#ifdef NAN_CHECK
check_float(data[0]);
check_float(data[1]);
check_float(data[2]);
check_float(data[3]);
#endif
COPY_CLEAN_4V(machine->VertAttribs[attr], size, data);
}
}
/* execute the program */
_mesa_execute_program(ctx, program, machine);
/* copy the output registers into the VB->attribs arrays */
for (j = 0; j < numOutputs; j++) {
const GLuint attr = outputs[j];
#ifdef NAN_CHECK
check_float(machine->Outputs[attr][0]);
check_float(machine->Outputs[attr][1]);
check_float(machine->Outputs[attr][2]);
check_float(machine->Outputs[attr][3]);
#endif
COPY_4V(store->results[attr].data[i], machine->Outputs[attr]);
}
/* FOGC is a special case. Fragment shader expects (f,0,0,1) */
if (program->info.outputs_written & BITFIELD64_BIT(VARYING_SLOT_FOGC)) {
store->results[VARYING_SLOT_FOGC].data[i][1] = 0.0;
store->results[VARYING_SLOT_FOGC].data[i][2] = 0.0;
store->results[VARYING_SLOT_FOGC].data[i][3] = 1.0;
}
#ifdef NAN_CHECK
assert(machine->Outputs[0][3] != 0.0F);
#endif
#if 0
printf("HPOS: %f %f %f %f\n",
machine->Outputs[0][0],
machine->Outputs[0][1],
machine->Outputs[0][2],
machine->Outputs[0][3]);
#endif
}
unmap_textures(ctx, program);
if (program->IsPositionInvariant) {
/* We need the exact same transform as in the fixed function path here
* to guarantee invariance, depending on compiler optimization flags
* results could be different otherwise.
*/
VB->ClipPtr = TransformRaw( &store->results[0],
&ctx->_ModelProjectMatrix,
VB->AttribPtr[0] );
/* Drivers expect this to be clean to element 4...
*/
switch (VB->ClipPtr->size) {
case 1:
/* impossible */
case 2:
_mesa_vector4f_clean_elem( VB->ClipPtr, VB->Count, 2 );
/* fall-through */
case 3:
_mesa_vector4f_clean_elem( VB->ClipPtr, VB->Count, 3 );
/* fall-through */
case 4:
break;
}
}
else {
/* Setup the VB pointers so that the next pipeline stages get
* their data from the right place (the program output arrays).
*/
VB->ClipPtr = &store->results[VARYING_SLOT_POS];
VB->ClipPtr->size = 4;
VB->ClipPtr->count = VB->Count;
}
VB->AttribPtr[VERT_ATTRIB_COLOR0] = &store->results[VARYING_SLOT_COL0];
VB->AttribPtr[VERT_ATTRIB_COLOR1] = &store->results[VARYING_SLOT_COL1];
VB->AttribPtr[VERT_ATTRIB_FOG] = &store->results[VARYING_SLOT_FOGC];
VB->AttribPtr[_TNL_ATTRIB_POINTSIZE] = &store->results[VARYING_SLOT_PSIZ];
VB->BackfaceColorPtr = &store->results[VARYING_SLOT_BFC0];
VB->BackfaceSecondaryColorPtr = &store->results[VARYING_SLOT_BFC1];
for (i = 0; i < ctx->Const.MaxTextureCoordUnits; i++) {
VB->AttribPtr[_TNL_ATTRIB_TEX0 + i]
= &store->results[VARYING_SLOT_TEX0 + i];
}
for (i = 0; i < ctx->Const.MaxVarying; i++) {
if (program->info.outputs_written &
BITFIELD64_BIT(VARYING_SLOT_VAR0 + i)) {
/* Note: varying results get put into the generic attributes */
VB->AttribPtr[VERT_ATTRIB_GENERIC0+i]
= &store->results[VARYING_SLOT_VAR0 + i];
}
}
/* Perform NDC and cliptest operations:
*/
return do_ndc_cliptest(ctx, store);
}
/**
* Flush existing data, set new attrib size, replay copied vertices.
* This is called when we transition from a small vertex attribute size
* to a larger one. Ex: glTexCoord2f -> glTexCoord4f.
* We need to go back over the previous 2-component texcoords and insert
* zero and one values.
*/
static void
vbo_exec_wrap_upgrade_vertex(struct vbo_exec_context *exec,
GLuint attr, GLuint newSize )
{
struct gl_context *ctx = exec->ctx;
struct vbo_context *vbo = vbo_context(ctx);
const GLint lastcount = exec->vtx.vert_count;
GLfloat *old_attrptr[VBO_ATTRIB_MAX];
const GLuint old_vtx_size = exec->vtx.vertex_size; /* floats per vertex */
const GLuint oldSize = exec->vtx.attrsz[attr];
GLuint i;
/* Run pipeline on current vertices, copy wrapped vertices
* to exec->vtx.copied.
*/
vbo_exec_wrap_buffers( exec );
if (unlikely(exec->vtx.copied.nr)) {
/* We're in the middle of a primitive, keep the old vertex
* format around to be able to translate the copied vertices to
* the new format.
*/
memcpy(old_attrptr, exec->vtx.attrptr, sizeof(old_attrptr));
}
if (unlikely(oldSize)) {
/* Do a COPY_TO_CURRENT to ensure back-copying works for the
* case when the attribute already exists in the vertex and is
* having its size increased.
*/
vbo_exec_copy_to_current( exec );
}
/* Heuristic: Attempt to isolate attributes received outside
* begin/end so that they don't bloat the vertices.
*/
if (ctx->Driver.CurrentExecPrimitive == PRIM_OUTSIDE_BEGIN_END &&
!oldSize && lastcount > 8 && exec->vtx.vertex_size) {
vbo_exec_copy_to_current( exec );
reset_attrfv( exec );
}
/* Fix up sizes:
*/
exec->vtx.attrsz[attr] = newSize;
exec->vtx.vertex_size += newSize - oldSize;
exec->vtx.max_vert = ((VBO_VERT_BUFFER_SIZE - exec->vtx.buffer_used) /
(exec->vtx.vertex_size * sizeof(GLfloat)));
exec->vtx.vert_count = 0;
exec->vtx.buffer_ptr = exec->vtx.buffer_map;
if (unlikely(oldSize)) {
/* Size changed, recalculate all the attrptr[] values
*/
GLfloat *tmp = exec->vtx.vertex;
for (i = 0 ; i < VBO_ATTRIB_MAX ; i++) {
if (exec->vtx.attrsz[i]) {
exec->vtx.attrptr[i] = tmp;
tmp += exec->vtx.attrsz[i];
}
else
exec->vtx.attrptr[i] = NULL; /* will not be dereferenced */
}
/* Copy from current to repopulate the vertex with correct
* values.
*/
vbo_exec_copy_from_current( exec );
}
else {
/* Just have to append the new attribute at the end */
exec->vtx.attrptr[attr] = exec->vtx.vertex +
exec->vtx.vertex_size - newSize;
}
/* Replay stored vertices to translate them
* to new format here.
*
* -- No need to replay - just copy piecewise
*/
if (unlikely(exec->vtx.copied.nr)) {
GLfloat *data = exec->vtx.copied.buffer;
GLfloat *dest = exec->vtx.buffer_ptr;
GLuint j;
assert(exec->vtx.buffer_ptr == exec->vtx.buffer_map);
for (i = 0 ; i < exec->vtx.copied.nr ; i++) {
for (j = 0 ; j < VBO_ATTRIB_MAX ; j++) {
GLuint sz = exec->vtx.attrsz[j];
if (sz) {
GLint old_offset = old_attrptr[j] - exec->vtx.vertex;
GLint new_offset = exec->vtx.attrptr[j] - exec->vtx.vertex;
if (j == attr) {
if (oldSize) {
GLfloat tmp[4];
COPY_CLEAN_4V(tmp, oldSize, data + old_offset);
COPY_SZ_4V(dest + new_offset, newSize, tmp);
} else {
GLfloat *current = (GLfloat *)vbo->currval[j].Ptr;
COPY_SZ_4V(dest + new_offset, sz, current);
}
}
else {
COPY_SZ_4V(dest + new_offset, sz, data + old_offset);
}
}
}
data += old_vtx_size;
dest += exec->vtx.vertex_size;
}
exec->vtx.buffer_ptr = dest;
exec->vtx.vert_count += exec->vtx.copied.nr;
exec->vtx.copied.nr = 0;
}
}
/**
* Flush existing data, set new attrib size, replay copied vertices.
*/
static void vbo_exec_wrap_upgrade_vertex( struct vbo_exec_context *exec,
GLuint attr,
GLuint newsz )
{
GLcontext *ctx = exec->ctx;
struct vbo_context *vbo = vbo_context(ctx);
GLint lastcount = exec->vtx.vert_count;
GLfloat *tmp;
GLuint oldsz;
GLuint i;
/* Run pipeline on current vertices, copy wrapped vertices
* to exec->vtx.copied.
*/
vbo_exec_wrap_buffers( exec );
/* Do a COPY_TO_CURRENT to ensure back-copying works for the case
* when the attribute already exists in the vertex and is having
* its size increased.
*/
vbo_exec_copy_to_current( exec );
/* Heuristic: Attempt to isolate attributes received outside
* begin/end so that they don't bloat the vertices.
*/
if (ctx->Driver.CurrentExecPrimitive == PRIM_OUTSIDE_BEGIN_END &&
exec->vtx.attrsz[attr] == 0 &&
lastcount > 8 &&
exec->vtx.vertex_size) {
reset_attrfv( exec );
}
/* Fix up sizes:
*/
oldsz = exec->vtx.attrsz[attr];
exec->vtx.attrsz[attr] = newsz;
exec->vtx.vertex_size += newsz - oldsz;
exec->vtx.max_vert = ((VBO_VERT_BUFFER_SIZE - exec->vtx.buffer_used) /
(exec->vtx.vertex_size * sizeof(GLfloat)));
exec->vtx.vert_count = 0;
exec->vtx.buffer_ptr = exec->vtx.buffer_map;
/* Recalculate all the attrptr[] values
*/
for (i = 0, tmp = exec->vtx.vertex ; i < VBO_ATTRIB_MAX ; i++) {
if (exec->vtx.attrsz[i]) {
exec->vtx.attrptr[i] = tmp;
tmp += exec->vtx.attrsz[i];
}
else
exec->vtx.attrptr[i] = NULL; /* will not be dereferenced */
}
/* Copy from current to repopulate the vertex with correct values.
*/
vbo_exec_copy_from_current( exec );
/* Replay stored vertices to translate them
* to new format here.
*
* -- No need to replay - just copy piecewise
*/
if (exec->vtx.copied.nr)
{
GLfloat *data = exec->vtx.copied.buffer;
GLfloat *dest = exec->vtx.buffer_ptr;
GLuint j;
assert(exec->vtx.buffer_ptr == exec->vtx.buffer_map);
for (i = 0 ; i < exec->vtx.copied.nr ; i++) {
for (j = 0 ; j < VBO_ATTRIB_MAX ; j++) {
if (exec->vtx.attrsz[j]) {
if (j == attr) {
if (oldsz) {
COPY_CLEAN_4V( dest, oldsz, data );
data += oldsz;
dest += newsz;
} else {
const GLfloat *current = (const GLfloat *)vbo->currval[j].Ptr;
COPY_SZ_4V( dest, newsz, current );
dest += newsz;
}
}
else {
GLuint sz = exec->vtx.attrsz[j];
COPY_SZ_4V( dest, sz, data );
dest += sz;
data += sz;
}
}
}
}
exec->vtx.buffer_ptr = dest;
exec->vtx.vert_count += exec->vtx.copied.nr;
exec->vtx.copied.nr = 0;
}
}
/**
* This function executes vertex programs
*/
static GLboolean
run_vp( GLcontext *ctx, struct tnl_pipeline_stage *stage )
{
TNLcontext *tnl = TNL_CONTEXT(ctx);
struct vp_stage_data *store = VP_STAGE_DATA(stage);
struct vertex_buffer *VB = &tnl->vb;
struct gl_vertex_program *program = ctx->VertexProgram._Current;
struct gl_program_machine machine;
GLuint outputs[VERT_RESULT_MAX], numOutputs;
GLuint i, j;
if (!program)
return GL_TRUE;
if (program->IsNVProgram) {
_mesa_load_tracked_matrices(ctx);
}
else {
/* ARB program or vertex shader */
_mesa_load_state_parameters(ctx, program->Base.Parameters);
}
numOutputs = 0;
for (i = 0; i < VERT_RESULT_MAX; i++) {
if (program->Base.OutputsWritten & (1 << i)) {
outputs[numOutputs++] = i;
}
}
for (i = 0; i < VB->Count; i++) {
GLuint attr;
init_machine(ctx, &machine);
#if 0
printf("Input %d: %f, %f, %f, %f\n", i,
VB->AttribPtr[0]->data[i][0],
VB->AttribPtr[0]->data[i][1],
VB->AttribPtr[0]->data[i][2],
VB->AttribPtr[0]->data[i][3]);
printf(" color: %f, %f, %f, %f\n",
VB->AttribPtr[3]->data[i][0],
VB->AttribPtr[3]->data[i][1],
VB->AttribPtr[3]->data[i][2],
VB->AttribPtr[3]->data[i][3]);
printf(" normal: %f, %f, %f, %f\n",
VB->AttribPtr[2]->data[i][0],
VB->AttribPtr[2]->data[i][1],
VB->AttribPtr[2]->data[i][2],
VB->AttribPtr[2]->data[i][3]);
#endif
/* the vertex array case */
for (attr = 0; attr < VERT_ATTRIB_MAX; attr++) {
if (program->Base.InputsRead & (1 << attr)) {
const GLubyte *ptr = (const GLubyte*) VB->AttribPtr[attr]->data;
const GLuint size = VB->AttribPtr[attr]->size;
const GLuint stride = VB->AttribPtr[attr]->stride;
const GLfloat *data = (GLfloat *) (ptr + stride * i);
COPY_CLEAN_4V(machine.VertAttribs[attr], size, data);
}
}
/* execute the program */
_mesa_execute_program(ctx, &program->Base, &machine);
/* copy the output registers into the VB->attribs arrays */
for (j = 0; j < numOutputs; j++) {
const GLuint attr = outputs[j];
COPY_4V(store->results[attr].data[i], machine.Outputs[attr]);
}
#if 0
printf("HPOS: %f %f %f %f\n",
machine.Outputs[0][0],
machine.Outputs[0][1],
machine.Outputs[0][2],
machine.Outputs[0][3]);
#endif
}
/* Fixup fog and point size results if needed */
if (program->IsNVProgram) {
if (ctx->Fog.Enabled &&
(program->Base.OutputsWritten & (1 << VERT_RESULT_FOGC)) == 0) {
for (i = 0; i < VB->Count; i++) {
store->results[VERT_RESULT_FOGC].data[i][0] = 1.0;
}
}
if (ctx->VertexProgram.PointSizeEnabled &&
(program->Base.OutputsWritten & (1 << VERT_RESULT_PSIZ)) == 0) {
for (i = 0; i < VB->Count; i++) {
store->results[VERT_RESULT_PSIZ].data[i][0] = ctx->Point.Size;
}
}
}
/* Setup the VB pointers so that the next pipeline stages get
* their data from the right place (the program output arrays).
*/
VB->ClipPtr = &store->results[VERT_RESULT_HPOS];
VB->ClipPtr->size = 4;
VB->ClipPtr->count = VB->Count;
VB->ColorPtr[0] = &store->results[VERT_RESULT_COL0];
VB->ColorPtr[1] = &store->results[VERT_RESULT_BFC0];
VB->SecondaryColorPtr[0] = &store->results[VERT_RESULT_COL1];
VB->SecondaryColorPtr[1] = &store->results[VERT_RESULT_BFC1];
VB->FogCoordPtr = &store->results[VERT_RESULT_FOGC];
VB->AttribPtr[VERT_ATTRIB_COLOR0] = &store->results[VERT_RESULT_COL0];
VB->AttribPtr[VERT_ATTRIB_COLOR1] = &store->results[VERT_RESULT_COL1];
VB->AttribPtr[VERT_ATTRIB_FOG] = &store->results[VERT_RESULT_FOGC];
VB->AttribPtr[_TNL_ATTRIB_POINTSIZE] = &store->results[VERT_RESULT_PSIZ];
for (i = 0; i < ctx->Const.MaxTextureCoordUnits; i++) {
VB->TexCoordPtr[i] =
VB->AttribPtr[_TNL_ATTRIB_TEX0 + i]
= &store->results[VERT_RESULT_TEX0 + i];
}
for (i = 0; i < ctx->Const.MaxVarying; i++) {
if (program->Base.OutputsWritten & (1 << (VERT_RESULT_VAR0 + i))) {
/* Note: varying results get put into the generic attributes */
VB->AttribPtr[VERT_ATTRIB_GENERIC0+i]
= &store->results[VERT_RESULT_VAR0 + i];
}
}
/* Cliptest and perspective divide. Clip functions must clear
* the clipmask.
*/
store->ormask = 0;
store->andmask = CLIP_FRUSTUM_BITS;
if (tnl->NeedNdcCoords) {
VB->NdcPtr =
_mesa_clip_tab[VB->ClipPtr->size]( VB->ClipPtr,
&store->ndcCoords,
store->clipmask,
&store->ormask,
&store->andmask );
}
else {
VB->NdcPtr = NULL;
_mesa_clip_np_tab[VB->ClipPtr->size]( VB->ClipPtr,
NULL,
store->clipmask,
&store->ormask,
&store->andmask );
}
if (store->andmask) /* All vertices are outside the frustum */
return GL_FALSE;
/* This is where we'd do clip testing against the user-defined
* clipping planes, but they're not supported by vertex programs.
*/
VB->ClipOrMask = store->ormask;
VB->ClipMask = store->clipmask;
return GL_TRUE;
}
/* Flush existing data, set new attrib size, replay copied vertices.
*/
static void
_save_upgrade_vertex(struct gl_context *ctx, GLuint attr, GLuint newsz)
{
struct vbo_save_context *save = &vbo_context(ctx)->save;
GLuint oldsz;
GLuint i;
GLfloat *tmp;
/* Store the current run of vertices, and emit a GL_END. Emit a
* BEGIN in the new buffer.
*/
if (save->vert_count)
_save_wrap_buffers(ctx);
else
assert(save->copied.nr == 0);
/* Do a COPY_TO_CURRENT to ensure back-copying works for the case
* when the attribute already exists in the vertex and is having
* its size increased.
*/
_save_copy_to_current(ctx);
/* Fix up sizes:
*/
oldsz = save->attrsz[attr];
save->attrsz[attr] = newsz;
save->vertex_size += newsz - oldsz;
save->max_vert = ((VBO_SAVE_BUFFER_SIZE - save->vertex_store->used) /
save->vertex_size);
save->vert_count = 0;
/* Recalculate all the attrptr[] values:
*/
for (i = 0, tmp = save->vertex; i < VBO_ATTRIB_MAX; i++) {
if (save->attrsz[i]) {
save->attrptr[i] = tmp;
tmp += save->attrsz[i];
}
else {
save->attrptr[i] = NULL; /* will not be dereferenced. */
}
}
/* Copy from current to repopulate the vertex with correct values.
*/
_save_copy_from_current(ctx);
/* Replay stored vertices to translate them to new format here.
*
* If there are copied vertices and the new (upgraded) attribute
* has not been defined before, this list is somewhat degenerate,
* and will need fixup at runtime.
*/
if (save->copied.nr) {
GLfloat *data = save->copied.buffer;
GLfloat *dest = save->buffer;
GLuint j;
/* Need to note this and fix up at runtime (or loopback):
*/
if (attr != VBO_ATTRIB_POS && save->currentsz[attr][0] == 0) {
assert(oldsz == 0);
save->dangling_attr_ref = GL_TRUE;
}
for (i = 0; i < save->copied.nr; i++) {
for (j = 0; j < VBO_ATTRIB_MAX; j++) {
if (save->attrsz[j]) {
if (j == attr) {
if (oldsz) {
COPY_CLEAN_4V(dest, oldsz, data);
data += oldsz;
dest += newsz;
}
else {
COPY_SZ_4V(dest, newsz, save->current[attr]);
dest += newsz;
}
}
else {
GLint sz = save->attrsz[j];
COPY_SZ_4V(dest, sz, data);
data += sz;
dest += sz;
}
}
}
}
save->buffer_ptr = dest;
save->vert_count += save->copied.nr;
}
}
/**
* This function executes vertex programs
*/
static GLboolean
run_vp( GLcontext *ctx, struct tnl_pipeline_stage *stage )
{
TNLcontext *tnl = TNL_CONTEXT(ctx);
struct vp_stage_data *store = VP_STAGE_DATA(stage);
struct vertex_buffer *VB = &tnl->vb;
struct gl_vertex_program *program = ctx->VertexProgram._Current;
struct gl_program_machine machine;
GLuint outputs[VERT_RESULT_MAX], numOutputs;
GLuint i, j;
if (!program)
return GL_TRUE;
if (program->IsNVProgram) {
_mesa_load_tracked_matrices(ctx);
}
else {
/* ARB program or vertex shader */
_mesa_load_state_parameters(ctx, program->Base.Parameters);
}
/* make list of outputs to save some time below */
numOutputs = 0;
for (i = 0; i < VERT_RESULT_MAX; i++) {
if (program->Base.OutputsWritten & (1 << i)) {
outputs[numOutputs++] = i;
}
}
map_textures(ctx, program);
for (i = 0; i < VB->Count; i++) {
GLuint attr;
init_machine(ctx, &machine);
#if 0
printf("Input %d: %f, %f, %f, %f\n", i,
VB->AttribPtr[0]->data[i][0],
VB->AttribPtr[0]->data[i][1],
VB->AttribPtr[0]->data[i][2],
VB->AttribPtr[0]->data[i][3]);
printf(" color: %f, %f, %f, %f\n",
VB->AttribPtr[3]->data[i][0],
VB->AttribPtr[3]->data[i][1],
VB->AttribPtr[3]->data[i][2],
VB->AttribPtr[3]->data[i][3]);
printf(" normal: %f, %f, %f, %f\n",
VB->AttribPtr[2]->data[i][0],
VB->AttribPtr[2]->data[i][1],
VB->AttribPtr[2]->data[i][2],
VB->AttribPtr[2]->data[i][3]);
#endif
/* the vertex array case */
for (attr = 0; attr < VERT_ATTRIB_MAX; attr++) {
if (program->Base.InputsRead & (1 << attr)) {
const GLubyte *ptr = (const GLubyte*) VB->AttribPtr[attr]->data;
const GLuint size = VB->AttribPtr[attr]->size;
const GLuint stride = VB->AttribPtr[attr]->stride;
const GLfloat *data = (GLfloat *) (ptr + stride * i);
COPY_CLEAN_4V(machine.VertAttribs[attr], size, data);
}
}
/* execute the program */
_mesa_execute_program(ctx, &program->Base, &machine);
/* copy the output registers into the VB->attribs arrays */
for (j = 0; j < numOutputs; j++) {
const GLuint attr = outputs[j];
COPY_4V(store->results[attr].data[i], machine.Outputs[attr]);
}
#if 0
printf("HPOS: %f %f %f %f\n",
machine.Outputs[0][0],
machine.Outputs[0][1],
machine.Outputs[0][2],
machine.Outputs[0][3]);
#endif
}
unmap_textures(ctx, program);
/* Fixup fog and point size results if needed */
if (program->IsNVProgram) {
if (ctx->Fog.Enabled &&
(program->Base.OutputsWritten & (1 << VERT_RESULT_FOGC)) == 0) {
for (i = 0; i < VB->Count; i++) {
store->results[VERT_RESULT_FOGC].data[i][0] = 1.0;
}
}
if (ctx->VertexProgram.PointSizeEnabled &&
(program->Base.OutputsWritten & (1 << VERT_RESULT_PSIZ)) == 0) {
for (i = 0; i < VB->Count; i++) {
store->results[VERT_RESULT_PSIZ].data[i][0] = ctx->Point.Size;
}
}
}
if (program->IsPositionInvariant) {
/* We need the exact same transform as in the fixed function path here
* to guarantee invariance, depending on compiler optimization flags
* results could be different otherwise.
*/
VB->ClipPtr = TransformRaw( &store->results[0],
&ctx->_ModelProjectMatrix,
VB->AttribPtr[0] );
/* Drivers expect this to be clean to element 4...
*/
switch (VB->ClipPtr->size) {
case 1:
/* impossible */
case 2:
_mesa_vector4f_clean_elem( VB->ClipPtr, VB->Count, 2 );
/* fall-through */
case 3:
_mesa_vector4f_clean_elem( VB->ClipPtr, VB->Count, 3 );
/* fall-through */
case 4:
break;
}
}
else {
/* Setup the VB pointers so that the next pipeline stages get
* their data from the right place (the program output arrays).
*/
VB->ClipPtr = &store->results[VERT_RESULT_HPOS];
VB->ClipPtr->size = 4;
VB->ClipPtr->count = VB->Count;
}
VB->ColorPtr[0] = &store->results[VERT_RESULT_COL0];
VB->ColorPtr[1] = &store->results[VERT_RESULT_BFC0];
VB->SecondaryColorPtr[0] = &store->results[VERT_RESULT_COL1];
VB->SecondaryColorPtr[1] = &store->results[VERT_RESULT_BFC1];
VB->FogCoordPtr = &store->results[VERT_RESULT_FOGC];
VB->AttribPtr[VERT_ATTRIB_COLOR0] = &store->results[VERT_RESULT_COL0];
VB->AttribPtr[VERT_ATTRIB_COLOR1] = &store->results[VERT_RESULT_COL1];
VB->AttribPtr[VERT_ATTRIB_FOG] = &store->results[VERT_RESULT_FOGC];
VB->AttribPtr[_TNL_ATTRIB_POINTSIZE] = &store->results[VERT_RESULT_PSIZ];
for (i = 0; i < ctx->Const.MaxTextureCoordUnits; i++) {
VB->TexCoordPtr[i] =
VB->AttribPtr[_TNL_ATTRIB_TEX0 + i]
= &store->results[VERT_RESULT_TEX0 + i];
}
for (i = 0; i < ctx->Const.MaxVarying; i++) {
if (program->Base.OutputsWritten & (1 << (VERT_RESULT_VAR0 + i))) {
/* Note: varying results get put into the generic attributes */
VB->AttribPtr[VERT_ATTRIB_GENERIC0+i]
= &store->results[VERT_RESULT_VAR0 + i];
}
}
/* Perform NDC and cliptest operations:
*/
return do_ndc_cliptest(ctx, store);
}
