OpenHarmony/third_party_mesa3d
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

llvmpipe: Compute interpolation coeffs directly into SoA layout.

Browse Source
This commit is contained in:
José Fonseca
2009-08-14 11:33:26 +01:00
parent b6f43b445b
commit 17aec9304c
4 changed files with 244 additions and 168 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
src/gallium/drivers/llvmpipe/lp_quad.h
View File

@@ -88,6 +88,17 @@ struct quad_header_output
}; };
/**
* Input interpolation coefficients
*/
struct quad_interp_coef
{
float ALIGN16_ATTRIB a0[PIPE_MAX_SHADER_INPUTS][NUM_CHANNELS];
float ALIGN16_ATTRIB dadx[PIPE_MAX_SHADER_INPUTS][NUM_CHANNELS];
float ALIGN16_ATTRIB dady[PIPE_MAX_SHADER_INPUTS][NUM_CHANNELS];
};
/** /**
* Encodes everything we need to know about a 2x2 pixel block. Uses * Encodes everything we need to know about a 2x2 pixel block. Uses
* "Channel-Serial" or "SoA" layout. * "Channel-Serial" or "SoA" layout.
@@ -100,7 +111,7 @@ struct quad_header {
/* Redundant/duplicated: /* Redundant/duplicated:
*/ */
const struct tgsi_interp_coef *posCoef; const struct tgsi_interp_coef *posCoef;
const struct tgsi_interp_coef *coef; const struct quad_interp_coef *coef;
}; };
#endif /* LP_QUAD_H */ #endif /* LP_QUAD_H */

27
src/gallium/drivers/llvmpipe/lp_quad_fs.c
View File

@@ -54,9 +54,6 @@ struct quad_shade_stage
struct quad_stage stage; /**< base class */ struct quad_stage stage; /**< base class */
union tgsi_exec_channel ALIGN16_ATTRIB pos[NUM_CHANNELS]; union tgsi_exec_channel ALIGN16_ATTRIB pos[NUM_CHANNELS];
float ALIGN16_ATTRIB a0[PIPE_MAX_SHADER_INPUTS][NUM_CHANNELS];
float ALIGN16_ATTRIB dadx[PIPE_MAX_SHADER_INPUTS][NUM_CHANNELS];
float ALIGN16_ATTRIB dady[PIPE_MAX_SHADER_INPUTS][NUM_CHANNELS];
struct tgsi_exec_vector ALIGN16_ATTRIB outputs[PIPE_MAX_ATTRIBS]; struct tgsi_exec_vector ALIGN16_ATTRIB outputs[PIPE_MAX_ATTRIBS];
}; };
@@ -102,23 +99,6 @@ setup_pos_vector(struct quad_shade_stage *qss,
} }
static void
setup_coef_vector(struct quad_shade_stage *qss,
const struct tgsi_interp_coef *coef)
{
unsigned num_inputs = qss->stage.llvmpipe->fs->info.num_inputs;
unsigned attrib, chan, i;
for (attrib = 0; attrib < num_inputs; ++attrib) {
for (chan = 0; chan < NUM_CHANNELS; ++chan) {
qss->a0[attrib][chan] = coef[attrib].a0[chan];
qss->dadx[attrib][chan] = coef[attrib].dadx[chan];
qss->dady[attrib][chan] = coef[attrib].dady[chan];
}
}
}
/** /**
* Execute fragment shader for the four fragments in the quad. * Execute fragment shader for the four fragments in the quad.
*/ */
@@ -142,7 +122,9 @@ shade_quad(struct quad_stage *qs, struct quad_header *quad)
/* run shader */ /* run shader */
llvmpipe->fs->jit_function( qss->pos, llvmpipe->fs->jit_function( qss->pos,
qss->a0, qss->dadx, qss->dady, quad->coef->a0,
quad->coef->dadx,
quad->coef->dady,
constants, constants,
qss->outputs, qss->outputs,
samplers); samplers);
@@ -217,9 +199,6 @@ shade_quads(struct quad_stage *qs,
struct quad_shade_stage *qss = quad_shade_stage( qs ); struct quad_shade_stage *qss = quad_shade_stage( qs );
unsigned i, pass = 0; unsigned i, pass = 0;
setup_coef_vector(qss,
quads[0]->coef);
for (i = 0; i < nr; i++) { for (i = 0; i < nr; i++) {
if (!shade_quad(qs, quads[i])) if (!shade_quad(qs, quads[i]))
continue; continue;

362
src/gallium/drivers/llvmpipe/lp_setup.c
View File

@@ -92,7 +92,7 @@ struct setup_context {
struct quad_header *quad_ptrs[MAX_QUADS]; struct quad_header *quad_ptrs[MAX_QUADS];
unsigned count; unsigned count;
struct tgsi_interp_coef coef[PIPE_MAX_SHADER_INPUTS]; struct quad_interp_coef coef;
struct tgsi_interp_coef posCoef; /* For Z, W */ struct tgsi_interp_coef posCoef; /* For Z, W */
struct { struct {
@@ -382,35 +382,12 @@ static boolean setup_sort_vertices( struct setup_context *setup,
} }
/**
* Compute a0 for a constant-valued coefficient (GL_FLAT shading).
* The value value comes from vertex[slot][i].
* The result will be put into setup->coef[slot].a0[i].
* \param slot which attribute slot
* \param i which component of the slot (0..3)
*/
static void const_coeff( struct setup_context *setup,
struct tgsi_interp_coef *coef,
uint vertSlot, uint i)
{
assert(i <= 3);
coef->dadx[i] = 0;
coef->dady[i] = 0;
/* need provoking vertex info!
*/
coef->a0[i] = setup->vprovoke[vertSlot][i];
}
/** /**
* Compute a0, dadx and dady for a linearly interpolated coefficient, * Compute a0, dadx and dady for a linearly interpolated coefficient,
* for a triangle. * for a triangle.
*/ */
static void tri_linear_coeff( struct setup_context *setup, static void tri_pos_coeff( struct setup_context *setup,
struct tgsi_interp_coef *coef, uint vertSlot, unsigned i)
uint vertSlot, uint i)
{ {
float botda = setup->vmid[vertSlot][i] - setup->vmin[vertSlot][i]; float botda = setup->vmid[vertSlot][i] - setup->vmin[vertSlot][i];
float majda = setup->vmax[vertSlot][i] - setup->vmin[vertSlot][i]; float majda = setup->vmax[vertSlot][i] - setup->vmin[vertSlot][i];
@@ -421,8 +398,8 @@ static void tri_linear_coeff( struct setup_context *setup,
assert(i <= 3); assert(i <= 3);
coef->dadx[i] = dadx; setup->posCoef.dadx[i] = dadx;
coef->dady[i] = dady; setup->posCoef.dady[i] = dady;
/* calculate a0 as the value which would be sampled for the /* calculate a0 as the value which would be sampled for the
* fragment at (0,0), taking into account that we want to sample at * fragment at (0,0), taking into account that we want to sample at
@@ -436,20 +413,111 @@ static void tri_linear_coeff( struct setup_context *setup,
* to define a0 as the sample at a pixel center somewhere near vmin * to define a0 as the sample at a pixel center somewhere near vmin
* instead - i'll switch to this later. * instead - i'll switch to this later.
*/ */
coef->a0[i] = (setup->vmin[vertSlot][i] - setup->posCoef.a0[i] = (setup->vmin[vertSlot][i] -
(dadx * (setup->vmin[0][0] - 0.5f) + (dadx * (setup->vmin[0][0] - 0.5f) +
dady * (setup->vmin[0][1] - 0.5f))); dady * (setup->vmin[0][1] - 0.5f)));
/* /*
debug_printf("attr[%d].%c: %f dx:%f dy:%f\n", debug_printf("attr[%d].%c: %f dx:%f dy:%f\n",
slot, "xyzw"[i], slot, "xyzw"[i],
setup->coef[slot].a0[i], setup->coef[slot].a0[i],
setup->coef[slot].dadx[i], setup->coef[slot].dadx[i],
setup->coef[slot].dady[i]); setup->coef[slot].dady[i]);
*/ */
} }
/**
* Compute a0 for a constant-valued coefficient (GL_FLAT shading).
* The value value comes from vertex[slot][i].
* The result will be put into setup->coef[slot].a0[i].
* \param slot which attribute slot
* \param i which component of the slot (0..3)
*/
static void const_pos_coeff( struct setup_context *setup,
uint vertSlot, unsigned i)
{
setup->posCoef.dadx[i] = 0;
setup->posCoef.dady[i] = 0;
/* need provoking vertex info!
*/
setup->posCoef.a0[i] = setup->vprovoke[vertSlot][i];
}
/**
* Compute a0 for a constant-valued coefficient (GL_FLAT shading).
* The value value comes from vertex[slot][i].
* The result will be put into setup->coef[slot].a0[i].
* \param slot which attribute slot
* \param i which component of the slot (0..3)
*/
static void const_coeff( struct setup_context *setup,
unsigned attrib,
uint vertSlot)
{
unsigned i;
for (i = 0; i < NUM_CHANNELS; ++i) {
setup->coef.dadx[attrib][i] = 0;
setup->coef.dady[attrib][i] = 0;
/* need provoking vertex info!
*/
setup->coef.a0[attrib][i] = setup->vprovoke[vertSlot][i];
}
}
/**
* Compute a0, dadx and dady for a linearly interpolated coefficient,
* for a triangle.
*/
static void tri_linear_coeff( struct setup_context *setup,
unsigned attrib,
uint vertSlot)
{
unsigned i;
for (i = 0; i < NUM_CHANNELS; ++i) {
float botda = setup->vmid[vertSlot][i] - setup->vmin[vertSlot][i];
float majda = setup->vmax[vertSlot][i] - setup->vmin[vertSlot][i];
float a = setup->ebot.dy * majda - botda * setup->emaj.dy;
float b = setup->emaj.dx * botda - majda * setup->ebot.dx;
float dadx = a * setup->oneoverarea;
float dady = b * setup->oneoverarea;
assert(i <= 3);
setup->coef.dadx[attrib][i] = dadx;
setup->coef.dady[attrib][i] = dady;
/* calculate a0 as the value which would be sampled for the
* fragment at (0,0), taking into account that we want to sample at
* pixel centers, in other words (0.5, 0.5).
*
* this is neat but unfortunately not a good way to do things for
* triangles with very large values of dadx or dady as it will
* result in the subtraction and re-addition from a0 of a very
* large number, which means we'll end up loosing a lot of the
* fractional bits and precision from a0. the way to fix this is
* to define a0 as the sample at a pixel center somewhere near vmin
* instead - i'll switch to this later.
*/
setup->coef.a0[attrib][i] = (setup->vmin[vertSlot][i] -
(dadx * (setup->vmin[0][0] - 0.5f) +
dady * (setup->vmin[0][1] - 0.5f)));
/*
debug_printf("attr[%d].%c: %f dx:%f dy:%f\n",
slot, "xyzw"[i],
setup->coef[slot].a0[i],
setup->coef[slot].dadx[i],
setup->coef[slot].dady[i]);
*/
}
}
/** /**
* Compute a0, dadx and dady for a perspective-corrected interpolant, * Compute a0, dadx and dady for a perspective-corrected interpolant,
* for a triangle. * for a triangle.
@@ -459,35 +527,38 @@ static void tri_linear_coeff( struct setup_context *setup,
* divide the interpolated value by the interpolated W at that fragment. * divide the interpolated value by the interpolated W at that fragment.
*/ */
static void tri_persp_coeff( struct setup_context *setup, static void tri_persp_coeff( struct setup_context *setup,
struct tgsi_interp_coef *coef, unsigned attrib,
uint vertSlot, uint i) uint vertSlot)
{ {
/* premultiply by 1/w (v[0][3] is always W): unsigned i;
*/ for (i = 0; i < NUM_CHANNELS; ++i) {
float mina = setup->vmin[vertSlot][i] * setup->vmin[0][3]; /* premultiply by 1/w (v[0][3] is always W):
float mida = setup->vmid[vertSlot][i] * setup->vmid[0][3]; */
float maxa = setup->vmax[vertSlot][i] * setup->vmax[0][3]; float mina = setup->vmin[vertSlot][i] * setup->vmin[0][3];
float botda = mida - mina; float mida = setup->vmid[vertSlot][i] * setup->vmid[0][3];
float majda = maxa - mina; float maxa = setup->vmax[vertSlot][i] * setup->vmax[0][3];
float a = setup->ebot.dy * majda - botda * setup->emaj.dy; float botda = mida - mina;
float b = setup->emaj.dx * botda - majda * setup->ebot.dx; float majda = maxa - mina;
float dadx = a * setup->oneoverarea; float a = setup->ebot.dy * majda - botda * setup->emaj.dy;
float dady = b * setup->oneoverarea; float b = setup->emaj.dx * botda - majda * setup->ebot.dx;
float dadx = a * setup->oneoverarea;
float dady = b * setup->oneoverarea;
/* /*
debug_printf("tri persp %d,%d: %f %f %f\n", vertSlot, i, debug_printf("tri persp %d,%d: %f %f %f\n", vertSlot, i,
setup->vmin[vertSlot][i], setup->vmin[vertSlot][i],
setup->vmid[vertSlot][i], setup->vmid[vertSlot][i],
setup->vmax[vertSlot][i] setup->vmax[vertSlot][i]
); );
*/ */
assert(i <= 3); assert(i <= 3);
coef->dadx[i] = dadx; setup->coef.dadx[attrib][i] = dadx;
coef->dady[i] = dady; setup->coef.dady[attrib][i] = dady;
coef->a0[i] = (mina - setup->coef.a0[attrib][i] = (mina -
(dadx * (setup->vmin[0][0] - 0.5f) + (dadx * (setup->vmin[0][0] - 0.5f) +
dady * (setup->vmin[0][1] - 0.5f))); dady * (setup->vmin[0][1] - 0.5f)));
}
} }
@@ -501,21 +572,21 @@ static void
setup_fragcoord_coeff(struct setup_context *setup, uint slot) setup_fragcoord_coeff(struct setup_context *setup, uint slot)
{ {
/*X*/ /*X*/
setup->coef[slot].a0[0] = 0; setup->coef.a0[slot][0] = 0;
setup->coef[slot].dadx[0] = 1.0; setup->coef.dadx[slot][0] = 1.0;
setup->coef[slot].dady[0] = 0.0; setup->coef.dady[slot][0] = 0.0;
/*Y*/ /*Y*/
setup->coef[slot].a0[1] = 0.0; setup->coef.a0[slot][1] = 0.0;
setup->coef[slot].dadx[1] = 0.0; setup->coef.dadx[slot][1] = 0.0;
setup->coef[slot].dady[1] = 1.0; setup->coef.dady[slot][1] = 1.0;
/*Z*/ /*Z*/
setup->coef[slot].a0[2] = setup->posCoef.a0[2]; setup->coef.a0[slot][2] = setup->posCoef.a0[2];
setup->coef[slot].dadx[2] = setup->posCoef.dadx[2]; setup->coef.dadx[slot][2] = setup->posCoef.dadx[2];
setup->coef[slot].dady[2] = setup->posCoef.dady[2]; setup->coef.dady[slot][2] = setup->posCoef.dady[2];
/*W*/ /*W*/
setup->coef[slot].a0[3] = setup->posCoef.a0[3]; setup->coef.a0[slot][3] = setup->posCoef.a0[3];
setup->coef[slot].dadx[3] = setup->posCoef.dadx[3]; setup->coef.dadx[slot][3] = setup->posCoef.dadx[3];
setup->coef[slot].dady[3] = setup->posCoef.dady[3]; setup->coef.dady[slot][3] = setup->posCoef.dady[3];
} }
@@ -533,27 +604,23 @@ static void setup_tri_coefficients( struct setup_context *setup )
/* z and w are done by linear interpolation: /* z and w are done by linear interpolation:
*/ */
tri_linear_coeff(setup, &setup->posCoef, 0, 2); tri_pos_coeff(setup, 0, 2);
tri_linear_coeff(setup, &setup->posCoef, 0, 3); tri_pos_coeff(setup, 0, 3);
/* setup interpolation for all the remaining attributes: /* setup interpolation for all the remaining attributes:
*/ */
for (fragSlot = 0; fragSlot < lpfs->info.num_inputs; fragSlot++) { for (fragSlot = 0; fragSlot < lpfs->info.num_inputs; fragSlot++) {
const uint vertSlot = vinfo->attrib[fragSlot].src_index; const uint vertSlot = vinfo->attrib[fragSlot].src_index;
uint j;
switch (vinfo->attrib[fragSlot].interp_mode) { switch (vinfo->attrib[fragSlot].interp_mode) {
case INTERP_CONSTANT: case INTERP_CONSTANT:
for (j = 0; j < NUM_CHANNELS; j++) const_coeff(setup, fragSlot, vertSlot);
const_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
break; break;
case INTERP_LINEAR: case INTERP_LINEAR:
for (j = 0; j < NUM_CHANNELS; j++) tri_linear_coeff(setup, fragSlot, vertSlot);
tri_linear_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
break; break;
case INTERP_PERSPECTIVE: case INTERP_PERSPECTIVE:
for (j = 0; j < NUM_CHANNELS; j++) tri_persp_coeff(setup, fragSlot, vertSlot);
tri_persp_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
break; break;
case INTERP_POS: case INTERP_POS:
setup_fragcoord_coeff(setup, fragSlot); setup_fragcoord_coeff(setup, fragSlot);
@@ -563,9 +630,9 @@ static void setup_tri_coefficients( struct setup_context *setup )
} }
if (lpfs->info.input_semantic_name[fragSlot] == TGSI_SEMANTIC_FACE) { if (lpfs->info.input_semantic_name[fragSlot] == TGSI_SEMANTIC_FACE) {
setup->coef[fragSlot].a0[0] = 1.0f - setup->facing; setup->coef.a0[fragSlot][0] = 1.0f - setup->facing;
setup->coef[fragSlot].dadx[0] = 0.0; setup->coef.dadx[fragSlot][0] = 0.0;
setup->coef[fragSlot].dady[0] = 0.0; setup->coef.dady[fragSlot][0] = 0.0;
} }
} }
} }
@@ -769,18 +836,40 @@ void setup_tri( struct setup_context *setup,
* for a line. * for a line.
*/ */
static void static void
line_linear_coeff(const struct setup_context *setup, linear_pos_coeff(struct setup_context *setup,
struct tgsi_interp_coef *coef, uint vertSlot, uint i)
uint vertSlot, uint i)
{ {
const float da = setup->vmax[vertSlot][i] - setup->vmin[vertSlot][i]; const float da = setup->vmax[vertSlot][i] - setup->vmin[vertSlot][i];
const float dadx = da * setup->emaj.dx * setup->oneoverarea; const float dadx = da * setup->emaj.dx * setup->oneoverarea;
const float dady = da * setup->emaj.dy * setup->oneoverarea; const float dady = da * setup->emaj.dy * setup->oneoverarea;
coef->dadx[i] = dadx; setup->posCoef.dadx[i] = dadx;
coef->dady[i] = dady; setup->posCoef.dady[i] = dady;
coef->a0[i] = (setup->vmin[vertSlot][i] - setup->posCoef.a0[i] = (setup->vmin[vertSlot][i] -
(dadx * (setup->vmin[0][0] - 0.5f) + (dadx * (setup->vmin[0][0] - 0.5f) +
dady * (setup->vmin[0][1] - 0.5f))); dady * (setup->vmin[0][1] - 0.5f)));
}
/**
* Compute a0, dadx and dady for a linearly interpolated coefficient,
* for a line.
*/
static void
line_linear_coeff(struct setup_context *setup,
unsigned attrib,
uint vertSlot)
{
unsigned i;
for (i = 0; i < NUM_CHANNELS; ++i) {
const float da = setup->vmax[vertSlot][i] - setup->vmin[vertSlot][i];
const float dadx = da * setup->emaj.dx * setup->oneoverarea;
const float dady = da * setup->emaj.dy * setup->oneoverarea;
setup->coef.dadx[attrib][i] = dadx;
setup->coef.dady[attrib][i] = dady;
setup->coef.a0[attrib][i] = (setup->vmin[vertSlot][i] -
(dadx * (setup->vmin[0][0] - 0.5f) +
dady * (setup->vmin[0][1] - 0.5f)));
}
} }
@@ -789,21 +878,24 @@ line_linear_coeff(const struct setup_context *setup,
* for a line. * for a line.
*/ */
static void static void
line_persp_coeff(const struct setup_context *setup, line_persp_coeff(struct setup_context *setup,
struct tgsi_interp_coef *coef, unsigned attrib,
uint vertSlot, uint i) uint vertSlot)
{ {
/* XXX double-check/verify this arithmetic */ unsigned i;
const float a0 = setup->vmin[vertSlot][i] * setup->vmin[0][3]; for (i = 0; i < NUM_CHANNELS; ++i) {
const float a1 = setup->vmax[vertSlot][i] * setup->vmax[0][3]; /* XXX double-check/verify this arithmetic */
const float da = a1 - a0; const float a0 = setup->vmin[vertSlot][i] * setup->vmin[0][3];
const float dadx = da * setup->emaj.dx * setup->oneoverarea; const float a1 = setup->vmax[vertSlot][i] * setup->vmax[0][3];
const float dady = da * setup->emaj.dy * setup->oneoverarea; const float da = a1 - a0;
coef->dadx[i] = dadx; const float dadx = da * setup->emaj.dx * setup->oneoverarea;
coef->dady[i] = dady; const float dady = da * setup->emaj.dy * setup->oneoverarea;
coef->a0[i] = (setup->vmin[vertSlot][i] - setup->coef.dadx[attrib][i] = dadx;
(dadx * (setup->vmin[0][0] - 0.5f) + setup->coef.dady[attrib][i] = dady;
dady * (setup->vmin[0][1] - 0.5f))); setup->coef.a0[attrib][i] = (setup->vmin[vertSlot][i] -
(dadx * (setup->vmin[0][0] - 0.5f) +
dady * (setup->vmin[0][1] - 0.5f)));
}
} }
@@ -841,27 +933,23 @@ setup_line_coefficients(struct setup_context *setup,
/* z and w are done by linear interpolation: /* z and w are done by linear interpolation:
*/ */
line_linear_coeff(setup, &setup->posCoef, 0, 2); linear_pos_coeff(setup, 0, 2);
line_linear_coeff(setup, &setup->posCoef, 0, 3); linear_pos_coeff(setup, 0, 3);
/* setup interpolation for all the remaining attributes: /* setup interpolation for all the remaining attributes:
*/ */
for (fragSlot = 0; fragSlot < lpfs->info.num_inputs; fragSlot++) { for (fragSlot = 0; fragSlot < lpfs->info.num_inputs; fragSlot++) {
const uint vertSlot = vinfo->attrib[fragSlot].src_index; const uint vertSlot = vinfo->attrib[fragSlot].src_index;
uint j;
switch (vinfo->attrib[fragSlot].interp_mode) { switch (vinfo->attrib[fragSlot].interp_mode) {
case INTERP_CONSTANT: case INTERP_CONSTANT:
for (j = 0; j < NUM_CHANNELS; j++) const_coeff(setup, fragSlot, vertSlot);
const_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
break; break;
case INTERP_LINEAR: case INTERP_LINEAR:
for (j = 0; j < NUM_CHANNELS; j++) line_linear_coeff(setup, fragSlot, vertSlot);
line_linear_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
break; break;
case INTERP_PERSPECTIVE: case INTERP_PERSPECTIVE:
for (j = 0; j < NUM_CHANNELS; j++) line_persp_coeff(setup, fragSlot, vertSlot);
line_persp_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
break; break;
case INTERP_POS: case INTERP_POS:
setup_fragcoord_coeff(setup, fragSlot); setup_fragcoord_coeff(setup, fragSlot);
@@ -871,9 +959,9 @@ setup_line_coefficients(struct setup_context *setup,
} }
if (lpfs->info.input_semantic_name[fragSlot] == TGSI_SEMANTIC_FACE) { if (lpfs->info.input_semantic_name[fragSlot] == TGSI_SEMANTIC_FACE) {
setup->coef[fragSlot].a0[0] = 1.0f - setup->facing; setup->coef.a0[fragSlot][0] = 1.0f - setup->facing;
setup->coef[fragSlot].dadx[0] = 0.0; setup->coef.dadx[fragSlot][0] = 0.0;
setup->coef[fragSlot].dady[0] = 0.0; setup->coef.dady[fragSlot][0] = 0.0;
} }
} }
return TRUE; return TRUE;
@@ -1027,15 +1115,17 @@ setup_line(struct setup_context *setup,
static void static void
point_persp_coeff(const struct setup_context *setup, point_persp_coeff(struct setup_context *setup,
const float (*vert)[4], const float (*vert)[4],
struct tgsi_interp_coef *coef, unsigned attrib,
uint vertSlot, uint i) uint vertSlot)
{ {
assert(i <= 3); unsigned i;
coef->dadx[i] = 0.0F; for(i = 0; i < NUM_CHANNELS; ++i) {
coef->dady[i] = 0.0F; setup->coef.dadx[attrib][i] = 0.0F;
coef->a0[i] = vert[vertSlot][i] * vert[0][3]; setup->coef.dady[attrib][i] = 0.0F;
setup->coef.a0[attrib][i] = vert[vertSlot][i] * vert[0][3];
}
} }
@@ -1090,24 +1180,20 @@ setup_point( struct setup_context *setup,
setup->vprovoke = v0; setup->vprovoke = v0;
/* setup Z, W */ /* setup Z, W */
const_coeff(setup, &setup->posCoef, 0, 2); const_pos_coeff(setup, 0, 2);
const_coeff(setup, &setup->posCoef, 0, 3); const_pos_coeff(setup, 0, 3);
for (fragSlot = 0; fragSlot < lpfs->info.num_inputs; fragSlot++) { for (fragSlot = 0; fragSlot < lpfs->info.num_inputs; fragSlot++) {
const uint vertSlot = vinfo->attrib[fragSlot].src_index; const uint vertSlot = vinfo->attrib[fragSlot].src_index;
uint j;
switch (vinfo->attrib[fragSlot].interp_mode) { switch (vinfo->attrib[fragSlot].interp_mode) {
case INTERP_CONSTANT: case INTERP_CONSTANT:
/* fall-through */ /* fall-through */
case INTERP_LINEAR: case INTERP_LINEAR:
for (j = 0; j < NUM_CHANNELS; j++) const_coeff(setup, fragSlot, vertSlot);
const_coeff(setup, &setup->coef[fragSlot], vertSlot, j);
break; break;
case INTERP_PERSPECTIVE: case INTERP_PERSPECTIVE:
for (j = 0; j < NUM_CHANNELS; j++) point_persp_coeff(setup, setup->vprovoke, fragSlot, vertSlot);
point_persp_coeff(setup, setup->vprovoke,
&setup->coef[fragSlot], vertSlot, j);
break; break;
case INTERP_POS: case INTERP_POS:
setup_fragcoord_coeff(setup, fragSlot); setup_fragcoord_coeff(setup, fragSlot);
@@ -1117,9 +1203,9 @@ setup_point( struct setup_context *setup,
} }
if (lpfs->info.input_semantic_name[fragSlot] == TGSI_SEMANTIC_FACE) { if (lpfs->info.input_semantic_name[fragSlot] == TGSI_SEMANTIC_FACE) {
setup->coef[fragSlot].a0[0] = 1.0f - setup->facing; setup->coef.a0[fragSlot][0] = 1.0f - setup->facing;
setup->coef[fragSlot].dadx[0] = 0.0; setup->coef.dadx[fragSlot][0] = 0.0;
setup->coef[fragSlot].dady[0] = 0.0; setup->coef.dady[fragSlot][0] = 0.0;
} }
} }
@@ -1287,7 +1373,7 @@ struct setup_context *setup_create_context( struct llvmpipe_context *llvmpipe )
setup->llvmpipe = llvmpipe; setup->llvmpipe = llvmpipe;
for (i = 0; i < MAX_QUADS; i++) { for (i = 0; i < MAX_QUADS; i++) {
setup->quad[i].coef = setup->coef; setup->quad[i].coef = &setup->coef;
setup->quad[i].posCoef = &setup->posCoef; setup->quad[i].posCoef = &setup->posCoef;
} }

10
src/gallium/drivers/llvmpipe/lp_state.h
View File

@@ -59,11 +59,11 @@ struct vertex_info;
typedef void typedef void
(*lp_shader_fs_func)(void *pos, (*lp_shader_fs_func)(const void *pos,
void *a0, const void *a0,
void *dadx, const void *dadx,
void *dady, const void *dady,
void *consts, const void *consts,
void *outputs, void *outputs,
struct tgsi_sampler **samplers); struct tgsi_sampler **samplers);