* of improperly weighted linear-filtered textures.
* The tests/texwrap.c demo is a good test.
*/
-static INLINE float
+static inline float
frac(float f)
{
return f - floorf(f);
/**
* Linear interpolation macro
*/
-static INLINE float
+static inline float
lerp(float a, float v0, float v1)
{
return v0 + a * (v1 - v0);
* optimization! If we find that's not true on some systems, convert
* to a macro.
*/
-static INLINE float
+static inline float
lerp_2d(float a, float b,
float v00, float v10, float v01, float v11)
{
/**
* As above, but 3D interpolation of 8 values.
*/
-static INLINE float
+static inline float
lerp_3d(float a, float b, float c,
float v000, float v100, float v010, float v110,
float v001, float v101, float v011, float v111)
* value. To avoid that problem we add a large multiple of the size
* (rather than using a conditional).
*/
-static INLINE int
+static inline int
repeat(int coord, unsigned size)
{
return (coord + size * 1024) % size;
* \param icoord returns the integer texcoords
*/
static void
-wrap_nearest_repeat(float s, unsigned size, int *icoord)
+wrap_nearest_repeat(float s, unsigned size, int offset, int *icoord)
{
/* s limited to [0,1) */
/* i limited to [0,size-1] */
- int i = util_ifloor(s * size);
- *icoord = repeat(i, size);
+ const int i = util_ifloor(s * size);
+ *icoord = repeat(i + offset, size);
}
static void
-wrap_nearest_clamp(float s, unsigned size, int *icoord)
+wrap_nearest_clamp(float s, unsigned size, int offset, int *icoord)
{
/* s limited to [0,1] */
/* i limited to [0,size-1] */
+ s *= size;
+ s += offset;
if (s <= 0.0F)
*icoord = 0;
- else if (s >= 1.0F)
+ else if (s >= size)
*icoord = size - 1;
else
- *icoord = util_ifloor(s * size);
+ *icoord = util_ifloor(s);
}
static void
-wrap_nearest_clamp_to_edge(float s, unsigned size, int *icoord)
+wrap_nearest_clamp_to_edge(float s, unsigned size, int offset, int *icoord)
{
/* s limited to [min,max] */
/* i limited to [0, size-1] */
- const float min = 1.0F / (2.0F * size);
- const float max = 1.0F - min;
+ const float min = 0.5F;
+ const float max = (float)size - 0.5F;
+
+ s *= size;
+ s += offset;
+
if (s < min)
*icoord = 0;
else if (s > max)
*icoord = size - 1;
else
- *icoord = util_ifloor(s * size);
+ *icoord = util_ifloor(s);
}
static void
-wrap_nearest_clamp_to_border(float s, unsigned size, int *icoord)
+wrap_nearest_clamp_to_border(float s, unsigned size, int offset, int *icoord)
{
/* s limited to [min,max] */
/* i limited to [-1, size] */
- const float min = -1.0F / (2.0F * size);
- const float max = 1.0F - min;
+ const float min = -0.5F;
+ const float max = size + 0.5F;
+
+ s *= size;
+ s += offset;
if (s <= min)
*icoord = -1;
else if (s >= max)
*icoord = size;
else
- *icoord = util_ifloor(s * size);
+ *icoord = util_ifloor(s);
}
-
static void
-wrap_nearest_mirror_repeat(float s, unsigned size, int *icoord)
+wrap_nearest_mirror_repeat(float s, unsigned size, int offset, int *icoord)
{
const float min = 1.0F / (2.0F * size);
const float max = 1.0F - min;
- const int flr = util_ifloor(s);
- float u = frac(s);
+ int flr;
+ float u;
+
+ s += (float)offset / size;
+ flr = util_ifloor(s);
+ u = frac(s);
if (flr & 1)
u = 1.0F - u;
if (u < min)
static void
-wrap_nearest_mirror_clamp(float s, unsigned size, int *icoord)
+wrap_nearest_mirror_clamp(float s, unsigned size, int offset, int *icoord)
{
/* s limited to [0,1] */
/* i limited to [0,size-1] */
- const float u = fabsf(s);
+ const float u = fabsf(s * size + offset);
if (u <= 0.0F)
*icoord = 0;
- else if (u >= 1.0F)
+ else if (u >= size)
*icoord = size - 1;
else
- *icoord = util_ifloor(u * size);
+ *icoord = util_ifloor(u);
}
static void
-wrap_nearest_mirror_clamp_to_edge(float s, unsigned size, int *icoord)
+wrap_nearest_mirror_clamp_to_edge(float s, unsigned size, int offset, int *icoord)
{
/* s limited to [min,max] */
/* i limited to [0, size-1] */
- const float min = 1.0F / (2.0F * size);
- const float max = 1.0F - min;
- const float u = fabsf(s);
+ const float min = 0.5F;
+ const float max = (float)size - 0.5F;
+ const float u = fabsf(s * size + offset);
+
if (u < min)
*icoord = 0;
else if (u > max)
*icoord = size - 1;
else
- *icoord = util_ifloor(u * size);
+ *icoord = util_ifloor(u);
}
static void
-wrap_nearest_mirror_clamp_to_border(float s, unsigned size, int *icoord)
+wrap_nearest_mirror_clamp_to_border(float s, unsigned size, int offset, int *icoord)
{
- /* s limited to [min,max] */
- /* i limited to [0, size-1] */
- const float min = -1.0F / (2.0F * size);
- const float max = 1.0F - min;
- const float u = fabsf(s);
+ /* u limited to [-0.5, size-0.5] */
+ const float min = -0.5F;
+ const float max = (float)size + 0.5F;
+ const float u = fabsf(s * size + offset);
+
if (u < min)
*icoord = -1;
else if (u > max)
*icoord = size;
else
- *icoord = util_ifloor(u * size);
+ *icoord = util_ifloor(u);
}
* \param icoord returns the computed integer texture coord
*/
static void
-wrap_linear_repeat(float s, unsigned size,
+wrap_linear_repeat(float s, unsigned size, int offset,
int *icoord0, int *icoord1, float *w)
{
- float u = s * size - 0.5F;
- *icoord0 = repeat(util_ifloor(u), size);
+ const float u = s * size - 0.5F;
+ *icoord0 = repeat(util_ifloor(u) + offset, size);
*icoord1 = repeat(*icoord0 + 1, size);
*w = frac(u);
}
static void
-wrap_linear_clamp(float s, unsigned size,
+wrap_linear_clamp(float s, unsigned size, int offset,
int *icoord0, int *icoord1, float *w)
{
- float u = CLAMP(s, 0.0F, 1.0F);
- u = u * size - 0.5f;
+ const float u = CLAMP(s * size + offset, 0.0F, (float)size) - 0.5f;
+
*icoord0 = util_ifloor(u);
*icoord1 = *icoord0 + 1;
*w = frac(u);
static void
-wrap_linear_clamp_to_edge(float s, unsigned size,
+wrap_linear_clamp_to_edge(float s, unsigned size, int offset,
int *icoord0, int *icoord1, float *w)
{
- float u = CLAMP(s, 0.0F, 1.0F);
- u = u * size - 0.5f;
+ const float u = CLAMP(s * size + offset, 0.0F, (float)size) - 0.5f;
*icoord0 = util_ifloor(u);
*icoord1 = *icoord0 + 1;
if (*icoord0 < 0)
static void
-wrap_linear_clamp_to_border(float s, unsigned size,
+wrap_linear_clamp_to_border(float s, unsigned size, int offset,
int *icoord0, int *icoord1, float *w)
{
- const float min = -1.0F / (2.0F * size);
- const float max = 1.0F - min;
- float u = CLAMP(s, min, max);
- u = u * size - 0.5f;
+ const float min = -0.5F;
+ const float max = (float)size + 0.5F;
+ const float u = CLAMP(s * size + offset, min, max) - 0.5f;
*icoord0 = util_ifloor(u);
*icoord1 = *icoord0 + 1;
*w = frac(u);
static void
-wrap_linear_mirror_repeat(float s, unsigned size,
+wrap_linear_mirror_repeat(float s, unsigned size, int offset,
int *icoord0, int *icoord1, float *w)
{
- const int flr = util_ifloor(s);
- float u = frac(s);
+ int flr;
+ float u;
+
+ s += (float)offset / size;
+ flr = util_ifloor(s);
+ u = frac(s);
if (flr & 1)
u = 1.0F - u;
u = u * size - 0.5F;
static void
-wrap_linear_mirror_clamp(float s, unsigned size,
+wrap_linear_mirror_clamp(float s, unsigned size, int offset,
int *icoord0, int *icoord1, float *w)
{
- float u = fabsf(s);
- if (u >= 1.0F)
+ float u = fabsf(s * size + offset);
+ if (u >= size)
u = (float) size;
- else
- u *= size;
u -= 0.5F;
*icoord0 = util_ifloor(u);
*icoord1 = *icoord0 + 1;
static void
-wrap_linear_mirror_clamp_to_edge(float s, unsigned size,
+wrap_linear_mirror_clamp_to_edge(float s, unsigned size, int offset,
int *icoord0, int *icoord1, float *w)
{
- float u = fabsf(s);
- if (u >= 1.0F)
+ float u = fabsf(s * size + offset);
+ if (u >= size)
u = (float) size;
- else
- u *= size;
u -= 0.5F;
*icoord0 = util_ifloor(u);
*icoord1 = *icoord0 + 1;
static void
-wrap_linear_mirror_clamp_to_border(float s, unsigned size,
+wrap_linear_mirror_clamp_to_border(float s, unsigned size, int offset,
int *icoord0, int *icoord1, float *w)
{
- const float min = -1.0F / (2.0F * size);
- const float max = 1.0F - min;
- float u = fabsf(s);
- if (u <= min)
- u = min * size;
- else if (u >= max)
- u = max * size;
- else
- u *= size;
- u -= 0.5F;
+ const float min = -0.5F;
+ const float max = size + 0.5F;
+ const float t = fabsf(s * size + offset);
+ const float u = CLAMP(t, min, max) - 0.5F;
*icoord0 = util_ifloor(u);
*icoord1 = *icoord0 + 1;
*w = frac(u);
* PIPE_TEX_WRAP_CLAMP for nearest sampling, unnormalized coords.
*/
static void
-wrap_nearest_unorm_clamp(float s, unsigned size, int *icoord)
+wrap_nearest_unorm_clamp(float s, unsigned size, int offset, int *icoord)
{
- int i = util_ifloor(s);
- *icoord = CLAMP(i, 0, (int) size-1);
+ const int i = util_ifloor(s);
+ *icoord = CLAMP(i + offset, 0, (int) size-1);
}
* PIPE_TEX_WRAP_CLAMP_TO_BORDER for nearest sampling, unnormalized coords.
*/
static void
-wrap_nearest_unorm_clamp_to_border(float s, unsigned size, int *icoord)
+wrap_nearest_unorm_clamp_to_border(float s, unsigned size, int offset, int *icoord)
{
- *icoord = util_ifloor( CLAMP(s, -0.5F, (float) size + 0.5F) );
+ *icoord = util_ifloor( CLAMP(s + offset, -0.5F, (float) size + 0.5F) );
}
* PIPE_TEX_WRAP_CLAMP_TO_EDGE for nearest sampling, unnormalized coords.
*/
static void
-wrap_nearest_unorm_clamp_to_edge(float s, unsigned size, int *icoord)
+wrap_nearest_unorm_clamp_to_edge(float s, unsigned size, int offset, int *icoord)
{
- *icoord = util_ifloor( CLAMP(s, 0.5F, (float) size - 0.5F) );
+ *icoord = util_ifloor( CLAMP(s + offset, 0.5F, (float) size - 0.5F) );
}
* PIPE_TEX_WRAP_CLAMP for linear sampling, unnormalized coords.
*/
static void
-wrap_linear_unorm_clamp(float s, unsigned size,
+wrap_linear_unorm_clamp(float s, unsigned size, int offset,
int *icoord0, int *icoord1, float *w)
{
/* Not exactly what the spec says, but it matches NVIDIA output */
- float u = CLAMP(s - 0.5F, 0.0f, (float) size - 1.0f);
+ const float u = CLAMP(s + offset - 0.5F, 0.0f, (float) size - 1.0f);
*icoord0 = util_ifloor(u);
*icoord1 = *icoord0 + 1;
*w = frac(u);
* PIPE_TEX_WRAP_CLAMP_TO_BORDER for linear sampling, unnormalized coords.
*/
static void
-wrap_linear_unorm_clamp_to_border(float s, unsigned size,
+wrap_linear_unorm_clamp_to_border(float s, unsigned size, int offset,
int *icoord0, int *icoord1, float *w)
{
- float u = CLAMP(s, -0.5F, (float) size + 0.5F);
- u -= 0.5F;
+ const float u = CLAMP(s + offset, -0.5F, (float) size + 0.5F) - 0.5F;
*icoord0 = util_ifloor(u);
*icoord1 = *icoord0 + 1;
if (*icoord1 > (int) size - 1)
* PIPE_TEX_WRAP_CLAMP_TO_EDGE for linear sampling, unnormalized coords.
*/
static void
-wrap_linear_unorm_clamp_to_edge(float s, unsigned size,
+wrap_linear_unorm_clamp_to_edge(float s, unsigned size, int offset,
int *icoord0, int *icoord1, float *w)
{
- float u = CLAMP(s, +0.5F, (float) size - 0.5F);
- u -= 0.5F;
+ const float u = CLAMP(s + offset, +0.5F, (float) size - 0.5F) - 0.5F;
*icoord0 = util_ifloor(u);
*icoord1 = *icoord0 + 1;
if (*icoord1 > (int) size - 1)
/**
* Do coordinate to array index conversion. For array textures.
*/
-static INLINE void
-wrap_array_layer(float coord, unsigned size, int *layer)
+static inline int
+coord_to_layer(float coord, unsigned first_layer, unsigned last_layer)
{
- int c = util_ifloor(coord + 0.5F);
- *layer = CLAMP(c, 0, (int) size - 1);
+ const int c = util_ifloor(coord + 0.5F);
+ return CLAMP(c, (int)first_layer, (int)last_layer);
}
const float p[TGSI_QUAD_SIZE])
{
const struct pipe_resource *texture = sview->base.texture;
- float dsdx = fabsf(s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]);
- float dsdy = fabsf(s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]);
- float rho = MAX2(dsdx, dsdy) * u_minify(texture->width0, sview->base.u.tex.first_level);
+ const float dsdx = fabsf(s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]);
+ const float dsdy = fabsf(s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]);
+ const float rho = MAX2(dsdx, dsdy) * u_minify(texture->width0, sview->base.u.tex.first_level);
return util_fast_log2(rho);
}
const float p[TGSI_QUAD_SIZE])
{
const struct pipe_resource *texture = sview->base.texture;
- float dsdx = fabsf(s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]);
- float dsdy = fabsf(s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]);
- float dtdx = fabsf(t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT]);
- float dtdy = fabsf(t[QUAD_TOP_LEFT] - t[QUAD_BOTTOM_LEFT]);
- float maxx = MAX2(dsdx, dsdy) * u_minify(texture->width0, sview->base.u.tex.first_level);
- float maxy = MAX2(dtdx, dtdy) * u_minify(texture->height0, sview->base.u.tex.first_level);
- float rho = MAX2(maxx, maxy);
+ const float dsdx = fabsf(s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]);
+ const float dsdy = fabsf(s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]);
+ const float dtdx = fabsf(t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT]);
+ const float dtdy = fabsf(t[QUAD_TOP_LEFT] - t[QUAD_BOTTOM_LEFT]);
+ const float maxx = MAX2(dsdx, dsdy) * u_minify(texture->width0, sview->base.u.tex.first_level);
+ const float maxy = MAX2(dtdx, dtdy) * u_minify(texture->height0, sview->base.u.tex.first_level);
+ const float rho = MAX2(maxx, maxy);
return util_fast_log2(rho);
}
const float p[TGSI_QUAD_SIZE])
{
const struct pipe_resource *texture = sview->base.texture;
- float dsdx = fabsf(s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]);
- float dsdy = fabsf(s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]);
- float dtdx = fabsf(t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT]);
- float dtdy = fabsf(t[QUAD_TOP_LEFT] - t[QUAD_BOTTOM_LEFT]);
- float dpdx = fabsf(p[QUAD_BOTTOM_RIGHT] - p[QUAD_BOTTOM_LEFT]);
- float dpdy = fabsf(p[QUAD_TOP_LEFT] - p[QUAD_BOTTOM_LEFT]);
- float maxx = MAX2(dsdx, dsdy) * u_minify(texture->width0, sview->base.u.tex.first_level);
- float maxy = MAX2(dtdx, dtdy) * u_minify(texture->height0, sview->base.u.tex.first_level);
- float maxz = MAX2(dpdx, dpdy) * u_minify(texture->depth0, sview->base.u.tex.first_level);
- float rho;
-
- rho = MAX2(maxx, maxy);
- rho = MAX2(rho, maxz);
+ const float dsdx = fabsf(s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]);
+ const float dsdy = fabsf(s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]);
+ const float dtdx = fabsf(t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT]);
+ const float dtdy = fabsf(t[QUAD_TOP_LEFT] - t[QUAD_BOTTOM_LEFT]);
+ const float dpdx = fabsf(p[QUAD_BOTTOM_RIGHT] - p[QUAD_BOTTOM_LEFT]);
+ const float dpdy = fabsf(p[QUAD_TOP_LEFT] - p[QUAD_BOTTOM_LEFT]);
+ const float maxx = MAX2(dsdx, dsdy) * u_minify(texture->width0, sview->base.u.tex.first_level);
+ const float maxy = MAX2(dtdx, dtdy) * u_minify(texture->height0, sview->base.u.tex.first_level);
+ const float maxz = MAX2(dpdx, dpdy) * u_minify(texture->depth0, sview->base.u.tex.first_level);
+ const float rho = MAX3(maxx, maxy, maxz);
return util_fast_log2(rho);
}
-static INLINE const float *
+static inline const float *
get_texel_2d_no_border(const struct sp_sampler_view *sp_sview,
union tex_tile_address addr, int x, int y)
{
}
-static INLINE const float *
+static inline const float *
get_texel_2d(const struct sp_sampler_view *sp_sview,
const struct sp_sampler *sp_samp,
union tex_tile_address addr, int x, int y)
{
const struct pipe_resource *texture = sp_sview->base.texture;
- unsigned level = addr.bits.level;
+ const unsigned level = addr.bits.level;
if (x < 0 || x >= (int) u_minify(texture->width0, level) ||
y < 0 || y >= (int) u_minify(texture->height0, level)) {
PIPE_TEX_FACE_POS_Y, PIPE_TEX_FACE_NEG_Y }
};
-static INLINE unsigned
+static inline unsigned
get_next_face(unsigned face, int idx)
{
return face_array[face][idx];
* return a new xcoord based on old face, old coords, cube size
* and fall_off_index (0 for x-, 1 for x+, 2 for y-, 3 for y+)
*/
-static INLINE int
+static inline int
get_next_xcoord(unsigned face, unsigned fall_off_index, int max, int xc, int yc)
{
if ((face == 0 && fall_off_index != 1) ||
* return a new ycoord based on old face, old coords, cube size
* and fall_off_index (0 for x-, 1 for x+, 2 for y-, 3 for y+)
*/
-static INLINE int
+static inline int
get_next_ycoord(unsigned face, unsigned fall_off_index, int max, int xc, int yc)
{
if ((fall_off_index <= 1) && (face <= 1 || face >= 4)) {
}
-static INLINE const float *
-get_texel_cube_seamless(const struct sp_sampler_view *sp_sview,
- union tex_tile_address addr, int x, int y,
- float *corner)
-{
- const struct pipe_resource *texture = sp_sview->base.texture;
- unsigned level = addr.bits.level;
- unsigned face = addr.bits.face;
- int new_x, new_y, max_x;
-
- max_x = (int) u_minify(texture->width0, level);
-
- assert(texture->width0 == texture->height0);
- new_x = x;
- new_y = y;
-
- /* change the face */
- if (x < 0) {
- /*
- * Cheat with corners. They are difficult and I believe because we don't get
- * per-pixel faces we can actually have multiple corner texels per pixel,
- * which screws things up majorly in any case (as the per spec behavior is
- * to average the 3 remaining texels, which we might not have).
- * Hence just make sure that the 2nd coord is clamped, will simply pick the
- * sample which would have fallen off the x coord, but not y coord.
- * So the filter weight of the samples will be wrong, but at least this
- * ensures that only valid texels near the corner are used.
- */
- if (y < 0 || y >= max_x) {
- y = CLAMP(y, 0, max_x - 1);
- }
- new_x = get_next_xcoord(face, 0, max_x -1, x, y);
- new_y = get_next_ycoord(face, 0, max_x -1, x, y);
- face = get_next_face(face, 0);
- } else if (x >= max_x) {
- if (y < 0 || y >= max_x) {
- y = CLAMP(y, 0, max_x - 1);
- }
- new_x = get_next_xcoord(face, 1, max_x -1, x, y);
- new_y = get_next_ycoord(face, 1, max_x -1, x, y);
- face = get_next_face(face, 1);
- } else if (y < 0) {
- new_x = get_next_xcoord(face, 2, max_x -1, x, y);
- new_y = get_next_ycoord(face, 2, max_x -1, x, y);
- face = get_next_face(face, 2);
- } else if (y >= max_x) {
- new_x = get_next_xcoord(face, 3, max_x -1, x, y);
- new_y = get_next_ycoord(face, 3, max_x -1, x, y);
- face = get_next_face(face, 3);
- }
-
- addr.bits.face = face;
- return get_texel_2d_no_border( sp_sview, addr, new_x, new_y );
-}
-
/* Gather a quad of adjacent texels within a tile:
*/
-static INLINE void
+static inline void
get_texel_quad_2d_no_border_single_tile(const struct sp_sampler_view *sp_sview,
union tex_tile_address addr,
unsigned x, unsigned y,
/* Gather a quad of potentially non-adjacent texels:
*/
-static INLINE void
+static inline void
get_texel_quad_2d_no_border(const struct sp_sampler_view *sp_sview,
union tex_tile_address addr,
int x0, int y0,
out[3] = get_texel_2d_no_border( sp_sview, addr, x1, y1 );
}
-/* Can involve a lot of unnecessary checks for border color:
- */
-static INLINE void
-get_texel_quad_2d(const struct sp_sampler_view *sp_sview,
- const struct sp_sampler *sp_samp,
- union tex_tile_address addr,
- int x0, int y0,
- int x1, int y1,
- const float *out[4])
-{
- out[0] = get_texel_2d( sp_sview, sp_samp, addr, x0, y0 );
- out[1] = get_texel_2d( sp_sview, sp_samp, addr, x1, y0 );
- out[3] = get_texel_2d( sp_sview, sp_samp, addr, x1, y1 );
- out[2] = get_texel_2d( sp_sview, sp_samp, addr, x0, y1 );
-}
-
-
/* 3d variants:
*/
-static INLINE const float *
+static inline const float *
get_texel_3d_no_border(const struct sp_sampler_view *sp_sview,
union tex_tile_address addr, int x, int y, int z)
{
}
-static INLINE const float *
+static inline const float *
get_texel_3d(const struct sp_sampler_view *sp_sview,
const struct sp_sampler *sp_samp,
union tex_tile_address addr, int x, int y, int z)
{
const struct pipe_resource *texture = sp_sview->base.texture;
- unsigned level = addr.bits.level;
+ const unsigned level = addr.bits.level;
if (x < 0 || x >= (int) u_minify(texture->width0, level) ||
y < 0 || y >= (int) u_minify(texture->height0, level) ||
/* Get texel pointer for 1D array texture */
-static INLINE const float *
+static inline const float *
get_texel_1d_array(const struct sp_sampler_view *sp_sview,
const struct sp_sampler *sp_samp,
union tex_tile_address addr, int x, int y)
{
const struct pipe_resource *texture = sp_sview->base.texture;
- unsigned level = addr.bits.level;
+ const unsigned level = addr.bits.level;
if (x < 0 || x >= (int) u_minify(texture->width0, level)) {
return sp_samp->base.border_color.f;
/* Get texel pointer for 2D array texture */
-static INLINE const float *
+static inline const float *
get_texel_2d_array(const struct sp_sampler_view *sp_sview,
const struct sp_sampler *sp_samp,
union tex_tile_address addr, int x, int y, int layer)
{
const struct pipe_resource *texture = sp_sview->base.texture;
- unsigned level = addr.bits.level;
+ const unsigned level = addr.bits.level;
assert(layer < (int) texture->array_size);
assert(layer >= 0);
}
+static inline const float *
+get_texel_cube_seamless(const struct sp_sampler_view *sp_sview,
+ union tex_tile_address addr, int x, int y,
+ float *corner, int layer, unsigned face)
+{
+ const struct pipe_resource *texture = sp_sview->base.texture;
+ const unsigned level = addr.bits.level;
+ int new_x, new_y, max_x;
+
+ max_x = (int) u_minify(texture->width0, level);
+
+ assert(texture->width0 == texture->height0);
+ new_x = x;
+ new_y = y;
+
+ /* change the face */
+ if (x < 0) {
+ /*
+ * Cheat with corners. They are difficult and I believe because we don't get
+ * per-pixel faces we can actually have multiple corner texels per pixel,
+ * which screws things up majorly in any case (as the per spec behavior is
+ * to average the 3 remaining texels, which we might not have).
+ * Hence just make sure that the 2nd coord is clamped, will simply pick the
+ * sample which would have fallen off the x coord, but not y coord.
+ * So the filter weight of the samples will be wrong, but at least this
+ * ensures that only valid texels near the corner are used.
+ */
+ if (y < 0 || y >= max_x) {
+ y = CLAMP(y, 0, max_x - 1);
+ }
+ new_x = get_next_xcoord(face, 0, max_x -1, x, y);
+ new_y = get_next_ycoord(face, 0, max_x -1, x, y);
+ face = get_next_face(face, 0);
+ } else if (x >= max_x) {
+ if (y < 0 || y >= max_x) {
+ y = CLAMP(y, 0, max_x - 1);
+ }
+ new_x = get_next_xcoord(face, 1, max_x -1, x, y);
+ new_y = get_next_ycoord(face, 1, max_x -1, x, y);
+ face = get_next_face(face, 1);
+ } else if (y < 0) {
+ new_x = get_next_xcoord(face, 2, max_x -1, x, y);
+ new_y = get_next_ycoord(face, 2, max_x -1, x, y);
+ face = get_next_face(face, 2);
+ } else if (y >= max_x) {
+ new_x = get_next_xcoord(face, 3, max_x -1, x, y);
+ new_y = get_next_ycoord(face, 3, max_x -1, x, y);
+ face = get_next_face(face, 3);
+ }
+
+ return get_texel_3d_no_border(sp_sview, addr, new_x, new_y, layer + face);
+}
+
+
/* Get texel pointer for cube array texture */
-static INLINE const float *
+static inline const float *
get_texel_cube_array(const struct sp_sampler_view *sp_sview,
const struct sp_sampler *sp_samp,
union tex_tile_address addr, int x, int y, int layer)
{
const struct pipe_resource *texture = sp_sview->base.texture;
- unsigned level = addr.bits.level;
+ const unsigned level = addr.bits.level;
assert(layer < (int) texture->array_size);
assert(layer >= 0);
* If level = 2, then we'll return 64 (the width at level=2).
* Return 1 if level > base_pot.
*/
-static INLINE unsigned
+static inline unsigned
pot_level_size(unsigned base_pot, unsigned level)
{
return (base_pot >= level) ? (1 << (base_pot - level)) : 1;
/* Some image-filter fastpaths:
*/
-static INLINE void
-img_filter_2d_linear_repeat_POT(struct sp_sampler_view *sp_sview,
- struct sp_sampler *sp_samp,
- float s,
- float t,
- float p,
- unsigned level,
- unsigned face_id,
+static inline void
+img_filter_2d_linear_repeat_POT(const struct sp_sampler_view *sp_sview,
+ const struct sp_sampler *sp_samp,
+ const struct img_filter_args *args,
float *rgba)
{
- unsigned xpot = pot_level_size(sp_sview->xpot, level);
- unsigned ypot = pot_level_size(sp_sview->ypot, level);
- int xmax = (xpot - 1) & (TEX_TILE_SIZE - 1); /* MIN2(TEX_TILE_SIZE, xpot) - 1; */
- int ymax = (ypot - 1) & (TEX_TILE_SIZE - 1); /* MIN2(TEX_TILE_SIZE, ypot) - 1; */
+ const unsigned xpot = pot_level_size(sp_sview->xpot, args->level);
+ const unsigned ypot = pot_level_size(sp_sview->ypot, args->level);
+ const int xmax = (xpot - 1) & (TEX_TILE_SIZE - 1); /* MIN2(TEX_TILE_SIZE, xpot) - 1; */
+ const int ymax = (ypot - 1) & (TEX_TILE_SIZE - 1); /* MIN2(TEX_TILE_SIZE, ypot) - 1; */
union tex_tile_address addr;
int c;
- float u = s * xpot - 0.5F;
- float v = t * ypot - 0.5F;
+ const float u = (args->s * xpot - 0.5F) + args->offset[0];
+ const float v = (args->t * ypot - 0.5F) + args->offset[1];
- int uflr = util_ifloor(u);
- int vflr = util_ifloor(v);
+ const int uflr = util_ifloor(u);
+ const int vflr = util_ifloor(v);
- float xw = u - (float)uflr;
- float yw = v - (float)vflr;
+ const float xw = u - (float)uflr;
+ const float yw = v - (float)vflr;
- int x0 = uflr & (xpot - 1);
- int y0 = vflr & (ypot - 1);
+ const int x0 = uflr & (xpot - 1);
+ const int y0 = vflr & (ypot - 1);
const float *tx[4];
addr.value = 0;
- addr.bits.level = level;
+ addr.bits.level = args->level;
+ addr.bits.z = sp_sview->base.u.tex.first_layer;
/* Can we fetch all four at once:
*/
get_texel_quad_2d_no_border_single_tile(sp_sview, addr, x0, y0, tx);
}
else {
- unsigned x1 = (x0 + 1) & (xpot - 1);
- unsigned y1 = (y0 + 1) & (ypot - 1);
+ const unsigned x1 = (x0 + 1) & (xpot - 1);
+ const unsigned y1 = (y0 + 1) & (ypot - 1);
get_texel_quad_2d_no_border(sp_sview, addr, x0, y0, x1, y1, tx);
}
/* interpolate R, G, B, A */
- for (c = 0; c < TGSI_QUAD_SIZE; c++) {
+ for (c = 0; c < TGSI_NUM_CHANNELS; c++) {
rgba[TGSI_NUM_CHANNELS*c] = lerp_2d(xw, yw,
tx[0][c], tx[1][c],
tx[2][c], tx[3][c]);
}
-static INLINE void
-img_filter_2d_nearest_repeat_POT(struct sp_sampler_view *sp_sview,
- struct sp_sampler *sp_samp,
- float s,
- float t,
- float p,
- unsigned level,
- unsigned face_id,
- float rgba[TGSI_QUAD_SIZE])
+static inline void
+img_filter_2d_nearest_repeat_POT(const struct sp_sampler_view *sp_sview,
+ const struct sp_sampler *sp_samp,
+ const struct img_filter_args *args,
+ float *rgba)
{
- unsigned xpot = pot_level_size(sp_sview->xpot, level);
- unsigned ypot = pot_level_size(sp_sview->ypot, level);
+ const unsigned xpot = pot_level_size(sp_sview->xpot, args->level);
+ const unsigned ypot = pot_level_size(sp_sview->ypot, args->level);
const float *out;
union tex_tile_address addr;
int c;
- float u = s * xpot;
- float v = t * ypot;
+ const float u = args->s * xpot + args->offset[0];
+ const float v = args->t * ypot + args->offset[1];
- int uflr = util_ifloor(u);
- int vflr = util_ifloor(v);
+ const int uflr = util_ifloor(u);
+ const int vflr = util_ifloor(v);
- int x0 = uflr & (xpot - 1);
- int y0 = vflr & (ypot - 1);
+ const int x0 = uflr & (xpot - 1);
+ const int y0 = vflr & (ypot - 1);
addr.value = 0;
- addr.bits.level = level;
+ addr.bits.level = args->level;
+ addr.bits.z = sp_sview->base.u.tex.first_layer;
out = get_texel_2d_no_border(sp_sview, addr, x0, y0);
- for (c = 0; c < TGSI_QUAD_SIZE; c++)
+ for (c = 0; c < TGSI_NUM_CHANNELS; c++)
rgba[TGSI_NUM_CHANNELS*c] = out[c];
if (DEBUG_TEX) {
}
-static INLINE void
-img_filter_2d_nearest_clamp_POT(struct sp_sampler_view *sp_sview,
- struct sp_sampler *sp_samp,
- float s,
- float t,
- float p,
- unsigned level,
- unsigned face_id,
- float rgba[TGSI_QUAD_SIZE])
+static inline void
+img_filter_2d_nearest_clamp_POT(const struct sp_sampler_view *sp_sview,
+ const struct sp_sampler *sp_samp,
+ const struct img_filter_args *args,
+ float *rgba)
{
- unsigned xpot = pot_level_size(sp_sview->xpot, level);
- unsigned ypot = pot_level_size(sp_sview->ypot, level);
+ const unsigned xpot = pot_level_size(sp_sview->xpot, args->level);
+ const unsigned ypot = pot_level_size(sp_sview->ypot, args->level);
union tex_tile_address addr;
int c;
- float u = s * xpot;
- float v = t * ypot;
+ const float u = args->s * xpot + args->offset[0];
+ const float v = args->t * ypot + args->offset[1];
int x0, y0;
const float *out;
addr.value = 0;
- addr.bits.level = level;
+ addr.bits.level = args->level;
+ addr.bits.z = sp_sview->base.u.tex.first_layer;
x0 = util_ifloor(u);
if (x0 < 0)
y0 = ypot - 1;
out = get_texel_2d_no_border(sp_sview, addr, x0, y0);
- for (c = 0; c < TGSI_QUAD_SIZE; c++)
+ for (c = 0; c < TGSI_NUM_CHANNELS; c++)
rgba[TGSI_NUM_CHANNELS*c] = out[c];
if (DEBUG_TEX) {
static void
-img_filter_1d_nearest(struct sp_sampler_view *sp_sview,
- struct sp_sampler *sp_samp,
- float s,
- float t,
- float p,
- unsigned level,
- unsigned face_id,
- float rgba[TGSI_QUAD_SIZE])
+img_filter_1d_nearest(const struct sp_sampler_view *sp_sview,
+ const struct sp_sampler *sp_samp,
+ const struct img_filter_args *args,
+ float *rgba)
{
const struct pipe_resource *texture = sp_sview->base.texture;
- int width;
+ const int width = u_minify(texture->width0, args->level);
int x;
union tex_tile_address addr;
const float *out;
int c;
- width = u_minify(texture->width0, level);
-
assert(width > 0);
addr.value = 0;
- addr.bits.level = level;
+ addr.bits.level = args->level;
- sp_samp->nearest_texcoord_s(s, width, &x);
+ sp_samp->nearest_texcoord_s(args->s, width, args->offset[0], &x);
- out = get_texel_2d(sp_sview, sp_samp, addr, x, 0);
- for (c = 0; c < TGSI_QUAD_SIZE; c++)
+ out = get_texel_1d_array(sp_sview, sp_samp, addr, x,
+ sp_sview->base.u.tex.first_layer);
+ for (c = 0; c < TGSI_NUM_CHANNELS; c++)
rgba[TGSI_NUM_CHANNELS*c] = out[c];
if (DEBUG_TEX) {
static void
-img_filter_1d_array_nearest(struct sp_sampler_view *sp_sview,
- struct sp_sampler *sp_samp,
- float s,
- float t,
- float p,
- unsigned level,
- unsigned face_id,
+img_filter_1d_array_nearest(const struct sp_sampler_view *sp_sview,
+ const struct sp_sampler *sp_samp,
+ const struct img_filter_args *args,
float *rgba)
{
const struct pipe_resource *texture = sp_sview->base.texture;
- int width;
- int x, layer;
+ const int width = u_minify(texture->width0, args->level);
+ const int layer = coord_to_layer(args->t, sp_sview->base.u.tex.first_layer,
+ sp_sview->base.u.tex.last_layer);
+ int x;
union tex_tile_address addr;
const float *out;
int c;
- width = u_minify(texture->width0, level);
-
assert(width > 0);
addr.value = 0;
- addr.bits.level = level;
+ addr.bits.level = args->level;
- sp_samp->nearest_texcoord_s(s, width, &x);
- wrap_array_layer(t, texture->array_size, &layer);
+ sp_samp->nearest_texcoord_s(args->s, width, args->offset[0], &x);
out = get_texel_1d_array(sp_sview, sp_samp, addr, x, layer);
- for (c = 0; c < TGSI_QUAD_SIZE; c++)
+ for (c = 0; c < TGSI_NUM_CHANNELS; c++)
rgba[TGSI_NUM_CHANNELS*c] = out[c];
if (DEBUG_TEX) {
static void
-img_filter_2d_nearest(struct sp_sampler_view *sp_sview,
- struct sp_sampler *sp_samp,
- float s,
- float t,
- float p,
- unsigned level,
- unsigned face_id,
+img_filter_2d_nearest(const struct sp_sampler_view *sp_sview,
+ const struct sp_sampler *sp_samp,
+ const struct img_filter_args *args,
float *rgba)
{
const struct pipe_resource *texture = sp_sview->base.texture;
- int width, height;
+ const int width = u_minify(texture->width0, args->level);
+ const int height = u_minify(texture->height0, args->level);
int x, y;
union tex_tile_address addr;
const float *out;
int c;
- width = u_minify(texture->width0, level);
- height = u_minify(texture->height0, level);
-
assert(width > 0);
assert(height > 0);
addr.value = 0;
- addr.bits.level = level;
+ addr.bits.level = args->level;
+ addr.bits.z = sp_sview->base.u.tex.first_layer;
- sp_samp->nearest_texcoord_s(s, width, &x);
- sp_samp->nearest_texcoord_t(t, height, &y);
+ sp_samp->nearest_texcoord_s(args->s, width, args->offset[0], &x);
+ sp_samp->nearest_texcoord_t(args->t, height, args->offset[1], &y);
out = get_texel_2d(sp_sview, sp_samp, addr, x, y);
- for (c = 0; c < TGSI_QUAD_SIZE; c++)
+ for (c = 0; c < TGSI_NUM_CHANNELS; c++)
rgba[TGSI_NUM_CHANNELS*c] = out[c];
if (DEBUG_TEX) {
static void
-img_filter_2d_array_nearest(struct sp_sampler_view *sp_sview,
- struct sp_sampler *sp_samp,
- float s,
- float t,
- float p,
- unsigned level,
- unsigned face_id,
+img_filter_2d_array_nearest(const struct sp_sampler_view *sp_sview,
+ const struct sp_sampler *sp_samp,
+ const struct img_filter_args *args,
float *rgba)
{
const struct pipe_resource *texture = sp_sview->base.texture;
- int width, height;
- int x, y, layer;
+ const int width = u_minify(texture->width0, args->level);
+ const int height = u_minify(texture->height0, args->level);
+ const int layer = coord_to_layer(args->p, sp_sview->base.u.tex.first_layer,
+ sp_sview->base.u.tex.last_layer);
+ int x, y;
union tex_tile_address addr;
const float *out;
int c;
- width = u_minify(texture->width0, level);
- height = u_minify(texture->height0, level);
-
assert(width > 0);
assert(height > 0);
addr.value = 0;
- addr.bits.level = level;
+ addr.bits.level = args->level;
- sp_samp->nearest_texcoord_s(s, width, &x);
- sp_samp->nearest_texcoord_t(t, height, &y);
- wrap_array_layer(p, texture->array_size, &layer);
+ sp_samp->nearest_texcoord_s(args->s, width, args->offset[0], &x);
+ sp_samp->nearest_texcoord_t(args->t, height, args->offset[1], &y);
out = get_texel_2d_array(sp_sview, sp_samp, addr, x, y, layer);
- for (c = 0; c < TGSI_QUAD_SIZE; c++)
+ for (c = 0; c < TGSI_NUM_CHANNELS; c++)
rgba[TGSI_NUM_CHANNELS*c] = out[c];
if (DEBUG_TEX) {
}
-static INLINE union tex_tile_address
-face(union tex_tile_address addr, unsigned face )
-{
- addr.bits.face = face;
- return addr;
-}
-
-
static void
-img_filter_cube_nearest(struct sp_sampler_view *sp_sview,
- struct sp_sampler *sp_samp,
- float s,
- float t,
- float p,
- unsigned level,
- unsigned face_id,
+img_filter_cube_nearest(const struct sp_sampler_view *sp_sview,
+ const struct sp_sampler *sp_samp,
+ const struct img_filter_args *args,
float *rgba)
{
const struct pipe_resource *texture = sp_sview->base.texture;
- int width, height;
+ const int width = u_minify(texture->width0, args->level);
+ const int height = u_minify(texture->height0, args->level);
+ const int layerface = args->face_id + sp_sview->base.u.tex.first_layer;
int x, y;
union tex_tile_address addr;
const float *out;
int c;
- width = u_minify(texture->width0, level);
- height = u_minify(texture->height0, level);
-
assert(width > 0);
assert(height > 0);
addr.value = 0;
- addr.bits.level = level;
+ addr.bits.level = args->level;
/*
* If NEAREST filtering is done within a miplevel, always apply wrap
* mode CLAMP_TO_EDGE.
*/
if (sp_samp->base.seamless_cube_map) {
- wrap_nearest_clamp_to_edge(s, width, &x);
- wrap_nearest_clamp_to_edge(t, height, &y);
+ wrap_nearest_clamp_to_edge(args->s, width, args->offset[0], &x);
+ wrap_nearest_clamp_to_edge(args->t, height, args->offset[1], &y);
} else {
/* Would probably make sense to ignore mode and just do edge clamp */
- sp_samp->nearest_texcoord_s(s, width, &x);
- sp_samp->nearest_texcoord_t(t, height, &y);
+ sp_samp->nearest_texcoord_s(args->s, width, args->offset[0], &x);
+ sp_samp->nearest_texcoord_t(args->t, height, args->offset[1], &y);
}
- out = get_texel_2d(sp_sview, sp_samp, face(addr, face_id), x, y);
- for (c = 0; c < TGSI_QUAD_SIZE; c++)
+ out = get_texel_cube_array(sp_sview, sp_samp, addr, x, y, layerface);
+ for (c = 0; c < TGSI_NUM_CHANNELS; c++)
rgba[TGSI_NUM_CHANNELS*c] = out[c];
if (DEBUG_TEX) {
}
static void
-img_filter_cube_array_nearest(struct sp_sampler_view *sp_sview,
- struct sp_sampler *sp_samp,
- float s,
- float t,
- float p,
- unsigned level,
- unsigned face_id,
+img_filter_cube_array_nearest(const struct sp_sampler_view *sp_sview,
+ const struct sp_sampler *sp_samp,
+ const struct img_filter_args *args,
float *rgba)
{
const struct pipe_resource *texture = sp_sview->base.texture;
- int width, height;
- int x, y, layer;
+ const int width = u_minify(texture->width0, args->level);
+ const int height = u_minify(texture->height0, args->level);
+ const int layerface =
+ coord_to_layer(6 * args->p + sp_sview->base.u.tex.first_layer,
+ sp_sview->base.u.tex.first_layer,
+ sp_sview->base.u.tex.last_layer - 5) + args->face_id;
+ int x, y;
union tex_tile_address addr;
const float *out;
int c;
- width = u_minify(texture->width0, level);
- height = u_minify(texture->height0, level);
-
assert(width > 0);
assert(height > 0);
addr.value = 0;
- addr.bits.level = level;
+ addr.bits.level = args->level;
- sp_samp->nearest_texcoord_s(s, width, &x);
- sp_samp->nearest_texcoord_t(t, height, &y);
- wrap_array_layer(p, texture->array_size, &layer);
+ sp_samp->nearest_texcoord_s(args->s, width, args->offset[0], &x);
+ sp_samp->nearest_texcoord_t(args->t, height, args->offset[1], &y);
- out = get_texel_cube_array(sp_sview, sp_samp, addr, x, y, layer * 6 + face_id);
- for (c = 0; c < TGSI_QUAD_SIZE; c++)
+ out = get_texel_cube_array(sp_sview, sp_samp, addr, x, y, layerface);
+ for (c = 0; c < TGSI_NUM_CHANNELS; c++)
rgba[TGSI_NUM_CHANNELS*c] = out[c];
if (DEBUG_TEX) {
}
static void
-img_filter_3d_nearest(struct sp_sampler_view *sp_sview,
- struct sp_sampler *sp_samp,
- float s,
- float t,
- float p,
- unsigned level,
- unsigned face_id,
+img_filter_3d_nearest(const struct sp_sampler_view *sp_sview,
+ const struct sp_sampler *sp_samp,
+ const struct img_filter_args *args,
float *rgba)
{
const struct pipe_resource *texture = sp_sview->base.texture;
- int width, height, depth;
+ const int width = u_minify(texture->width0, args->level);
+ const int height = u_minify(texture->height0, args->level);
+ const int depth = u_minify(texture->depth0, args->level);
int x, y, z;
union tex_tile_address addr;
const float *out;
int c;
- width = u_minify(texture->width0, level);
- height = u_minify(texture->height0, level);
- depth = u_minify(texture->depth0, level);
-
assert(width > 0);
assert(height > 0);
assert(depth > 0);
- sp_samp->nearest_texcoord_s(s, width, &x);
- sp_samp->nearest_texcoord_t(t, height, &y);
- sp_samp->nearest_texcoord_p(p, depth, &z);
+ sp_samp->nearest_texcoord_s(args->s, width, args->offset[0], &x);
+ sp_samp->nearest_texcoord_t(args->t, height, args->offset[1], &y);
+ sp_samp->nearest_texcoord_p(args->p, depth, args->offset[2], &z);
addr.value = 0;
- addr.bits.level = level;
+ addr.bits.level = args->level;
out = get_texel_3d(sp_sview, sp_samp, addr, x, y, z);
- for (c = 0; c < TGSI_QUAD_SIZE; c++)
+ for (c = 0; c < TGSI_NUM_CHANNELS; c++)
rgba[TGSI_NUM_CHANNELS*c] = out[c];
}
static void
-img_filter_1d_linear(struct sp_sampler_view *sp_sview,
- struct sp_sampler *sp_samp,
- float s,
- float t,
- float p,
- unsigned level,
- unsigned face_id,
+img_filter_1d_linear(const struct sp_sampler_view *sp_sview,
+ const struct sp_sampler *sp_samp,
+ const struct img_filter_args *args,
float *rgba)
{
const struct pipe_resource *texture = sp_sview->base.texture;
- int width;
+ const int width = u_minify(texture->width0, args->level);
int x0, x1;
float xw; /* weights */
union tex_tile_address addr;
const float *tx0, *tx1;
int c;
- width = u_minify(texture->width0, level);
-
assert(width > 0);
addr.value = 0;
- addr.bits.level = level;
+ addr.bits.level = args->level;
- sp_samp->linear_texcoord_s(s, width, &x0, &x1, &xw);
+ sp_samp->linear_texcoord_s(args->s, width, args->offset[0], &x0, &x1, &xw);
- tx0 = get_texel_2d(sp_sview, sp_samp, addr, x0, 0);
- tx1 = get_texel_2d(sp_sview, sp_samp, addr, x1, 0);
+ tx0 = get_texel_1d_array(sp_sview, sp_samp, addr, x0,
+ sp_sview->base.u.tex.first_layer);
+ tx1 = get_texel_1d_array(sp_sview, sp_samp, addr, x1,
+ sp_sview->base.u.tex.first_layer);
/* interpolate R, G, B, A */
- for (c = 0; c < TGSI_QUAD_SIZE; c++)
+ for (c = 0; c < TGSI_NUM_CHANNELS; c++)
rgba[TGSI_NUM_CHANNELS*c] = lerp(xw, tx0[c], tx1[c]);
}
static void
-img_filter_1d_array_linear(struct sp_sampler_view *sp_sview,
- struct sp_sampler *sp_samp,
- float s,
- float t,
- float p,
- unsigned level,
- unsigned face_id,
+img_filter_1d_array_linear(const struct sp_sampler_view *sp_sview,
+ const struct sp_sampler *sp_samp,
+ const struct img_filter_args *args,
float *rgba)
{
const struct pipe_resource *texture = sp_sview->base.texture;
- int width;
- int x0, x1, layer;
+ const int width = u_minify(texture->width0, args->level);
+ const int layer = coord_to_layer(args->t, sp_sview->base.u.tex.first_layer,
+ sp_sview->base.u.tex.last_layer);
+ int x0, x1;
float xw; /* weights */
union tex_tile_address addr;
const float *tx0, *tx1;
int c;
- width = u_minify(texture->width0, level);
-
assert(width > 0);
addr.value = 0;
- addr.bits.level = level;
+ addr.bits.level = args->level;
- sp_samp->linear_texcoord_s(s, width, &x0, &x1, &xw);
- wrap_array_layer(t, texture->array_size, &layer);
+ sp_samp->linear_texcoord_s(args->s, width, args->offset[0], &x0, &x1, &xw);
tx0 = get_texel_1d_array(sp_sview, sp_samp, addr, x0, layer);
tx1 = get_texel_1d_array(sp_sview, sp_samp, addr, x1, layer);
/* interpolate R, G, B, A */
- for (c = 0; c < TGSI_QUAD_SIZE; c++)
+ for (c = 0; c < TGSI_NUM_CHANNELS; c++)
rgba[TGSI_NUM_CHANNELS*c] = lerp(xw, tx0[c], tx1[c]);
}
+/*
+ * Retrieve the gathered value, need to convert to the
+ * TGSI expected interface, and take component select
+ * and swizzling into account.
+ */
+static float
+get_gather_value(const struct sp_sampler_view *sp_sview,
+ int chan_in, int comp_sel,
+ const float *tx[4])
+{
+ int chan;
+ unsigned swizzle;
+
+ /*
+ * softpipe samples in a different order
+ * to TGSI expects, so we need to swizzle,
+ * the samples into the correct slots.
+ */
+ switch (chan_in) {
+ case 0:
+ chan = 2;
+ break;
+ case 1:
+ chan = 3;
+ break;
+ case 2:
+ chan = 1;
+ break;
+ case 3:
+ chan = 0;
+ break;
+ default:
+ assert(0);
+ return 0.0;
+ }
+
+ /* pick which component to use for the swizzle */
+ switch (comp_sel) {
+ case 0:
+ swizzle = sp_sview->base.swizzle_r;
+ break;
+ case 1:
+ swizzle = sp_sview->base.swizzle_g;
+ break;
+ case 2:
+ swizzle = sp_sview->base.swizzle_b;
+ break;
+ case 3:
+ swizzle = sp_sview->base.swizzle_a;
+ break;
+ default:
+ assert(0);
+ return 0.0;
+ }
+
+ /* get correct result using the channel and swizzle */
+ switch (swizzle) {
+ case PIPE_SWIZZLE_0:
+ return 0.0;
+ case PIPE_SWIZZLE_1:
+ return 1.0;
+ default:
+ return tx[chan][swizzle];
+ }
+}
+
static void
-img_filter_2d_linear(struct sp_sampler_view *sp_sview,
- struct sp_sampler *sp_samp,
- float s,
- float t,
- float p,
- unsigned level,
- unsigned face_id,
+img_filter_2d_linear(const struct sp_sampler_view *sp_sview,
+ const struct sp_sampler *sp_samp,
+ const struct img_filter_args *args,
float *rgba)
{
const struct pipe_resource *texture = sp_sview->base.texture;
- int width, height;
+ const int width = u_minify(texture->width0, args->level);
+ const int height = u_minify(texture->height0, args->level);
int x0, y0, x1, y1;
float xw, yw; /* weights */
union tex_tile_address addr;
- const float *tx0, *tx1, *tx2, *tx3;
+ const float *tx[4];
int c;
- width = u_minify(texture->width0, level);
- height = u_minify(texture->height0, level);
-
assert(width > 0);
assert(height > 0);
addr.value = 0;
- addr.bits.level = level;
-
- sp_samp->linear_texcoord_s(s, width, &x0, &x1, &xw);
- sp_samp->linear_texcoord_t(t, height, &y0, &y1, &yw);
-
- tx0 = get_texel_2d(sp_sview, sp_samp, addr, x0, y0);
- tx1 = get_texel_2d(sp_sview, sp_samp, addr, x1, y0);
- tx2 = get_texel_2d(sp_sview, sp_samp, addr, x0, y1);
- tx3 = get_texel_2d(sp_sview, sp_samp, addr, x1, y1);
-
- /* interpolate R, G, B, A */
- for (c = 0; c < TGSI_QUAD_SIZE; c++)
- rgba[TGSI_NUM_CHANNELS*c] = lerp_2d(xw, yw,
- tx0[c], tx1[c],
- tx2[c], tx3[c]);
+ addr.bits.level = args->level;
+ addr.bits.z = sp_sview->base.u.tex.first_layer;
+
+ sp_samp->linear_texcoord_s(args->s, width, args->offset[0], &x0, &x1, &xw);
+ sp_samp->linear_texcoord_t(args->t, height, args->offset[1], &y0, &y1, &yw);
+
+ tx[0] = get_texel_2d(sp_sview, sp_samp, addr, x0, y0);
+ tx[1] = get_texel_2d(sp_sview, sp_samp, addr, x1, y0);
+ tx[2] = get_texel_2d(sp_sview, sp_samp, addr, x0, y1);
+ tx[3] = get_texel_2d(sp_sview, sp_samp, addr, x1, y1);
+
+ if (args->gather_only) {
+ for (c = 0; c < TGSI_NUM_CHANNELS; c++)
+ rgba[TGSI_NUM_CHANNELS*c] = get_gather_value(sp_sview, c,
+ args->gather_comp,
+ tx);
+ } else {
+ /* interpolate R, G, B, A */
+ for (c = 0; c < TGSI_NUM_CHANNELS; c++)
+ rgba[TGSI_NUM_CHANNELS*c] = lerp_2d(xw, yw,
+ tx[0][c], tx[1][c],
+ tx[2][c], tx[3][c]);
+ }
}
static void
-img_filter_2d_array_linear(struct sp_sampler_view *sp_sview,
- struct sp_sampler *sp_samp,
- float s,
- float t,
- float p,
- unsigned level,
- unsigned face_id,
+img_filter_2d_array_linear(const struct sp_sampler_view *sp_sview,
+ const struct sp_sampler *sp_samp,
+ const struct img_filter_args *args,
float *rgba)
{
const struct pipe_resource *texture = sp_sview->base.texture;
- int width, height;
- int x0, y0, x1, y1, layer;
+ const int width = u_minify(texture->width0, args->level);
+ const int height = u_minify(texture->height0, args->level);
+ const int layer = coord_to_layer(args->p, sp_sview->base.u.tex.first_layer,
+ sp_sview->base.u.tex.last_layer);
+ int x0, y0, x1, y1;
float xw, yw; /* weights */
union tex_tile_address addr;
- const float *tx0, *tx1, *tx2, *tx3;
+ const float *tx[4];
int c;
- width = u_minify(texture->width0, level);
- height = u_minify(texture->height0, level);
-
assert(width > 0);
assert(height > 0);
addr.value = 0;
- addr.bits.level = level;
+ addr.bits.level = args->level;
- sp_samp->linear_texcoord_s(s, width, &x0, &x1, &xw);
- sp_samp->linear_texcoord_t(t, height, &y0, &y1, &yw);
- wrap_array_layer(p, texture->array_size, &layer);
+ sp_samp->linear_texcoord_s(args->s, width, args->offset[0], &x0, &x1, &xw);
+ sp_samp->linear_texcoord_t(args->t, height, args->offset[1], &y0, &y1, &yw);
- tx0 = get_texel_2d_array(sp_sview, sp_samp, addr, x0, y0, layer);
- tx1 = get_texel_2d_array(sp_sview, sp_samp, addr, x1, y0, layer);
- tx2 = get_texel_2d_array(sp_sview, sp_samp, addr, x0, y1, layer);
- tx3 = get_texel_2d_array(sp_sview, sp_samp, addr, x1, y1, layer);
+ tx[0] = get_texel_2d_array(sp_sview, sp_samp, addr, x0, y0, layer);
+ tx[1] = get_texel_2d_array(sp_sview, sp_samp, addr, x1, y0, layer);
+ tx[2] = get_texel_2d_array(sp_sview, sp_samp, addr, x0, y1, layer);
+ tx[3] = get_texel_2d_array(sp_sview, sp_samp, addr, x1, y1, layer);
- /* interpolate R, G, B, A */
- for (c = 0; c < TGSI_QUAD_SIZE; c++)
- rgba[TGSI_NUM_CHANNELS*c] = lerp_2d(xw, yw,
- tx0[c], tx1[c],
- tx2[c], tx3[c]);
+ if (args->gather_only) {
+ for (c = 0; c < TGSI_NUM_CHANNELS; c++)
+ rgba[TGSI_NUM_CHANNELS*c] = get_gather_value(sp_sview, c,
+ args->gather_comp,
+ tx);
+ } else {
+ /* interpolate R, G, B, A */
+ for (c = 0; c < TGSI_NUM_CHANNELS; c++)
+ rgba[TGSI_NUM_CHANNELS*c] = lerp_2d(xw, yw,
+ tx[0][c], tx[1][c],
+ tx[2][c], tx[3][c]);
+ }
}
static void
-img_filter_cube_linear(struct sp_sampler_view *sp_sview,
- struct sp_sampler *sp_samp,
- float s,
- float t,
- float p,
- unsigned level,
- unsigned face_id,
+img_filter_cube_linear(const struct sp_sampler_view *sp_sview,
+ const struct sp_sampler *sp_samp,
+ const struct img_filter_args *args,
float *rgba)
{
const struct pipe_resource *texture = sp_sview->base.texture;
- int width, height;
+ const int width = u_minify(texture->width0, args->level);
+ const int height = u_minify(texture->height0, args->level);
+ const int layer = sp_sview->base.u.tex.first_layer;
int x0, y0, x1, y1;
float xw, yw; /* weights */
- union tex_tile_address addr, addrj;
- const float *tx0, *tx1, *tx2, *tx3;
+ union tex_tile_address addr;
+ const float *tx[4];
float corner0[TGSI_QUAD_SIZE], corner1[TGSI_QUAD_SIZE],
corner2[TGSI_QUAD_SIZE], corner3[TGSI_QUAD_SIZE];
int c;
- width = u_minify(texture->width0, level);
- height = u_minify(texture->height0, level);
-
assert(width > 0);
assert(height > 0);
addr.value = 0;
- addr.bits.level = level;
+ addr.bits.level = args->level;
/*
* For seamless if LINEAR filtering is done within a miplevel,
*/
if (sp_samp->base.seamless_cube_map) {
/* Note this is a bit overkill, actual clamping is not required */
- wrap_linear_clamp_to_border(s, width, &x0, &x1, &xw);
- wrap_linear_clamp_to_border(t, height, &y0, &y1, &yw);
+ wrap_linear_clamp_to_border(args->s, width, args->offset[0], &x0, &x1, &xw);
+ wrap_linear_clamp_to_border(args->t, height, args->offset[1], &y0, &y1, &yw);
} else {
/* Would probably make sense to ignore mode and just do edge clamp */
- sp_samp->linear_texcoord_s(s, width, &x0, &x1, &xw);
- sp_samp->linear_texcoord_t(t, height, &y0, &y1, &yw);
+ sp_samp->linear_texcoord_s(args->s, width, args->offset[0], &x0, &x1, &xw);
+ sp_samp->linear_texcoord_t(args->t, height, args->offset[1], &y0, &y1, &yw);
}
- addrj = face(addr, face_id);
-
if (sp_samp->base.seamless_cube_map) {
- tx0 = get_texel_cube_seamless(sp_sview, addrj, x0, y0, corner0);
- tx1 = get_texel_cube_seamless(sp_sview, addrj, x1, y0, corner1);
- tx2 = get_texel_cube_seamless(sp_sview, addrj, x0, y1, corner2);
- tx3 = get_texel_cube_seamless(sp_sview, addrj, x1, y1, corner3);
+ tx[0] = get_texel_cube_seamless(sp_sview, addr, x0, y0, corner0, layer, args->face_id);
+ tx[1] = get_texel_cube_seamless(sp_sview, addr, x1, y0, corner1, layer, args->face_id);
+ tx[2] = get_texel_cube_seamless(sp_sview, addr, x0, y1, corner2, layer, args->face_id);
+ tx[3] = get_texel_cube_seamless(sp_sview, addr, x1, y1, corner3, layer, args->face_id);
} else {
- tx0 = get_texel_2d(sp_sview, sp_samp, addrj, x0, y0);
- tx1 = get_texel_2d(sp_sview, sp_samp, addrj, x1, y0);
- tx2 = get_texel_2d(sp_sview, sp_samp, addrj, x0, y1);
- tx3 = get_texel_2d(sp_sview, sp_samp, addrj, x1, y1);
+ tx[0] = get_texel_cube_array(sp_sview, sp_samp, addr, x0, y0, layer + args->face_id);
+ tx[1] = get_texel_cube_array(sp_sview, sp_samp, addr, x1, y0, layer + args->face_id);
+ tx[2] = get_texel_cube_array(sp_sview, sp_samp, addr, x0, y1, layer + args->face_id);
+ tx[3] = get_texel_cube_array(sp_sview, sp_samp, addr, x1, y1, layer + args->face_id);
+ }
+
+ if (args->gather_only) {
+ for (c = 0; c < TGSI_NUM_CHANNELS; c++)
+ rgba[TGSI_NUM_CHANNELS*c] = get_gather_value(sp_sview, c,
+ args->gather_comp,
+ tx);
+ } else {
+ /* interpolate R, G, B, A */
+ for (c = 0; c < TGSI_NUM_CHANNELS; c++)
+ rgba[TGSI_NUM_CHANNELS*c] = lerp_2d(xw, yw,
+ tx[0][c], tx[1][c],
+ tx[2][c], tx[3][c]);
}
- /* interpolate R, G, B, A */
- for (c = 0; c < TGSI_QUAD_SIZE; c++)
- rgba[TGSI_NUM_CHANNELS*c] = lerp_2d(xw, yw,
- tx0[c], tx1[c],
- tx2[c], tx3[c]);
}
static void
-img_filter_cube_array_linear(struct sp_sampler_view *sp_sview,
- struct sp_sampler *sp_samp,
- float s,
- float t,
- float p,
- unsigned level,
- unsigned face_id,
+img_filter_cube_array_linear(const struct sp_sampler_view *sp_sview,
+ const struct sp_sampler *sp_samp,
+ const struct img_filter_args *args,
float *rgba)
{
const struct pipe_resource *texture = sp_sview->base.texture;
- int width, height;
- int x0, y0, x1, y1, layer;
+ const int width = u_minify(texture->width0, args->level);
+ const int height = u_minify(texture->height0, args->level);
+ const int layer =
+ coord_to_layer(6 * args->p + sp_sview->base.u.tex.first_layer,
+ sp_sview->base.u.tex.first_layer,
+ sp_sview->base.u.tex.last_layer - 5);
+ int x0, y0, x1, y1;
float xw, yw; /* weights */
union tex_tile_address addr;
- const float *tx0, *tx1, *tx2, *tx3;
+ const float *tx[4];
+ float corner0[TGSI_QUAD_SIZE], corner1[TGSI_QUAD_SIZE],
+ corner2[TGSI_QUAD_SIZE], corner3[TGSI_QUAD_SIZE];
int c;
- width = u_minify(texture->width0, level);
- height = u_minify(texture->height0, level);
-
assert(width > 0);
assert(height > 0);
addr.value = 0;
- addr.bits.level = level;
+ addr.bits.level = args->level;
- sp_samp->linear_texcoord_s(s, width, &x0, &x1, &xw);
- sp_samp->linear_texcoord_t(t, height, &y0, &y1, &yw);
- wrap_array_layer(p, texture->array_size, &layer);
+ /*
+ * For seamless if LINEAR filtering is done within a miplevel,
+ * always apply wrap mode CLAMP_TO_BORDER.
+ */
+ if (sp_samp->base.seamless_cube_map) {
+ /* Note this is a bit overkill, actual clamping is not required */
+ wrap_linear_clamp_to_border(args->s, width, args->offset[0], &x0, &x1, &xw);
+ wrap_linear_clamp_to_border(args->t, height, args->offset[1], &y0, &y1, &yw);
+ } else {
+ /* Would probably make sense to ignore mode and just do edge clamp */
+ sp_samp->linear_texcoord_s(args->s, width, args->offset[0], &x0, &x1, &xw);
+ sp_samp->linear_texcoord_t(args->t, height, args->offset[1], &y0, &y1, &yw);
+ }
- tx0 = get_texel_cube_array(sp_sview, sp_samp, addr, x0, y0, layer * 6 + face_id);
- tx1 = get_texel_cube_array(sp_sview, sp_samp, addr, x1, y0, layer * 6 + face_id);
- tx2 = get_texel_cube_array(sp_sview, sp_samp, addr, x0, y1, layer * 6 + face_id);
- tx3 = get_texel_cube_array(sp_sview, sp_samp, addr, x1, y1, layer * 6 + face_id);
+ if (sp_samp->base.seamless_cube_map) {
+ tx[0] = get_texel_cube_seamless(sp_sview, addr, x0, y0, corner0, layer, args->face_id);
+ tx[1] = get_texel_cube_seamless(sp_sview, addr, x1, y0, corner1, layer, args->face_id);
+ tx[2] = get_texel_cube_seamless(sp_sview, addr, x0, y1, corner2, layer, args->face_id);
+ tx[3] = get_texel_cube_seamless(sp_sview, addr, x1, y1, corner3, layer, args->face_id);
+ } else {
+ tx[0] = get_texel_cube_array(sp_sview, sp_samp, addr, x0, y0, layer + args->face_id);
+ tx[1] = get_texel_cube_array(sp_sview, sp_samp, addr, x1, y0, layer + args->face_id);
+ tx[2] = get_texel_cube_array(sp_sview, sp_samp, addr, x0, y1, layer + args->face_id);
+ tx[3] = get_texel_cube_array(sp_sview, sp_samp, addr, x1, y1, layer + args->face_id);
+ }
- /* interpolate R, G, B, A */
- for (c = 0; c < TGSI_QUAD_SIZE; c++)
- rgba[TGSI_NUM_CHANNELS*c] = lerp_2d(xw, yw,
- tx0[c], tx1[c],
- tx2[c], tx3[c]);
+ if (args->gather_only) {
+ for (c = 0; c < TGSI_NUM_CHANNELS; c++)
+ rgba[TGSI_NUM_CHANNELS*c] = get_gather_value(sp_sview, c,
+ args->gather_comp,
+ tx);
+ } else {
+ /* interpolate R, G, B, A */
+ for (c = 0; c < TGSI_NUM_CHANNELS; c++)
+ rgba[TGSI_NUM_CHANNELS*c] = lerp_2d(xw, yw,
+ tx[0][c], tx[1][c],
+ tx[2][c], tx[3][c]);
+ }
}
static void
-img_filter_3d_linear(struct sp_sampler_view *sp_sview,
- struct sp_sampler *sp_samp,
- float s,
- float t,
- float p,
- unsigned level,
- unsigned face_id,
+img_filter_3d_linear(const struct sp_sampler_view *sp_sview,
+ const struct sp_sampler *sp_samp,
+ const struct img_filter_args *args,
float *rgba)
{
const struct pipe_resource *texture = sp_sview->base.texture;
- int width, height, depth;
+ const int width = u_minify(texture->width0, args->level);
+ const int height = u_minify(texture->height0, args->level);
+ const int depth = u_minify(texture->depth0, args->level);
int x0, x1, y0, y1, z0, z1;
float xw, yw, zw; /* interpolation weights */
union tex_tile_address addr;
const float *tx00, *tx01, *tx02, *tx03, *tx10, *tx11, *tx12, *tx13;
int c;
- width = u_minify(texture->width0, level);
- height = u_minify(texture->height0, level);
- depth = u_minify(texture->depth0, level);
-
addr.value = 0;
- addr.bits.level = level;
+ addr.bits.level = args->level;
assert(width > 0);
assert(height > 0);
assert(depth > 0);
- sp_samp->linear_texcoord_s(s, width, &x0, &x1, &xw);
- sp_samp->linear_texcoord_t(t, height, &y0, &y1, &yw);
- sp_samp->linear_texcoord_p(p, depth, &z0, &z1, &zw);
-
+ sp_samp->linear_texcoord_s(args->s, width, args->offset[0], &x0, &x1, &xw);
+ sp_samp->linear_texcoord_t(args->t, height, args->offset[1], &y0, &y1, &yw);
+ sp_samp->linear_texcoord_p(args->p, depth, args->offset[2], &z0, &z1, &zw);
tx00 = get_texel_3d(sp_sview, sp_samp, addr, x0, y0, z0);
tx01 = get_texel_3d(sp_sview, sp_samp, addr, x1, y0, z0);
tx13 = get_texel_3d(sp_sview, sp_samp, addr, x1, y1, z1);
/* interpolate R, G, B, A */
- for (c = 0; c < TGSI_QUAD_SIZE; c++)
+ for (c = 0; c < TGSI_NUM_CHANNELS; c++)
rgba[TGSI_NUM_CHANNELS*c] = lerp_3d(xw, yw, zw,
tx00[c], tx01[c],
tx02[c], tx03[c],
* \param lod_in per-fragment lod_bias or explicit_lod.
* \param lod returns the per-fragment lod.
*/
-static INLINE void
+static inline void
compute_lod(const struct pipe_sampler_state *sampler,
enum tgsi_sampler_control control,
const float biased_lambda,
const float lod_in[TGSI_QUAD_SIZE],
float lod[TGSI_QUAD_SIZE])
{
- float min_lod = sampler->min_lod;
- float max_lod = sampler->max_lod;
+ const float min_lod = sampler->min_lod;
+ const float max_lod = sampler->max_lod;
uint i;
switch (control) {
- case tgsi_sampler_lod_none:
- case tgsi_sampler_lod_zero:
+ case TGSI_SAMPLER_LOD_NONE:
+ case TGSI_SAMPLER_LOD_ZERO:
/* XXX FIXME */
- case tgsi_sampler_derivs_explicit:
+ case TGSI_SAMPLER_DERIVS_EXPLICIT:
lod[0] = lod[1] = lod[2] = lod[3] = CLAMP(biased_lambda, min_lod, max_lod);
break;
- case tgsi_sampler_lod_bias:
+ case TGSI_SAMPLER_LOD_BIAS:
for (i = 0; i < TGSI_QUAD_SIZE; i++) {
lod[i] = biased_lambda + lod_in[i];
lod[i] = CLAMP(lod[i], min_lod, max_lod);
}
break;
- case tgsi_sampler_lod_explicit:
+ case TGSI_SAMPLER_LOD_EXPLICIT:
for (i = 0; i < TGSI_QUAD_SIZE; i++) {
lod[i] = CLAMP(lod_in[i], min_lod, max_lod);
}
}
-/* Calculate level of detail for every fragment.
+/* Calculate level of detail for every fragment. The computed value is not
+ * clamped to lod_min and lod_max.
* \param lod_in per-fragment lod_bias or explicit_lod.
* \param lod results per-fragment lod.
*/
-static INLINE void
-compute_lambda_lod(struct sp_sampler_view *sp_sview,
- struct sp_sampler *sp_samp,
- const float s[TGSI_QUAD_SIZE],
- const float t[TGSI_QUAD_SIZE],
- const float p[TGSI_QUAD_SIZE],
- const float lod_in[TGSI_QUAD_SIZE],
- enum tgsi_sampler_control control,
- float lod[TGSI_QUAD_SIZE])
+static inline void
+compute_lambda_lod_unclamped(const struct sp_sampler_view *sp_sview,
+ const struct sp_sampler *sp_samp,
+ const float s[TGSI_QUAD_SIZE],
+ const float t[TGSI_QUAD_SIZE],
+ const float p[TGSI_QUAD_SIZE],
+ const float lod_in[TGSI_QUAD_SIZE],
+ enum tgsi_sampler_control control,
+ float lod[TGSI_QUAD_SIZE])
{
const struct pipe_sampler_state *sampler = &sp_samp->base;
- float lod_bias = sampler->lod_bias;
- float min_lod = sampler->min_lod;
- float max_lod = sampler->max_lod;
+ const float lod_bias = sampler->lod_bias;
float lambda;
uint i;
switch (control) {
- case tgsi_sampler_lod_none:
+ case TGSI_SAMPLER_LOD_NONE:
/* XXX FIXME */
- case tgsi_sampler_derivs_explicit:
+ case TGSI_SAMPLER_DERIVS_EXPLICIT:
lambda = sp_sview->compute_lambda(sp_sview, s, t, p) + lod_bias;
- lod[0] = lod[1] = lod[2] = lod[3] = CLAMP(lambda, min_lod, max_lod);
+ lod[0] = lod[1] = lod[2] = lod[3] = lambda;
break;
- case tgsi_sampler_lod_bias:
+ case TGSI_SAMPLER_LOD_BIAS:
lambda = sp_sview->compute_lambda(sp_sview, s, t, p) + lod_bias;
for (i = 0; i < TGSI_QUAD_SIZE; i++) {
lod[i] = lambda + lod_in[i];
- lod[i] = CLAMP(lod[i], min_lod, max_lod);
}
break;
- case tgsi_sampler_lod_explicit:
+ case TGSI_SAMPLER_LOD_EXPLICIT:
for (i = 0; i < TGSI_QUAD_SIZE; i++) {
- lod[i] = CLAMP(lod_in[i], min_lod, max_lod);
+ lod[i] = lod_in[i] + lod_bias;
}
break;
- case tgsi_sampler_lod_zero:
- /* this is all static state in the sampler really need clamp here? */
- lod[0] = lod[1] = lod[2] = lod[3] = CLAMP(lod_bias, min_lod, max_lod);
+ case TGSI_SAMPLER_LOD_ZERO:
+ case TGSI_SAMPLER_GATHER:
+ lod[0] = lod[1] = lod[2] = lod[3] = lod_bias;
break;
default:
assert(0);
}
}
+/* Calculate level of detail for every fragment.
+ * \param lod_in per-fragment lod_bias or explicit_lod.
+ * \param lod results per-fragment lod.
+ */
+static inline void
+compute_lambda_lod(const struct sp_sampler_view *sp_sview,
+ const struct sp_sampler *sp_samp,
+ const float s[TGSI_QUAD_SIZE],
+ const float t[TGSI_QUAD_SIZE],
+ const float p[TGSI_QUAD_SIZE],
+ const float lod_in[TGSI_QUAD_SIZE],
+ enum tgsi_sampler_control control,
+ float lod[TGSI_QUAD_SIZE])
+{
+ const struct pipe_sampler_state *sampler = &sp_samp->base;
+ const float min_lod = sampler->min_lod;
+ const float max_lod = sampler->max_lod;
+ int i;
+
+ compute_lambda_lod_unclamped(sp_sview, sp_samp,
+ s, t, p, lod_in, control, lod);
+ for (i = 0; i < TGSI_QUAD_SIZE; i++) {
+ lod[i] = CLAMP(lod[i], min_lod, max_lod);
+ }
+}
+
+static inline unsigned
+get_gather_component(const float lod_in[TGSI_QUAD_SIZE])
+{
+ /* gather component is stored in lod_in slot as unsigned */
+ return (*(unsigned int *)lod_in) & 0x3;
+}
+
+/**
+ * Clamps given lod to both lod limits and mip level limits. Clamping to the
+ * latter limits is done so that lod is relative to the first (base) level.
+ */
+static void
+clamp_lod(const struct sp_sampler_view *sp_sview,
+ const struct sp_sampler *sp_samp,
+ const float lod[TGSI_QUAD_SIZE],
+ float clamped[TGSI_QUAD_SIZE])
+{
+ const float min_lod = sp_samp->base.min_lod;
+ const float max_lod = sp_samp->base.max_lod;
+ const float min_level = sp_sview->base.u.tex.first_level;
+ const float max_level = sp_sview->base.u.tex.last_level;
+ int i;
+
+ for (i = 0; i < TGSI_QUAD_SIZE; i++) {
+ float cl = lod[i];
+
+ cl = CLAMP(cl, min_lod, max_lod);
+ cl = CLAMP(cl, 0, max_level - min_level);
+ clamped[i] = cl;
+ }
+}
+
+/**
+ * Get mip level relative to base level for linear mip filter
+ */
+static void
+mip_rel_level_linear(const struct sp_sampler_view *sp_sview,
+ const struct sp_sampler *sp_samp,
+ const float lod[TGSI_QUAD_SIZE],
+ float level[TGSI_QUAD_SIZE])
+{
+ clamp_lod(sp_sview, sp_samp, lod, level);
+}
static void
-mip_filter_linear(struct sp_sampler_view *sp_sview,
- struct sp_sampler *sp_samp,
+mip_filter_linear(const struct sp_sampler_view *sp_sview,
+ const struct sp_sampler *sp_samp,
img_filter_func min_filter,
img_filter_func mag_filter,
const float s[TGSI_QUAD_SIZE],
const float p[TGSI_QUAD_SIZE],
const float c0[TGSI_QUAD_SIZE],
const float lod_in[TGSI_QUAD_SIZE],
- enum tgsi_sampler_control control,
+ const struct filter_args *filt_args,
float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
{
const struct pipe_sampler_view *psview = &sp_sview->base;
int j;
float lod[TGSI_QUAD_SIZE];
+ struct img_filter_args args;
- compute_lambda_lod(sp_sview, sp_samp, s, t, p, lod_in, control, lod);
+ compute_lambda_lod(sp_sview, sp_samp, s, t, p, lod_in, filt_args->control, lod);
- for (j = 0; j < TGSI_QUAD_SIZE; j++) {
- int level0 = psview->u.tex.first_level + (int)lod[j];
+ args.offset = filt_args->offset;
+ args.gather_only = filt_args->control == TGSI_SAMPLER_GATHER;
+ args.gather_comp = get_gather_component(lod_in);
- if (lod[j] < 0.0)
- mag_filter(sp_sview, sp_samp, s[j], t[j], p[j],
- psview->u.tex.first_level,
- sp_sview->faces[j], &rgba[0][j]);
+ for (j = 0; j < TGSI_QUAD_SIZE; j++) {
+ const int level0 = psview->u.tex.first_level + (int)lod[j];
- else if (level0 >= (int) psview->u.tex.last_level)
- min_filter(sp_sview, sp_samp, s[j], t[j], p[j], psview->u.tex.last_level,
- sp_sview->faces[j], &rgba[0][j]);
+ args.s = s[j];
+ args.t = t[j];
+ args.p = p[j];
+ args.face_id = filt_args->faces[j];
+ if (lod[j] < 0.0) {
+ args.level = psview->u.tex.first_level;
+ mag_filter(sp_sview, sp_samp, &args, &rgba[0][j]);
+ }
+ else if (level0 >= (int) psview->u.tex.last_level) {
+ args.level = psview->u.tex.last_level;
+ min_filter(sp_sview, sp_samp, &args, &rgba[0][j]);
+ }
else {
float levelBlend = frac(lod[j]);
float rgbax[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE];
int c;
- min_filter(sp_sview, sp_samp, s[j], t[j], p[j], level0,
- sp_sview->faces[j], &rgbax[0][0]);
- min_filter(sp_sview, sp_samp, s[j], t[j], p[j], level0+1,
- sp_sview->faces[j], &rgbax[0][1]);
+ args.level = level0;
+ min_filter(sp_sview, sp_samp, &args, &rgbax[0][0]);
+ args.level = level0+1;
+ min_filter(sp_sview, sp_samp, &args, &rgbax[0][1]);
for (c = 0; c < 4; c++) {
rgba[c][j] = lerp(levelBlend, rgbax[c][0], rgbax[c][1]);
}
+/**
+ * Get mip level relative to base level for nearest mip filter
+ */
+static void
+mip_rel_level_nearest(const struct sp_sampler_view *sp_sview,
+ const struct sp_sampler *sp_samp,
+ const float lod[TGSI_QUAD_SIZE],
+ float level[TGSI_QUAD_SIZE])
+{
+ int j;
+
+ clamp_lod(sp_sview, sp_samp, lod, level);
+ for (j = 0; j < TGSI_QUAD_SIZE; j++)
+ /* TODO: It should rather be:
+ * level[j] = ceil(level[j] + 0.5F) - 1.0F;
+ */
+ level[j] = (int)(level[j] + 0.5F);
+}
+
/**
* Compute nearest mipmap level from texcoords.
* Then sample the texture level for four elements of a quad.
* \param c0 the LOD bias factors, or absolute LODs (depending on control)
*/
static void
-mip_filter_nearest(struct sp_sampler_view *sp_sview,
- struct sp_sampler *sp_samp,
+mip_filter_nearest(const struct sp_sampler_view *sp_sview,
+ const struct sp_sampler *sp_samp,
img_filter_func min_filter,
img_filter_func mag_filter,
const float s[TGSI_QUAD_SIZE],
const float p[TGSI_QUAD_SIZE],
const float c0[TGSI_QUAD_SIZE],
const float lod_in[TGSI_QUAD_SIZE],
- enum tgsi_sampler_control control,
+ const struct filter_args *filt_args,
float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
{
const struct pipe_sampler_view *psview = &sp_sview->base;
float lod[TGSI_QUAD_SIZE];
int j;
+ struct img_filter_args args;
- compute_lambda_lod(sp_sview, sp_samp, s, t, p, lod_in, control, lod);
+ args.offset = filt_args->offset;
+ args.gather_only = filt_args->control == TGSI_SAMPLER_GATHER;
+ args.gather_comp = get_gather_component(lod_in);
+
+ compute_lambda_lod(sp_sview, sp_samp, s, t, p, lod_in, filt_args->control, lod);
for (j = 0; j < TGSI_QUAD_SIZE; j++) {
- if (lod[j] < 0.0)
- mag_filter(sp_sview, sp_samp, s[j], t[j], p[j],
- psview->u.tex.first_level,
- sp_sview->faces[j], &rgba[0][j]);
- else {
- int level = psview->u.tex.first_level + (int)(lod[j] + 0.5F);
- level = MIN2(level, (int)psview->u.tex.last_level);
- min_filter(sp_sview, sp_samp, s[j], t[j], p[j],
- level, sp_sview->faces[j], &rgba[0][j]);
+ args.s = s[j];
+ args.t = t[j];
+ args.p = p[j];
+ args.face_id = filt_args->faces[j];
+
+ if (lod[j] < 0.0) {
+ args.level = psview->u.tex.first_level;
+ mag_filter(sp_sview, sp_samp, &args, &rgba[0][j]);
+ } else {
+ const int level = psview->u.tex.first_level + (int)(lod[j] + 0.5F);
+ args.level = MIN2(level, (int)psview->u.tex.last_level);
+ min_filter(sp_sview, sp_samp, &args, &rgba[0][j]);
}
}
}
+/**
+ * Get mip level relative to base level for none mip filter
+ */
+static void
+mip_rel_level_none(const struct sp_sampler_view *sp_sview,
+ const struct sp_sampler *sp_samp,
+ const float lod[TGSI_QUAD_SIZE],
+ float level[TGSI_QUAD_SIZE])
+{
+ int j;
+
+ for (j = 0; j < TGSI_QUAD_SIZE; j++) {
+ level[j] = 0;
+ }
+}
+
static void
-mip_filter_none(struct sp_sampler_view *sp_sview,
- struct sp_sampler *sp_samp,
+mip_filter_none(const struct sp_sampler_view *sp_sview,
+ const struct sp_sampler *sp_samp,
img_filter_func min_filter,
img_filter_func mag_filter,
const float s[TGSI_QUAD_SIZE],
const float p[TGSI_QUAD_SIZE],
const float c0[TGSI_QUAD_SIZE],
const float lod_in[TGSI_QUAD_SIZE],
- enum tgsi_sampler_control control,
+ const struct filter_args *filt_args,
float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
{
float lod[TGSI_QUAD_SIZE];
int j;
+ struct img_filter_args args;
+
+ args.level = sp_sview->base.u.tex.first_level;
+ args.offset = filt_args->offset;
+ args.gather_only = filt_args->control == TGSI_SAMPLER_GATHER;
- compute_lambda_lod(sp_sview, sp_samp, s, t, p, lod_in, control, lod);
+ compute_lambda_lod(sp_sview, sp_samp, s, t, p, lod_in, filt_args->control, lod);
for (j = 0; j < TGSI_QUAD_SIZE; j++) {
- if (lod[j] < 0.0) {
- mag_filter(sp_sview, sp_samp, s[j], t[j], p[j],
- sp_sview->base.u.tex.first_level,
- sp_sview->faces[j], &rgba[0][j]);
+ args.s = s[j];
+ args.t = t[j];
+ args.p = p[j];
+ args.face_id = filt_args->faces[j];
+ if (lod[j] < 0.0) {
+ mag_filter(sp_sview, sp_samp, &args, &rgba[0][j]);
}
else {
- min_filter(sp_sview, sp_samp, s[j], t[j], p[j],
- sp_sview->base.u.tex.first_level,
- sp_sview->faces[j], &rgba[0][j]);
+ min_filter(sp_sview, sp_samp, &args, &rgba[0][j]);
}
}
}
+/**
+ * Get mip level relative to base level for none mip filter
+ */
+static void
+mip_rel_level_none_no_filter_select(const struct sp_sampler_view *sp_sview,
+ const struct sp_sampler *sp_samp,
+ const float lod[TGSI_QUAD_SIZE],
+ float level[TGSI_QUAD_SIZE])
+{
+ mip_rel_level_none(sp_sview, sp_samp, lod, level);
+}
+
static void
-mip_filter_none_no_filter_select(struct sp_sampler_view *sp_sview,
- struct sp_sampler *sp_samp,
+mip_filter_none_no_filter_select(const struct sp_sampler_view *sp_sview,
+ const struct sp_sampler *sp_samp,
img_filter_func min_filter,
img_filter_func mag_filter,
const float s[TGSI_QUAD_SIZE],
const float p[TGSI_QUAD_SIZE],
const float c0[TGSI_QUAD_SIZE],
const float lod_in[TGSI_QUAD_SIZE],
- enum tgsi_sampler_control control,
+ const struct filter_args *filt_args,
float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
{
int j;
-
- for (j = 0; j < TGSI_QUAD_SIZE; j++)
- mag_filter(sp_sview, sp_samp, s[j], t[j], p[j],
- sp_sview->base.u.tex.first_level,
- sp_sview->faces[j], &rgba[0][j]);
+ struct img_filter_args args;
+ args.level = sp_sview->base.u.tex.first_level;
+ args.offset = filt_args->offset;
+ args.gather_only = filt_args->control == TGSI_SAMPLER_GATHER;
+ for (j = 0; j < TGSI_QUAD_SIZE; j++) {
+ args.s = s[j];
+ args.t = t[j];
+ args.p = p[j];
+ args.face_id = filt_args->faces[j];
+ mag_filter(sp_sview, sp_samp, &args, &rgba[0][j]);
+ }
}
/* For anisotropic filtering */
#define WEIGHT_LUT_SIZE 1024
-static float *weightLut = NULL;
+static const float *weightLut = NULL;
/**
* Creates the look-up table used to speed-up EWA sampling
{
unsigned i;
if (!weightLut) {
- weightLut = (float *) MALLOC(WEIGHT_LUT_SIZE * sizeof(float));
+ float *lut = (float *) MALLOC(WEIGHT_LUT_SIZE * sizeof(float));
for (i = 0; i < WEIGHT_LUT_SIZE; ++i) {
- float alpha = 2;
- float r2 = (float) i / (float) (WEIGHT_LUT_SIZE - 1);
- float weight = (float) exp(-alpha * r2);
- weightLut[i] = weight;
+ const float alpha = 2;
+ const float r2 = (float) i / (float) (WEIGHT_LUT_SIZE - 1);
+ const float weight = (float) exp(-alpha * r2);
+ lut[i] = weight;
}
+ weightLut = lut;
}
}
* "Fundamentals of Texture Mapping and Image Warping" (1989)
*/
static void
-img_filter_2d_ewa(struct sp_sampler_view *sp_sview,
- struct sp_sampler *sp_samp,
+img_filter_2d_ewa(const struct sp_sampler_view *sp_sview,
+ const struct sp_sampler *sp_samp,
img_filter_func min_filter,
img_filter_func mag_filter,
const float s[TGSI_QUAD_SIZE],
const float t[TGSI_QUAD_SIZE],
const float p[TGSI_QUAD_SIZE],
+ const uint faces[TGSI_QUAD_SIZE],
+ const int8_t *offset,
unsigned level,
const float dudx, const float dvdx,
const float dudy, const float dvdy,
const struct pipe_resource *texture = sp_sview->base.texture;
// ??? Won't the image filters blow up if level is negative?
- unsigned level0 = level > 0 ? level : 0;
- float scaling = 1.0f / (1 << level0);
- int width = u_minify(texture->width0, level0);
- int height = u_minify(texture->height0, level0);
-
- float ux = dudx * scaling;
- float vx = dvdx * scaling;
- float uy = dudy * scaling;
- float vy = dvdy * scaling;
+ const unsigned level0 = level > 0 ? level : 0;
+ const float scaling = 1.0f / (1 << level0);
+ const int width = u_minify(texture->width0, level0);
+ const int height = u_minify(texture->height0, level0);
+ struct img_filter_args args;
+ const float ux = dudx * scaling;
+ const float vx = dvdx * scaling;
+ const float uy = dudy * scaling;
+ const float vy = dvdy * scaling;
/* compute ellipse coefficients to bound the region:
* A*x*x + B*x*y + C*y*y = F.
float F = A*C-B*B/4.0f;
/* check if it is an ellipse */
- /* ASSERT(F > 0.0); */
+ /* assert(F > 0.0); */
/* Compute the ellipse's (u,v) bounding box in texture space */
- float d = -B*B+4.0f*C*A;
- float box_u = 2.0f / d * sqrtf(d*C*F); /* box_u -> half of bbox with */
- float box_v = 2.0f / d * sqrtf(A*d*F); /* box_v -> half of bbox height */
+ const float d = -B*B+4.0f*C*A;
+ const float box_u = 2.0f / d * sqrtf(d*C*F); /* box_u -> half of bbox with */
+ const float box_v = 2.0f / d * sqrtf(A*d*F); /* box_v -> half of bbox height */
float rgba_temp[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE];
float s_buffer[TGSI_QUAD_SIZE];
float t_buffer[TGSI_QUAD_SIZE];
float weight_buffer[TGSI_QUAD_SIZE];
- unsigned buffer_next;
int j;
- float den; /* = 0.0F; */
- float ddq;
- float U; /* = u0 - tex_u; */
- int v;
+
+ /* For each quad, the du and dx values are the same and so the ellipse is
+ * also the same. Note that texel/image access can only be performed using
+ * a quad, i.e. it is not possible to get the pixel value for a single
+ * tex coord. In order to have a better performance, the access is buffered
+ * using the s_buffer/t_buffer and weight_buffer. Only when the buffer is
+ * full, then the pixel values are read from the image.
+ */
+ const float ddq = 2 * A;
/* Scale ellipse formula to directly index the Filter Lookup Table.
* i.e. scale so that F = WEIGHT_LUT_SIZE-1
*/
- double formScale = (double) (WEIGHT_LUT_SIZE - 1) / F;
+ const double formScale = (double) (WEIGHT_LUT_SIZE - 1) / F;
A *= formScale;
B *= formScale;
C *= formScale;
/* F *= formScale; */ /* no need to scale F as we don't use it below here */
- /* For each quad, the du and dx values are the same and so the ellipse is
- * also the same. Note that texel/image access can only be performed using
- * a quad, i.e. it is not possible to get the pixel value for a single
- * tex coord. In order to have a better performance, the access is buffered
- * using the s_buffer/t_buffer and weight_buffer. Only when the buffer is
- * full, then the pixel values are read from the image.
- */
- ddq = 2 * A;
-
+ args.level = level;
+ args.offset = offset;
+
for (j = 0; j < TGSI_QUAD_SIZE; j++) {
/* Heckbert MS thesis, p. 59; scan over the bounding box of the ellipse
* and incrementally update the value of Ax^2+Bxy*Cy^2; when this
* value, q, is less than F, we're inside the ellipse
*/
- float tex_u = -0.5F + s[j] * texture->width0 * scaling;
- float tex_v = -0.5F + t[j] * texture->height0 * scaling;
+ const float tex_u = -0.5F + s[j] * texture->width0 * scaling;
+ const float tex_v = -0.5F + t[j] * texture->height0 * scaling;
- int u0 = (int) floorf(tex_u - box_u);
- int u1 = (int) ceilf(tex_u + box_u);
- int v0 = (int) floorf(tex_v - box_v);
- int v1 = (int) ceilf(tex_v + box_v);
+ const int u0 = (int) floorf(tex_u - box_u);
+ const int u1 = (int) ceilf(tex_u + box_u);
+ const int v0 = (int) floorf(tex_v - box_v);
+ const int v1 = (int) ceilf(tex_v + box_v);
+ const float U = u0 - tex_u;
float num[4] = {0.0F, 0.0F, 0.0F, 0.0F};
- buffer_next = 0;
- den = 0;
- U = u0 - tex_u;
+ unsigned buffer_next = 0;
+ float den = 0;
+ int v;
+ args.face_id = faces[j];
+
for (v = v0; v <= v1; ++v) {
- float V = v - tex_v;
+ const float V = v - tex_v;
float dq = A * (2 * U + 1) + B * V;
float q = (C * V + B * U) * V + A * U * U;
* should not happen, though
*/
const int qClamped = q >= 0.0F ? q : 0;
- float weight = weightLut[qClamped];
+ const float weight = weightLut[qClamped];
weight_buffer[buffer_next] = weight;
s_buffer[buffer_next] = u / ((float) width);
* accelerated img_filter_2d_nearest_XXX functions.
*/
for (jj = 0; jj < buffer_next; jj++) {
- min_filter(sp_sview, sp_samp, s_buffer[jj], t_buffer[jj], p[jj],
- level, sp_sview->faces[j], &rgba_temp[0][jj]);
+ args.s = s_buffer[jj];
+ args.t = t_buffer[jj];
+ args.p = p[jj];
+ min_filter(sp_sview, sp_samp, &args, &rgba_temp[0][jj]);
num[0] += weight_buffer[jj] * rgba_temp[0][jj];
num[1] += weight_buffer[jj] * rgba_temp[1][jj];
num[2] += weight_buffer[jj] * rgba_temp[2][jj];
* accelerated img_filter_2d_nearest_XXX functions.
*/
for (jj = 0; jj < buffer_next; jj++) {
- min_filter(sp_sview, sp_samp, s_buffer[jj], t_buffer[jj], p[jj],
- level, sp_sview->faces[j], &rgba_temp[0][jj]);
+ args.s = s_buffer[jj];
+ args.t = t_buffer[jj];
+ args.p = p[jj];
+ min_filter(sp_sview, sp_samp, &args, &rgba_temp[0][jj]);
num[0] += weight_buffer[jj] * rgba_temp[0][jj];
num[1] += weight_buffer[jj] * rgba_temp[1][jj];
num[2] += weight_buffer[jj] * rgba_temp[2][jj];
rgba[2]=0;
rgba[3]=0;*/
/* not enough pixels in resampling, resort to direct interpolation */
- min_filter(sp_sview, sp_samp, s[j], t[j], p[j], level,
- sp_sview->faces[j], &rgba_temp[0][j]);
+ args.s = s[j];
+ args.t = t[j];
+ args.p = p[j];
+ min_filter(sp_sview, sp_samp, &args, &rgba_temp[0][j]);
den = 1;
num[0] = rgba_temp[0][j];
num[1] = rgba_temp[1][j];
}
+/**
+ * Get mip level relative to base level for linear mip filter
+ */
+static void
+mip_rel_level_linear_aniso(const struct sp_sampler_view *sp_sview,
+ const struct sp_sampler *sp_samp,
+ const float lod[TGSI_QUAD_SIZE],
+ float level[TGSI_QUAD_SIZE])
+{
+ mip_rel_level_linear(sp_sview, sp_samp, lod, level);
+}
+
/**
* Sample 2D texture using an anisotropic filter.
*/
static void
-mip_filter_linear_aniso(struct sp_sampler_view *sp_sview,
- struct sp_sampler *sp_samp,
+mip_filter_linear_aniso(const struct sp_sampler_view *sp_sview,
+ const struct sp_sampler *sp_samp,
img_filter_func min_filter,
img_filter_func mag_filter,
const float s[TGSI_QUAD_SIZE],
const float p[TGSI_QUAD_SIZE],
const float c0[TGSI_QUAD_SIZE],
const float lod_in[TGSI_QUAD_SIZE],
- enum tgsi_sampler_control control,
+ const struct filter_args *filt_args,
float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
{
const struct pipe_resource *texture = sp_sview->base.texture;
float lambda;
float lod[TGSI_QUAD_SIZE];
- float s_to_u = u_minify(texture->width0, psview->u.tex.first_level);
- float t_to_v = u_minify(texture->height0, psview->u.tex.first_level);
- float dudx = (s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]) * s_to_u;
- float dudy = (s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]) * s_to_u;
- float dvdx = (t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT]) * t_to_v;
- float dvdy = (t[QUAD_TOP_LEFT] - t[QUAD_BOTTOM_LEFT]) * t_to_v;
-
- if (control == tgsi_sampler_lod_bias ||
- control == tgsi_sampler_lod_none ||
+ const float s_to_u = u_minify(texture->width0, psview->u.tex.first_level);
+ const float t_to_v = u_minify(texture->height0, psview->u.tex.first_level);
+ const float dudx = (s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]) * s_to_u;
+ const float dudy = (s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]) * s_to_u;
+ const float dvdx = (t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT]) * t_to_v;
+ const float dvdy = (t[QUAD_TOP_LEFT] - t[QUAD_BOTTOM_LEFT]) * t_to_v;
+ struct img_filter_args args;
+
+ args.offset = filt_args->offset;
+
+ if (filt_args->control == TGSI_SAMPLER_LOD_BIAS ||
+ filt_args->control == TGSI_SAMPLER_LOD_NONE ||
/* XXX FIXME */
- control == tgsi_sampler_derivs_explicit) {
+ filt_args->control == TGSI_SAMPLER_DERIVS_EXPLICIT) {
/* note: instead of working with Px and Py, we will use the
* squared length instead, to avoid sqrt.
*/
- float Px2 = dudx * dudx + dvdx * dvdx;
- float Py2 = dudy * dudy + dvdy * dvdy;
+ const float Px2 = dudx * dudx + dvdx * dvdx;
+ const float Py2 = dudy * dudy + dvdy * dvdy;
float Pmax2;
float Pmin2;
* this since 0.5*log(x) = log(sqrt(x))
*/
lambda = 0.5F * util_fast_log2(Pmin2) + sp_samp->base.lod_bias;
- compute_lod(&sp_samp->base, control, lambda, lod_in, lod);
+ compute_lod(&sp_samp->base, filt_args->control, lambda, lod_in, lod);
}
else {
- assert(control == tgsi_sampler_lod_explicit ||
- control == tgsi_sampler_lod_zero);
- compute_lod(&sp_samp->base, control, sp_samp->base.lod_bias, lod_in, lod);
+ assert(filt_args->control == TGSI_SAMPLER_LOD_EXPLICIT ||
+ filt_args->control == TGSI_SAMPLER_LOD_ZERO);
+ compute_lod(&sp_samp->base, filt_args->control, sp_samp->base.lod_bias, lod_in, lod);
}
/* XXX: Take into account all lod values.
*/
if (level0 >= (int) psview->u.tex.last_level) {
int j;
- for (j = 0; j < TGSI_QUAD_SIZE; j++)
- min_filter(sp_sview, sp_samp, s[j], t[j], p[j], psview->u.tex.last_level,
- sp_sview->faces[j], &rgba[0][j]);
+ for (j = 0; j < TGSI_QUAD_SIZE; j++) {
+ args.s = s[j];
+ args.t = t[j];
+ args.p = p[j];
+ args.level = psview->u.tex.last_level;
+ args.face_id = filt_args->faces[j];
+ /*
+ * XXX: we overwrote any linear filter with nearest, so this
+ * isn't right (albeit if last level is 1x1 and no border it
+ * will work just the same).
+ */
+ min_filter(sp_sview, sp_samp, &args, &rgba[0][j]);
+ }
}
else {
/* don't bother interpolating between multiple LODs; it doesn't
* seem to be worth the extra running time.
*/
img_filter_2d_ewa(sp_sview, sp_samp, min_filter, mag_filter,
- s, t, p, level0,
- dudx, dvdx, dudy, dvdy, rgba);
+ s, t, p, filt_args->faces, filt_args->offset,
+ level0, dudx, dvdx, dudy, dvdy, rgba);
}
if (DEBUG_TEX) {
}
}
+/**
+ * Get mip level relative to base level for linear mip filter
+ */
+static void
+mip_rel_level_linear_2d_linear_repeat_POT(
+ const struct sp_sampler_view *sp_sview,
+ const struct sp_sampler *sp_samp,
+ const float lod[TGSI_QUAD_SIZE],
+ float level[TGSI_QUAD_SIZE])
+{
+ mip_rel_level_linear(sp_sview, sp_samp, lod, level);
+}
/**
* Specialized version of mip_filter_linear with hard-wired calls to
*/
static void
mip_filter_linear_2d_linear_repeat_POT(
- struct sp_sampler_view *sp_sview,
- struct sp_sampler *sp_samp,
+ const struct sp_sampler_view *sp_sview,
+ const struct sp_sampler *sp_samp,
img_filter_func min_filter,
img_filter_func mag_filter,
const float s[TGSI_QUAD_SIZE],
const float p[TGSI_QUAD_SIZE],
const float c0[TGSI_QUAD_SIZE],
const float lod_in[TGSI_QUAD_SIZE],
- enum tgsi_sampler_control control,
+ const struct filter_args *filt_args,
float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
{
const struct pipe_sampler_view *psview = &sp_sview->base;
int j;
float lod[TGSI_QUAD_SIZE];
- compute_lambda_lod(sp_sview, sp_samp, s, t, p, lod_in, control, lod);
+ compute_lambda_lod(sp_sview, sp_samp, s, t, p, lod_in, filt_args->control, lod);
for (j = 0; j < TGSI_QUAD_SIZE; j++) {
- int level0 = psview->u.tex.first_level + (int)lod[j];
-
+ const int level0 = psview->u.tex.first_level + (int)lod[j];
+ struct img_filter_args args;
/* Catches both negative and large values of level0:
*/
+ args.s = s[j];
+ args.t = t[j];
+ args.p = p[j];
+ args.face_id = filt_args->faces[j];
+ args.offset = filt_args->offset;
+ args.gather_only = filt_args->control == TGSI_SAMPLER_GATHER;
if ((unsigned)level0 >= psview->u.tex.last_level) {
if (level0 < 0)
- img_filter_2d_linear_repeat_POT(sp_sview, sp_samp, s[j], t[j], p[j],
- psview->u.tex.first_level,
- sp_sview->faces[j], &rgba[0][j]);
+ args.level = psview->u.tex.first_level;
else
- img_filter_2d_linear_repeat_POT(sp_sview, sp_samp, s[j], t[j], p[j],
- psview->u.tex.last_level,
- sp_sview->faces[j], &rgba[0][j]);
+ args.level = psview->u.tex.last_level;
+ img_filter_2d_linear_repeat_POT(sp_sview, sp_samp, &args,
+ &rgba[0][j]);
}
else {
- float levelBlend = frac(lod[j]);
+ const float levelBlend = frac(lod[j]);
float rgbax[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE];
int c;
- img_filter_2d_linear_repeat_POT(sp_sview, sp_samp, s[j], t[j], p[j], level0,
- sp_sview->faces[j], &rgbax[0][0]);
- img_filter_2d_linear_repeat_POT(sp_sview, sp_samp, s[j], t[j], p[j], level0+1,
- sp_sview->faces[j], &rgbax[0][1]);
+ args.level = level0;
+ img_filter_2d_linear_repeat_POT(sp_sview, sp_samp, &args, &rgbax[0][0]);
+ args.level = level0+1;
+ img_filter_2d_linear_repeat_POT(sp_sview, sp_samp, &args, &rgbax[0][1]);
for (c = 0; c < TGSI_NUM_CHANNELS; c++)
rgba[c][j] = lerp(levelBlend, rgbax[c][0], rgbax[c][1]);
}
}
+static const struct sp_filter_funcs funcs_linear = {
+ mip_rel_level_linear,
+ mip_filter_linear
+};
+
+static const struct sp_filter_funcs funcs_nearest = {
+ mip_rel_level_nearest,
+ mip_filter_nearest
+};
+
+static const struct sp_filter_funcs funcs_none = {
+ mip_rel_level_none,
+ mip_filter_none
+};
+
+static const struct sp_filter_funcs funcs_none_no_filter_select = {
+ mip_rel_level_none_no_filter_select,
+ mip_filter_none_no_filter_select
+};
+
+static const struct sp_filter_funcs funcs_linear_aniso = {
+ mip_rel_level_linear_aniso,
+ mip_filter_linear_aniso
+};
+
+static const struct sp_filter_funcs funcs_linear_2d_linear_repeat_POT = {
+ mip_rel_level_linear_2d_linear_repeat_POT,
+ mip_filter_linear_2d_linear_repeat_POT
+};
/**
* Do shadow/depth comparisons.
*/
static void
-sample_compare(struct sp_sampler_view *sp_sview,
- struct sp_sampler *sp_samp,
+sample_compare(const struct sp_sampler_view *sp_sview,
+ const struct sp_sampler *sp_samp,
const float s[TGSI_QUAD_SIZE],
const float t[TGSI_QUAD_SIZE],
const float p[TGSI_QUAD_SIZE],
float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
{
const struct pipe_sampler_state *sampler = &sp_samp->base;
- int j;
- int k[4];
+ int j, v;
+ int k[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE];
float pc[4];
- const struct util_format_description *format_desc;
- unsigned chan_type;
+ const struct util_format_description *format_desc =
+ util_format_description(sp_sview->base.format);
+ /* not entirely sure we couldn't end up with non-valid swizzle here */
+ const unsigned chan_type =
+ format_desc->swizzle[0] <= PIPE_SWIZZLE_W ?
+ format_desc->channel[format_desc->swizzle[0]].type :
+ UTIL_FORMAT_TYPE_FLOAT;
+ const bool is_gather = (control == TGSI_SAMPLER_GATHER);
/**
* Compare texcoord 'p' (aka R) against texture value 'rgba[0]'
* RGBA channels. We look at the red channel here.
*/
- if (sp_sview->base.texture->target == PIPE_TEXTURE_2D_ARRAY ||
- sp_sview->base.texture->target == PIPE_TEXTURE_CUBE) {
+ if (sp_sview->base.target == PIPE_TEXTURE_2D_ARRAY ||
+ sp_sview->base.target == PIPE_TEXTURE_CUBE) {
pc[0] = c0[0];
pc[1] = c0[1];
pc[2] = c0[2];
pc[3] = c0[3];
- } else if (sp_sview->base.texture->target == PIPE_TEXTURE_CUBE_ARRAY) {
+ } else if (sp_sview->base.target == PIPE_TEXTURE_CUBE_ARRAY) {
pc[0] = c1[0];
pc[1] = c1[1];
pc[2] = c1[2];
pc[3] = p[3];
}
- format_desc = util_format_description(sp_sview->base.format);
- /* not entirely sure we couldn't end up with non-valid swizzle here */
- chan_type = format_desc->swizzle[0] <= UTIL_FORMAT_SWIZZLE_W ?
- format_desc->channel[format_desc->swizzle[0]].type :
- UTIL_FORMAT_TYPE_FLOAT;
if (chan_type != UTIL_FORMAT_TYPE_FLOAT) {
/*
* clamping is a result of conversion to texture format, hence
pc[3] = CLAMP(pc[3], 0.0F, 1.0F);
}
- /* compare four texcoords vs. four texture samples */
- switch (sampler->compare_func) {
- case PIPE_FUNC_LESS:
- k[0] = pc[0] < rgba[0][0];
- k[1] = pc[1] < rgba[0][1];
- k[2] = pc[2] < rgba[0][2];
- k[3] = pc[3] < rgba[0][3];
- break;
- case PIPE_FUNC_LEQUAL:
- k[0] = pc[0] <= rgba[0][0];
- k[1] = pc[1] <= rgba[0][1];
- k[2] = pc[2] <= rgba[0][2];
- k[3] = pc[3] <= rgba[0][3];
- break;
- case PIPE_FUNC_GREATER:
- k[0] = pc[0] > rgba[0][0];
- k[1] = pc[1] > rgba[0][1];
- k[2] = pc[2] > rgba[0][2];
- k[3] = pc[3] > rgba[0][3];
- break;
- case PIPE_FUNC_GEQUAL:
- k[0] = pc[0] >= rgba[0][0];
- k[1] = pc[1] >= rgba[0][1];
- k[2] = pc[2] >= rgba[0][2];
- k[3] = pc[3] >= rgba[0][3];
- break;
- case PIPE_FUNC_EQUAL:
- k[0] = pc[0] == rgba[0][0];
- k[1] = pc[1] == rgba[0][1];
- k[2] = pc[2] == rgba[0][2];
- k[3] = pc[3] == rgba[0][3];
- break;
- case PIPE_FUNC_NOTEQUAL:
- k[0] = pc[0] != rgba[0][0];
- k[1] = pc[1] != rgba[0][1];
- k[2] = pc[2] != rgba[0][2];
- k[3] = pc[3] != rgba[0][3];
- break;
- case PIPE_FUNC_ALWAYS:
- k[0] = k[1] = k[2] = k[3] = 1;
- break;
- case PIPE_FUNC_NEVER:
- k[0] = k[1] = k[2] = k[3] = 0;
- break;
- default:
- k[0] = k[1] = k[2] = k[3] = 0;
- assert(0);
- break;
+ for (v = 0; v < (is_gather ? TGSI_NUM_CHANNELS : 1); v++) {
+ /* compare four texcoords vs. four texture samples */
+ switch (sampler->compare_func) {
+ case PIPE_FUNC_LESS:
+ k[v][0] = pc[0] < rgba[v][0];
+ k[v][1] = pc[1] < rgba[v][1];
+ k[v][2] = pc[2] < rgba[v][2];
+ k[v][3] = pc[3] < rgba[v][3];
+ break;
+ case PIPE_FUNC_LEQUAL:
+ k[v][0] = pc[0] <= rgba[v][0];
+ k[v][1] = pc[1] <= rgba[v][1];
+ k[v][2] = pc[2] <= rgba[v][2];
+ k[v][3] = pc[3] <= rgba[v][3];
+ break;
+ case PIPE_FUNC_GREATER:
+ k[v][0] = pc[0] > rgba[v][0];
+ k[v][1] = pc[1] > rgba[v][1];
+ k[v][2] = pc[2] > rgba[v][2];
+ k[v][3] = pc[3] > rgba[v][3];
+ break;
+ case PIPE_FUNC_GEQUAL:
+ k[v][0] = pc[0] >= rgba[v][0];
+ k[v][1] = pc[1] >= rgba[v][1];
+ k[v][2] = pc[2] >= rgba[v][2];
+ k[v][3] = pc[3] >= rgba[v][3];
+ break;
+ case PIPE_FUNC_EQUAL:
+ k[v][0] = pc[0] == rgba[v][0];
+ k[v][1] = pc[1] == rgba[v][1];
+ k[v][2] = pc[2] == rgba[v][2];
+ k[v][3] = pc[3] == rgba[v][3];
+ break;
+ case PIPE_FUNC_NOTEQUAL:
+ k[v][0] = pc[0] != rgba[v][0];
+ k[v][1] = pc[1] != rgba[v][1];
+ k[v][2] = pc[2] != rgba[v][2];
+ k[v][3] = pc[3] != rgba[v][3];
+ break;
+ case PIPE_FUNC_ALWAYS:
+ k[v][0] = k[v][1] = k[v][2] = k[v][3] = 1;
+ break;
+ case PIPE_FUNC_NEVER:
+ k[v][0] = k[v][1] = k[v][2] = k[v][3] = 0;
+ break;
+ default:
+ k[v][0] = k[v][1] = k[v][2] = k[v][3] = 0;
+ assert(0);
+ break;
+ }
}
- for (j = 0; j < TGSI_QUAD_SIZE; j++) {
- rgba[0][j] = k[j];
- rgba[1][j] = k[j];
- rgba[2][j] = k[j];
- rgba[3][j] = 1.0F;
+ if (is_gather) {
+ for (j = 0; j < TGSI_QUAD_SIZE; j++) {
+ for (v = 0; v < TGSI_NUM_CHANNELS; v++) {
+ rgba[v][j] = k[v][j];
+ }
+ }
+ } else {
+ for (j = 0; j < TGSI_QUAD_SIZE; j++) {
+ rgba[0][j] = k[0][j];
+ rgba[1][j] = k[0][j];
+ rgba[2][j] = k[0][j];
+ rgba[3][j] = 1.0F;
+ }
}
}
-
static void
do_swizzling(const struct pipe_sampler_view *sview,
float in[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE],
const unsigned swizzle_a = sview->swizzle_a;
switch (swizzle_r) {
- case PIPE_SWIZZLE_ZERO:
+ case PIPE_SWIZZLE_0:
for (j = 0; j < 4; j++)
out[0][j] = 0.0f;
break;
- case PIPE_SWIZZLE_ONE:
+ case PIPE_SWIZZLE_1:
for (j = 0; j < 4; j++)
out[0][j] = 1.0f;
break;
}
switch (swizzle_g) {
- case PIPE_SWIZZLE_ZERO:
+ case PIPE_SWIZZLE_0:
for (j = 0; j < 4; j++)
out[1][j] = 0.0f;
break;
- case PIPE_SWIZZLE_ONE:
+ case PIPE_SWIZZLE_1:
for (j = 0; j < 4; j++)
out[1][j] = 1.0f;
break;
}
switch (swizzle_b) {
- case PIPE_SWIZZLE_ZERO:
+ case PIPE_SWIZZLE_0:
for (j = 0; j < 4; j++)
out[2][j] = 0.0f;
break;
- case PIPE_SWIZZLE_ONE:
+ case PIPE_SWIZZLE_1:
for (j = 0; j < 4; j++)
out[2][j] = 1.0f;
break;
}
switch (swizzle_a) {
- case PIPE_SWIZZLE_ZERO:
+ case PIPE_SWIZZLE_0:
for (j = 0; j < 4; j++)
out[3][j] = 0.0f;
break;
- case PIPE_SWIZZLE_ONE:
+ case PIPE_SWIZZLE_1:
for (j = 0; j < 4; j++)
out[3][j] = 1.0f;
break;
case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
return wrap_nearest_unorm_clamp_to_border;
default:
- assert(0);
+ debug_printf("illegal wrap mode %d with non-normalized coords\n", mode);
return wrap_nearest_unorm_clamp;
}
}
case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
return wrap_linear_unorm_clamp_to_border;
default:
- assert(0);
+ debug_printf("illegal wrap mode %d with non-normalized coords\n", mode);
return wrap_linear_unorm_clamp;
}
}
/**
* Is swizzling needed for the given state key?
*/
-static INLINE bool
+static inline bool
any_swizzle(const struct pipe_sampler_view *view)
{
- return (view->swizzle_r != PIPE_SWIZZLE_RED ||
- view->swizzle_g != PIPE_SWIZZLE_GREEN ||
- view->swizzle_b != PIPE_SWIZZLE_BLUE ||
- view->swizzle_a != PIPE_SWIZZLE_ALPHA);
+ return (view->swizzle_r != PIPE_SWIZZLE_X ||
+ view->swizzle_g != PIPE_SWIZZLE_Y ||
+ view->swizzle_b != PIPE_SWIZZLE_Z ||
+ view->swizzle_a != PIPE_SWIZZLE_W);
}
static img_filter_func
get_img_filter(const struct sp_sampler_view *sp_sview,
const struct pipe_sampler_state *sampler,
- unsigned filter)
+ unsigned filter, bool gather)
{
- switch (sp_sview->base.texture->target) {
+ switch (sp_sview->base.target) {
case PIPE_BUFFER:
case PIPE_TEXTURE_1D:
if (filter == PIPE_TEX_FILTER_NEAREST)
case PIPE_TEXTURE_RECT:
/* Try for fast path:
*/
- if (sp_sview->pot2d &&
+ if (!gather && sp_sview->pot2d &&
sampler->wrap_s == sampler->wrap_t &&
sampler->normalized_coords)
{
}
}
+/**
+ * Get mip filter funcs, and optionally both img min filter and img mag
+ * filter. Note that both img filter function pointers must be either non-NULL
+ * or NULL.
+ */
+static void
+get_filters(const struct sp_sampler_view *sp_sview,
+ const struct sp_sampler *sp_samp,
+ const enum tgsi_sampler_control control,
+ const struct sp_filter_funcs **funcs,
+ img_filter_func *min,
+ img_filter_func *mag)
+{
+ assert(funcs);
+ if (control == TGSI_SAMPLER_GATHER) {
+ *funcs = &funcs_nearest;
+ if (min) {
+ *min = get_img_filter(sp_sview, &sp_samp->base,
+ PIPE_TEX_FILTER_LINEAR, true);
+ }
+ } else if (sp_sview->pot2d & sp_samp->min_mag_equal_repeat_linear) {
+ *funcs = &funcs_linear_2d_linear_repeat_POT;
+ } else {
+ *funcs = sp_samp->filter_funcs;
+ if (min) {
+ assert(mag);
+ *min = get_img_filter(sp_sview, &sp_samp->base,
+ sp_samp->min_img_filter, false);
+ if (sp_samp->min_mag_equal) {
+ *mag = *min;
+ } else {
+ *mag = get_img_filter(sp_sview, &sp_samp->base,
+ sp_samp->base.mag_img_filter, false);
+ }
+ }
+ }
+}
static void
-sample_mip(struct sp_sampler_view *sp_sview,
- struct sp_sampler *sp_samp,
+sample_mip(const struct sp_sampler_view *sp_sview,
+ const struct sp_sampler *sp_samp,
const float s[TGSI_QUAD_SIZE],
const float t[TGSI_QUAD_SIZE],
const float p[TGSI_QUAD_SIZE],
const float c0[TGSI_QUAD_SIZE],
const float lod[TGSI_QUAD_SIZE],
- enum tgsi_sampler_control control,
+ const struct filter_args *filt_args,
float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
{
- mip_filter_func mip_filter;
+ const struct sp_filter_funcs *funcs = NULL;
img_filter_func min_img_filter = NULL;
img_filter_func mag_img_filter = NULL;
- if (sp_sview->pot2d & sp_samp->min_mag_equal_repeat_linear) {
- mip_filter = mip_filter_linear_2d_linear_repeat_POT;
- }
- else {
- mip_filter = sp_samp->mip_filter;
- min_img_filter = get_img_filter(sp_sview, &sp_samp->base, sp_samp->min_img_filter);
- if (sp_samp->min_mag_equal) {
- mag_img_filter = min_img_filter;
- }
- else {
- mag_img_filter = get_img_filter(sp_sview, &sp_samp->base, sp_samp->base.mag_img_filter);
- }
- }
+ get_filters(sp_sview, sp_samp, filt_args->control,
+ &funcs, &min_img_filter, &mag_img_filter);
- mip_filter(sp_sview, sp_samp, min_img_filter, mag_img_filter,
- s, t, p, c0, lod, control, rgba);
+ funcs->filter(sp_sview, sp_samp, min_img_filter, mag_img_filter,
+ s, t, p, c0, lod, filt_args, rgba);
if (sp_samp->base.compare_mode != PIPE_TEX_COMPARE_NONE) {
- sample_compare(sp_sview, sp_samp, s, t, p, c0, lod, control, rgba);
+ sample_compare(sp_sview, sp_samp, s, t, p, c0,
+ lod, filt_args->control, rgba);
}
- if (sp_sview->need_swizzle) {
+ if (sp_sview->need_swizzle && filt_args->control != TGSI_SAMPLER_GATHER) {
float rgba_temp[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE];
memcpy(rgba_temp, rgba, sizeof(rgba_temp));
do_swizzling(&sp_sview->base, rgba_temp, rgba);
/**
- * Use 3D texcoords to choose a cube face, then sample the 2D cube faces.
- * Put face info into the sampler faces[] array.
+ * This function uses cube texture coordinates to choose a face of a cube and
+ * computes the 2D cube face coordinates. Puts face info into the sampler
+ * faces[] array.
*/
static void
-sample_cube(struct sp_sampler_view *sp_sview,
- struct sp_sampler *sp_samp,
- const float s[TGSI_QUAD_SIZE],
- const float t[TGSI_QUAD_SIZE],
- const float p[TGSI_QUAD_SIZE],
- const float c0[TGSI_QUAD_SIZE],
- const float c1[TGSI_QUAD_SIZE],
- enum tgsi_sampler_control control,
- float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
+convert_cube(const struct sp_sampler_view *sp_sview,
+ const struct sp_sampler *sp_samp,
+ const float s[TGSI_QUAD_SIZE],
+ const float t[TGSI_QUAD_SIZE],
+ const float p[TGSI_QUAD_SIZE],
+ const float c0[TGSI_QUAD_SIZE],
+ float ssss[TGSI_QUAD_SIZE],
+ float tttt[TGSI_QUAD_SIZE],
+ float pppp[TGSI_QUAD_SIZE],
+ uint faces[TGSI_QUAD_SIZE])
{
unsigned j;
- float ssss[4], tttt[4];
-
- /* Not actually used, but the intermediate steps that do the
- * dereferencing don't know it.
- */
- static float pppp[4] = { 0, 0, 0, 0 };
pppp[0] = c0[0];
pppp[1] = c0[1];
const float arx = fabsf(rx), ary = fabsf(ry), arz = fabsf(rz);
if (arx >= ary && arx >= arz) {
- float sign = (rx >= 0.0F) ? 1.0F : -1.0F;
- uint face = (rx >= 0.0F) ? PIPE_TEX_FACE_POS_X : PIPE_TEX_FACE_NEG_X;
+ const float sign = (rx >= 0.0F) ? 1.0F : -1.0F;
+ const uint face = (rx >= 0.0F) ?
+ PIPE_TEX_FACE_POS_X : PIPE_TEX_FACE_NEG_X;
for (j = 0; j < TGSI_QUAD_SIZE; j++) {
const float ima = -0.5F / fabsf(s[j]);
ssss[j] = sign * p[j] * ima + 0.5F;
tttt[j] = t[j] * ima + 0.5F;
- sp_sview->faces[j] = face;
+ faces[j] = face;
}
}
else if (ary >= arx && ary >= arz) {
- float sign = (ry >= 0.0F) ? 1.0F : -1.0F;
- uint face = (ry >= 0.0F) ? PIPE_TEX_FACE_POS_Y : PIPE_TEX_FACE_NEG_Y;
+ const float sign = (ry >= 0.0F) ? 1.0F : -1.0F;
+ const uint face = (ry >= 0.0F) ?
+ PIPE_TEX_FACE_POS_Y : PIPE_TEX_FACE_NEG_Y;
for (j = 0; j < TGSI_QUAD_SIZE; j++) {
const float ima = -0.5F / fabsf(t[j]);
ssss[j] = -s[j] * ima + 0.5F;
tttt[j] = sign * -p[j] * ima + 0.5F;
- sp_sview->faces[j] = face;
+ faces[j] = face;
}
}
else {
- float sign = (rz >= 0.0F) ? 1.0F : -1.0F;
- uint face = (rz >= 0.0F) ? PIPE_TEX_FACE_POS_Z : PIPE_TEX_FACE_NEG_Z;
+ const float sign = (rz >= 0.0F) ? 1.0F : -1.0F;
+ const uint face = (rz >= 0.0F) ?
+ PIPE_TEX_FACE_POS_Z : PIPE_TEX_FACE_NEG_Z;
for (j = 0; j < TGSI_QUAD_SIZE; j++) {
const float ima = -0.5F / fabsf(p[j]);
ssss[j] = sign * -s[j] * ima + 0.5F;
tttt[j] = t[j] * ima + 0.5F;
- sp_sview->faces[j] = face;
+ faces[j] = face;
}
}
}
-
- sample_mip(sp_sview, sp_samp, ssss, tttt, pppp, c0, c1, control, rgba);
}
static void
-sp_get_dims(struct sp_sampler_view *sp_sview, int level,
+sp_get_dims(const struct sp_sampler_view *sp_sview,
+ int level,
int dims[4])
{
const struct pipe_sampler_view *view = &sp_sview->base;
const struct pipe_resource *texture = view->texture;
- if (texture->target == PIPE_BUFFER) {
- dims[0] = (view->u.buf.last_element - view->u.buf.first_element) + 1;
+ if (view->target == PIPE_BUFFER) {
+ dims[0] = view->u.buf.size / util_format_get_blocksize(view->format);
/* the other values are undefined, but let's avoid potential valgrind
* warnings.
*/
dims[3] = view->u.tex.last_level - view->u.tex.first_level + 1;
dims[0] = u_minify(texture->width0, level);
- switch(texture->target) {
+ switch (view->target) {
case PIPE_TEXTURE_1D_ARRAY:
dims[1] = view->u.tex.last_layer - view->u.tex.first_layer + 1;
/* fallthrough */
* coords to the texture image size.
*/
static void
-sp_get_texels(struct sp_sampler_view *sp_sview,
+sp_get_texels(const struct sp_sampler_view *sp_sview,
const int v_i[TGSI_QUAD_SIZE],
const int v_j[TGSI_QUAD_SIZE],
const int v_k[TGSI_QUAD_SIZE],
const struct pipe_resource *texture = sp_sview->base.texture;
int j, c;
const float *tx;
- int width, height, depth;
-
- addr.value = 0;
/* TODO write a better test for LOD */
- addr.bits.level = lod[0];
+ const unsigned level =
+ sp_sview->base.target == PIPE_BUFFER ? 0 :
+ CLAMP(lod[0] + sp_sview->base.u.tex.first_level,
+ sp_sview->base.u.tex.first_level,
+ sp_sview->base.u.tex.last_level);
+ const int width = u_minify(texture->width0, level);
+ const int height = u_minify(texture->height0, level);
+ const int depth = u_minify(texture->depth0, level);
+ unsigned elem_size, first_element, last_element;
- width = u_minify(texture->width0, addr.bits.level);
- height = u_minify(texture->height0, addr.bits.level);
- depth = u_minify(texture->depth0, addr.bits.level);
+ addr.value = 0;
+ addr.bits.level = level;
- switch(texture->target) {
+ switch (sp_sview->base.target) {
case PIPE_BUFFER:
- case PIPE_TEXTURE_1D:
+ elem_size = util_format_get_blocksize(sp_sview->base.format);
+ first_element = sp_sview->base.u.buf.offset / elem_size;
+ last_element = (sp_sview->base.u.buf.offset +
+ sp_sview->base.u.buf.size) / elem_size - 1;
for (j = 0; j < TGSI_QUAD_SIZE; j++) {
- int x = CLAMP(v_i[j] + offset[0], 0, width - 1);
+ const int x = CLAMP(v_i[j] + offset[0] +
+ first_element,
+ first_element,
+ last_element);
tx = get_texel_2d_no_border(sp_sview, addr, x, 0);
for (c = 0; c < 4; c++) {
rgba[c][j] = tx[c];
}
}
break;
+ case PIPE_TEXTURE_1D:
+ for (j = 0; j < TGSI_QUAD_SIZE; j++) {
+ const int x = CLAMP(v_i[j] + offset[0], 0, width - 1);
+ tx = get_texel_2d_no_border(sp_sview, addr, x,
+ sp_sview->base.u.tex.first_layer);
+ for (c = 0; c < 4; c++) {
+ rgba[c][j] = tx[c];
+ }
+ }
+ break;
case PIPE_TEXTURE_1D_ARRAY:
for (j = 0; j < TGSI_QUAD_SIZE; j++) {
- int x = CLAMP(v_i[j] + offset[0], 0, width - 1);
- int y = CLAMP(v_j[j], sp_sview->base.u.tex.first_layer, sp_sview->base.u.tex.last_layer);
+ const int x = CLAMP(v_i[j] + offset[0], 0, width - 1);
+ const int y = CLAMP(v_j[j], sp_sview->base.u.tex.first_layer,
+ sp_sview->base.u.tex.last_layer);
tx = get_texel_2d_no_border(sp_sview, addr, x, y);
for (c = 0; c < 4; c++) {
rgba[c][j] = tx[c];
case PIPE_TEXTURE_2D:
case PIPE_TEXTURE_RECT:
for (j = 0; j < TGSI_QUAD_SIZE; j++) {
- int x = CLAMP(v_i[j] + offset[0], 0, width - 1);
- int y = CLAMP(v_j[j] + offset[1], 0, height - 1);
- tx = get_texel_2d_no_border(sp_sview, addr, x, y);
+ const int x = CLAMP(v_i[j] + offset[0], 0, width - 1);
+ const int y = CLAMP(v_j[j] + offset[1], 0, height - 1);
+ tx = get_texel_3d_no_border(sp_sview, addr, x, y,
+ sp_sview->base.u.tex.first_layer);
for (c = 0; c < 4; c++) {
rgba[c][j] = tx[c];
}
break;
case PIPE_TEXTURE_2D_ARRAY:
for (j = 0; j < TGSI_QUAD_SIZE; j++) {
- int x = CLAMP(v_i[j] + offset[0], 0, width - 1);
- int y = CLAMP(v_j[j] + offset[1], 0, height - 1);
- int layer = CLAMP(v_k[j], sp_sview->base.u.tex.first_layer, sp_sview->base.u.tex.last_layer);
+ const int x = CLAMP(v_i[j] + offset[0], 0, width - 1);
+ const int y = CLAMP(v_j[j] + offset[1], 0, height - 1);
+ const int layer = CLAMP(v_k[j], sp_sview->base.u.tex.first_layer,
+ sp_sview->base.u.tex.last_layer);
tx = get_texel_3d_no_border(sp_sview, addr, x, y, layer);
for (c = 0; c < 4; c++) {
rgba[c][j] = tx[c];
}
break;
case PIPE_TEXTURE_CUBE: /* TXF can't work on CUBE according to spec */
+ case PIPE_TEXTURE_CUBE_ARRAY:
default:
assert(!"Unknown or CUBE texture type in TXF processing\n");
break;
switch (sampler->min_mip_filter) {
case PIPE_TEX_MIPFILTER_NONE:
if (sampler->min_img_filter == sampler->mag_img_filter)
- samp->mip_filter = mip_filter_none_no_filter_select;
+ samp->filter_funcs = &funcs_none_no_filter_select;
else
- samp->mip_filter = mip_filter_none;
+ samp->filter_funcs = &funcs_none;
break;
case PIPE_TEX_MIPFILTER_NEAREST:
- samp->mip_filter = mip_filter_nearest;
+ samp->filter_funcs = &funcs_nearest;
break;
case PIPE_TEX_MIPFILTER_LINEAR:
sampler->max_anisotropy <= 1) {
samp->min_mag_equal_repeat_linear = TRUE;
}
- samp->mip_filter = mip_filter_linear;
+ samp->filter_funcs = &funcs_linear;
/* Anisotropic filtering extension. */
if (sampler->max_anisotropy > 1) {
- samp->mip_filter = mip_filter_linear_aniso;
+ samp->filter_funcs = &funcs_linear_aniso;
/* Override min_img_filter:
* min_img_filter needs to be set to NEAREST since we need to access
compute_lambda_func
-softpipe_get_lambda_func(const struct pipe_sampler_view *view, unsigned shader)
+softpipe_get_lambda_func(const struct pipe_sampler_view *view,
+ enum pipe_shader_type shader)
{
if (shader != PIPE_SHADER_FRAGMENT)
return compute_lambda_vert;
- switch (view->texture->target) {
+ switch (view->target) {
case PIPE_BUFFER:
case PIPE_TEXTURE_1D:
case PIPE_TEXTURE_1D_ARRAY:
const struct pipe_sampler_view *templ)
{
struct sp_sampler_view *sview = CALLOC_STRUCT(sp_sampler_view);
- struct softpipe_resource *spr = (struct softpipe_resource *)resource;
+ const struct softpipe_resource *spr = (struct softpipe_resource *)resource;
if (sview) {
struct pipe_sampler_view *view = &sview->base;
pipe_resource_reference(&view->texture, resource);
view->context = pipe;
+#ifdef DEBUG
+ /*
+ * This is possibly too lenient, but the primary reason is just
+ * to catch state trackers which forget to initialize this, so
+ * it only catches clearly impossible view targets.
+ */
+ if (view->target != resource->target) {
+ if (view->target == PIPE_TEXTURE_1D)
+ assert(resource->target == PIPE_TEXTURE_1D_ARRAY);
+ else if (view->target == PIPE_TEXTURE_1D_ARRAY)
+ assert(resource->target == PIPE_TEXTURE_1D);
+ else if (view->target == PIPE_TEXTURE_2D)
+ assert(resource->target == PIPE_TEXTURE_2D_ARRAY ||
+ resource->target == PIPE_TEXTURE_CUBE ||
+ resource->target == PIPE_TEXTURE_CUBE_ARRAY);
+ else if (view->target == PIPE_TEXTURE_2D_ARRAY)
+ assert(resource->target == PIPE_TEXTURE_2D ||
+ resource->target == PIPE_TEXTURE_CUBE ||
+ resource->target == PIPE_TEXTURE_CUBE_ARRAY);
+ else if (view->target == PIPE_TEXTURE_CUBE)
+ assert(resource->target == PIPE_TEXTURE_CUBE_ARRAY ||
+ resource->target == PIPE_TEXTURE_2D_ARRAY);
+ else if (view->target == PIPE_TEXTURE_CUBE_ARRAY)
+ assert(resource->target == PIPE_TEXTURE_CUBE ||
+ resource->target == PIPE_TEXTURE_2D_ARRAY);
+ else
+ assert(0);
+ }
+#endif
+
if (any_swizzle(view)) {
sview->need_swizzle = TRUE;
}
- if (resource->target == PIPE_TEXTURE_CUBE ||
- resource->target == PIPE_TEXTURE_CUBE_ARRAY)
- sview->get_samples = sample_cube;
- else {
- sview->get_samples = sample_mip;
- }
+ sview->need_cube_convert = (view->target == PIPE_TEXTURE_CUBE ||
+ view->target == PIPE_TEXTURE_CUBE_ARRAY);
sview->pot2d = spr->pot &&
- (resource->target == PIPE_TEXTURE_2D ||
- resource->target == PIPE_TEXTURE_RECT);
+ (view->target == PIPE_TEXTURE_2D ||
+ view->target == PIPE_TEXTURE_RECT);
sview->xpot = util_logbase2( resource->width0 );
sview->ypot = util_logbase2( resource->height0 );
}
+static inline const struct sp_tgsi_sampler *
+sp_tgsi_sampler_cast_c(const struct tgsi_sampler *sampler)
+{
+ return (const struct sp_tgsi_sampler *)sampler;
+}
+
+
static void
sp_tgsi_get_dims(struct tgsi_sampler *tgsi_sampler,
const unsigned sview_index,
int level, int dims[4])
{
- struct sp_tgsi_sampler *sp_samp = (struct sp_tgsi_sampler *)tgsi_sampler;
+ const struct sp_tgsi_sampler *sp_samp =
+ sp_tgsi_sampler_cast_c(tgsi_sampler);
assert(sview_index < PIPE_MAX_SHADER_SAMPLER_VIEWS);
/* always have a view here but texture is NULL if no sampler view was set. */
enum tgsi_sampler_control control,
float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
{
- struct sp_tgsi_sampler *sp_samp = (struct sp_tgsi_sampler *)tgsi_sampler;
+ const struct sp_tgsi_sampler *sp_tgsi_samp =
+ sp_tgsi_sampler_cast_c(tgsi_sampler);
+ const struct sp_sampler_view *sp_sview;
+ const struct sp_sampler *sp_samp;
+ struct filter_args filt_args;
assert(sview_index < PIPE_MAX_SHADER_SAMPLER_VIEWS);
assert(sampler_index < PIPE_MAX_SAMPLERS);
- assert(sp_samp->sp_sampler[sampler_index]);
+ assert(sp_tgsi_samp->sp_sampler[sampler_index]);
+
+ sp_sview = &sp_tgsi_samp->sp_sview[sview_index];
+ sp_samp = sp_tgsi_samp->sp_sampler[sampler_index];
/* always have a view here but texture is NULL if no sampler view was set. */
- if (!sp_samp->sp_sview[sview_index].base.texture) {
+ if (!sp_sview->base.texture) {
int i, j;
for (j = 0; j < TGSI_NUM_CHANNELS; j++) {
for (i = 0; i < TGSI_QUAD_SIZE; i++) {
}
return;
}
- sp_samp->sp_sview[sview_index].get_samples(&sp_samp->sp_sview[sview_index],
- sp_samp->sp_sampler[sampler_index],
- s, t, p, c0, lod, control, rgba);
+
+ filt_args.control = control;
+ filt_args.offset = offset;
+
+ if (sp_sview->need_cube_convert) {
+ float cs[TGSI_QUAD_SIZE];
+ float ct[TGSI_QUAD_SIZE];
+ float cp[TGSI_QUAD_SIZE];
+ uint faces[TGSI_QUAD_SIZE];
+
+ convert_cube(sp_sview, sp_samp, s, t, p, c0, cs, ct, cp, faces);
+
+ filt_args.faces = faces;
+ sample_mip(sp_sview, sp_samp, cs, ct, cp, c0, lod, &filt_args, rgba);
+ } else {
+ static const uint zero_faces[TGSI_QUAD_SIZE] = {0, 0, 0, 0};
+
+ filt_args.faces = zero_faces;
+ sample_mip(sp_sview, sp_samp, s, t, p, c0, lod, &filt_args, rgba);
+ }
}
+static void
+sp_tgsi_query_lod(const struct tgsi_sampler *tgsi_sampler,
+ const unsigned sview_index,
+ const unsigned sampler_index,
+ const float s[TGSI_QUAD_SIZE],
+ const float t[TGSI_QUAD_SIZE],
+ const float p[TGSI_QUAD_SIZE],
+ const float c0[TGSI_QUAD_SIZE],
+ const enum tgsi_sampler_control control,
+ float mipmap[TGSI_QUAD_SIZE],
+ float lod[TGSI_QUAD_SIZE])
+{
+ static const float lod_in[TGSI_QUAD_SIZE] = { 0.0, 0.0, 0.0, 0.0 };
+
+ const struct sp_tgsi_sampler *sp_tgsi_samp =
+ sp_tgsi_sampler_cast_c(tgsi_sampler);
+ const struct sp_sampler_view *sp_sview;
+ const struct sp_sampler *sp_samp;
+ const struct sp_filter_funcs *funcs;
+ int i;
+
+ assert(sview_index < PIPE_MAX_SHADER_SAMPLER_VIEWS);
+ assert(sampler_index < PIPE_MAX_SAMPLERS);
+ assert(sp_tgsi_samp->sp_sampler[sampler_index]);
+
+ sp_sview = &sp_tgsi_samp->sp_sview[sview_index];
+ sp_samp = sp_tgsi_samp->sp_sampler[sampler_index];
+ /* always have a view here but texture is NULL if no sampler view was
+ * set. */
+ if (!sp_sview->base.texture) {
+ for (i = 0; i < TGSI_QUAD_SIZE; i++) {
+ mipmap[i] = 0.0f;
+ lod[i] = 0.0f;
+ }
+ return;
+ }
+
+ if (sp_sview->need_cube_convert) {
+ float cs[TGSI_QUAD_SIZE];
+ float ct[TGSI_QUAD_SIZE];
+ float cp[TGSI_QUAD_SIZE];
+ uint unused_faces[TGSI_QUAD_SIZE];
+
+ convert_cube(sp_sview, sp_samp, s, t, p, c0, cs, ct, cp, unused_faces);
+ compute_lambda_lod_unclamped(sp_sview, sp_samp,
+ cs, ct, cp, lod_in, control, lod);
+ } else {
+ compute_lambda_lod_unclamped(sp_sview, sp_samp,
+ s, t, p, lod_in, control, lod);
+ }
+
+ get_filters(sp_sview, sp_samp, control, &funcs, NULL, NULL);
+ funcs->relative_level(sp_sview, sp_samp, lod, mipmap);
+}
static void
sp_tgsi_get_texel(struct tgsi_sampler *tgsi_sampler,
const int lod[TGSI_QUAD_SIZE], const int8_t offset[3],
float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
{
- struct sp_tgsi_sampler *sp_samp = (struct sp_tgsi_sampler *)tgsi_sampler;
+ const struct sp_tgsi_sampler *sp_samp =
+ sp_tgsi_sampler_cast_c(tgsi_sampler);
assert(sview_index < PIPE_MAX_SHADER_SAMPLER_VIEWS);
/* always have a view here but texture is NULL if no sampler view was set. */
samp->base.get_dims = sp_tgsi_get_dims;
samp->base.get_samples = sp_tgsi_get_samples;
samp->base.get_texel = sp_tgsi_get_texel;
+ samp->base.query_lod = sp_tgsi_query_lod;
return samp;
}
-