*/
static void
lp_build_sample_texel_soa(struct lp_build_sample_context *bld,
- unsigned unit,
+ unsigned sampler_unit,
LLVMValueRef width,
LLVMValueRef height,
LLVMValueRef depth,
LLVMValueRef y_stride,
LLVMValueRef z_stride,
LLVMValueRef data_ptr,
+ LLVMValueRef mipoffsets,
LLVMValueRef texel_out[4])
{
- const struct lp_sampler_static_state *static_state = bld->static_state;
+ const struct lp_static_sampler_state *static_state = bld->static_sampler_state;
const unsigned dims = bld->dims;
struct lp_build_context *int_coord_bld = &bld->int_coord_bld;
LLVMBuilderRef builder = bld->gallivm->builder;
bld->format_desc,
x, y, z, y_stride, z_stride,
&offset, &i, &j);
+ if (mipoffsets) {
+ offset = lp_build_add(&bld->int_coord_bld, offset, mipoffsets);
+ }
if (use_border) {
/* If we can sample the border color, it means that texcoords may
/* select texel color or border color depending on use_border */
LLVMValueRef border_color_ptr =
bld->dynamic_state->border_color(bld->dynamic_state,
- bld->gallivm, unit);
+ bld->gallivm, sampler_unit);
int chan;
for (chan = 0; chan < 4; chan++) {
LLVMValueRef border_chan =
lp_build_const_int32(bld->gallivm, chan));
LLVMValueRef border_chan_vec =
lp_build_broadcast_scalar(&bld->float_vec_bld, border_chan);
+
+ if (!bld->texel_type.floating) {
+ border_chan_vec = LLVMBuildBitCast(builder, border_chan_vec,
+ bld->texel_bld.vec_type, "");
+ }
texel_out[chan] = lp_build_select(&bld->texel_bld, use_border,
border_chan_vec, texel_out[chan]);
}
break;
case PIPE_TEX_WRAP_CLAMP:
- if (bld->static_state->normalized_coords) {
+ if (bld->static_sampler_state->normalized_coords) {
/* scale coord to length */
coord = lp_build_mul(coord_bld, coord, length_f);
}
struct lp_build_context abs_coord_bld = bld->coord_bld;
abs_coord_bld.type.sign = FALSE;
- if (bld->static_state->normalized_coords) {
+ if (bld->static_sampler_state->normalized_coords) {
/* mul by tex size */
coord = lp_build_mul(coord_bld, coord, length_f);
}
}
case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
- {
- LLVMValueRef min;
- if (bld->static_state->normalized_coords) {
- /* scale coord to length */
- coord = lp_build_mul(coord_bld, coord, length_f);
- }
- /* was: clamp to [-0.5, length + 0.5], then sub 0.5 */
- coord = lp_build_sub(coord_bld, coord, half);
- min = lp_build_const_vec(bld->gallivm, coord_bld->type, -1.0F);
- coord = lp_build_clamp(coord_bld, coord, min, length_f);
- /* convert to int, compute lerp weight */
- lp_build_ifloor_fract(coord_bld, coord, &coord0, &weight);
- coord1 = lp_build_add(int_coord_bld, coord0, int_coord_bld->one);
+ if (bld->static_sampler_state->normalized_coords) {
+ /* scale coord to length */
+ coord = lp_build_mul(coord_bld, coord, length_f);
}
+ /* was: clamp to [-0.5, length + 0.5], then sub 0.5 */
+ /* can skip clamp (though might not work for very large coord values */
+ coord = lp_build_sub(coord_bld, coord, half);
+ /* convert to int, compute lerp weight */
+ lp_build_ifloor_fract(coord_bld, coord, &coord0, &weight);
+ coord1 = lp_build_add(int_coord_bld, coord0, int_coord_bld->one);
break;
case PIPE_TEX_WRAP_MIRROR_REPEAT:
case PIPE_TEX_WRAP_MIRROR_CLAMP:
coord = lp_build_abs(coord_bld, coord);
- if (bld->static_state->normalized_coords) {
+ if (bld->static_sampler_state->normalized_coords) {
/* scale coord to length */
coord = lp_build_mul(coord_bld, coord, length_f);
}
abs_coord_bld.type.sign = FALSE;
coord = lp_build_abs(coord_bld, coord);
- if (bld->static_state->normalized_coords) {
+ if (bld->static_sampler_state->normalized_coords) {
/* scale coord to length */
coord = lp_build_mul(coord_bld, coord, length_f);
}
{
coord = lp_build_abs(coord_bld, coord);
- if (bld->static_state->normalized_coords) {
+ if (bld->static_sampler_state->normalized_coords) {
/* scale coord to length */
coord = lp_build_mul(coord_bld, coord, length_f);
}
/* was: clamp to [-0.5, length + 0.5] then sub 0.5 */
- /* skip -0.5 clamp (always positive), do sub first */
+ /* skip clamp - always positive, and other side
+ only potentially matters for very large coords */
coord = lp_build_sub(coord_bld, coord, half);
- coord = lp_build_min(coord_bld, coord, length_f);
/* convert to int, compute lerp weight */
lp_build_ifloor_fract(coord_bld, coord, &coord0, &weight);
case PIPE_TEX_WRAP_CLAMP:
case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
- if (bld->static_state->normalized_coords) {
+ if (bld->static_sampler_state->normalized_coords) {
/* scale coord to length */
coord = lp_build_mul(coord_bld, coord, length_f);
}
break;
case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
- /* Note: this is the same as CLAMP_TO_EDGE, except min = -1 */
- {
- LLVMValueRef min, max;
-
- if (bld->static_state->normalized_coords) {
- /* scale coord to length */
- coord = lp_build_mul(coord_bld, coord, length_f);
- }
-
- icoord = lp_build_ifloor(coord_bld, coord);
-
- /* clamp to [-1, length] */
- min = lp_build_negate(int_coord_bld, int_coord_bld->one);
- max = length;
- icoord = lp_build_clamp(int_coord_bld, icoord, min, max);
+ if (bld->static_sampler_state->normalized_coords) {
+ /* scale coord to length */
+ coord = lp_build_mul(coord_bld, coord, length_f);
}
+ /* no clamp necessary, border masking will handle this */
+ icoord = lp_build_ifloor(coord_bld, coord);
break;
case PIPE_TEX_WRAP_MIRROR_REPEAT:
coord = lp_build_coord_mirror(bld, coord);
/* scale coord to length */
- assert(bld->static_state->normalized_coords);
+ assert(bld->static_sampler_state->normalized_coords);
coord = lp_build_mul(coord_bld, coord, length_f);
/* itrunc == ifloor here */
case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
coord = lp_build_abs(coord_bld, coord);
- if (bld->static_state->normalized_coords) {
+ if (bld->static_sampler_state->normalized_coords) {
/* scale coord to length */
coord = lp_build_mul(coord_bld, coord, length_f);
}
case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
coord = lp_build_abs(coord_bld, coord);
- if (bld->static_state->normalized_coords) {
+ if (bld->static_sampler_state->normalized_coords) {
/* scale coord to length */
coord = lp_build_mul(coord_bld, coord, length_f);
}
/* itrunc == ifloor here */
icoord = lp_build_itrunc(coord_bld, coord);
-
- /* clamp to [0, length] */
- icoord = lp_build_min(int_coord_bld, icoord, length);
break;
default:
*/
static void
lp_build_sample_image_nearest(struct lp_build_sample_context *bld,
- unsigned unit,
+ unsigned sampler_unit,
LLVMValueRef size,
LLVMValueRef row_stride_vec,
LLVMValueRef img_stride_vec,
LLVMValueRef data_ptr,
+ LLVMValueRef mipoffsets,
LLVMValueRef s,
LLVMValueRef t,
LLVMValueRef r,
LLVMValueRef flt_width_vec;
LLVMValueRef flt_height_vec;
LLVMValueRef flt_depth_vec;
- LLVMValueRef x, y, z;
+ LLVMValueRef x, y = NULL, z = NULL;
lp_build_extract_image_sizes(bld,
- bld->int_size_type,
+ &bld->int_size_bld,
bld->int_coord_type,
size,
&width_vec, &height_vec, &depth_vec);
flt_size = lp_build_int_to_float(&bld->float_size_bld, size);
lp_build_extract_image_sizes(bld,
- bld->float_size_type,
+ &bld->float_size_bld,
bld->coord_type,
flt_size,
&flt_width_vec, &flt_height_vec, &flt_depth_vec);
* Compute integer texcoords.
*/
x = lp_build_sample_wrap_nearest(bld, s, width_vec, flt_width_vec,
- bld->static_state->pot_width,
- bld->static_state->wrap_s);
+ bld->static_texture_state->pot_width,
+ bld->static_sampler_state->wrap_s);
lp_build_name(x, "tex.x.wrapped");
if (dims >= 2) {
y = lp_build_sample_wrap_nearest(bld, t, height_vec, flt_height_vec,
- bld->static_state->pot_height,
- bld->static_state->wrap_t);
+ bld->static_texture_state->pot_height,
+ bld->static_sampler_state->wrap_t);
lp_build_name(y, "tex.y.wrapped");
if (dims == 3) {
z = lp_build_sample_wrap_nearest(bld, r, depth_vec, flt_depth_vec,
- bld->static_state->pot_depth,
- bld->static_state->wrap_r);
+ bld->static_texture_state->pot_depth,
+ bld->static_sampler_state->wrap_r);
lp_build_name(z, "tex.z.wrapped");
}
- else if (bld->static_state->target == PIPE_TEXTURE_CUBE) {
- z = r;
- }
- else {
- z = NULL;
- }
}
- else {
- y = z = NULL;
+ if (bld->static_texture_state->target == PIPE_TEXTURE_CUBE ||
+ bld->static_texture_state->target == PIPE_TEXTURE_1D_ARRAY ||
+ bld->static_texture_state->target == PIPE_TEXTURE_2D_ARRAY) {
+ z = r;
+ lp_build_name(z, "tex.z.layer");
}
/*
* Get texture colors.
*/
- lp_build_sample_texel_soa(bld, unit,
+ lp_build_sample_texel_soa(bld, sampler_unit,
width_vec, height_vec, depth_vec,
x, y, z,
row_stride_vec, img_stride_vec,
- data_ptr, colors_out);
+ data_ptr, mipoffsets, colors_out);
}
*/
static void
lp_build_sample_image_linear(struct lp_build_sample_context *bld,
- unsigned unit,
+ unsigned sampler_unit,
LLVMValueRef size,
LLVMValueRef row_stride_vec,
LLVMValueRef img_stride_vec,
LLVMValueRef data_ptr,
+ LLVMValueRef mipoffsets,
LLVMValueRef s,
LLVMValueRef t,
LLVMValueRef r,
LLVMValueRef flt_width_vec;
LLVMValueRef flt_height_vec;
LLVMValueRef flt_depth_vec;
- LLVMValueRef x0, y0, z0, x1, y1, z1;
- LLVMValueRef s_fpart, t_fpart, r_fpart;
+ LLVMValueRef x0, y0 = NULL, z0 = NULL, x1, y1 = NULL, z1 = NULL;
+ LLVMValueRef s_fpart, t_fpart = NULL, r_fpart = NULL;
LLVMValueRef neighbors[2][2][4];
int chan;
lp_build_extract_image_sizes(bld,
- bld->int_size_type,
+ &bld->int_size_bld,
bld->int_coord_type,
size,
&width_vec, &height_vec, &depth_vec);
flt_size = lp_build_int_to_float(&bld->float_size_bld, size);
lp_build_extract_image_sizes(bld,
- bld->float_size_type,
+ &bld->float_size_bld,
bld->coord_type,
flt_size,
&flt_width_vec, &flt_height_vec, &flt_depth_vec);
* Compute integer texcoords.
*/
lp_build_sample_wrap_linear(bld, s, width_vec, flt_width_vec,
- bld->static_state->pot_width,
- bld->static_state->wrap_s,
+ bld->static_texture_state->pot_width,
+ bld->static_sampler_state->wrap_s,
&x0, &x1, &s_fpart);
lp_build_name(x0, "tex.x0.wrapped");
lp_build_name(x1, "tex.x1.wrapped");
if (dims >= 2) {
lp_build_sample_wrap_linear(bld, t, height_vec, flt_height_vec,
- bld->static_state->pot_height,
- bld->static_state->wrap_t,
+ bld->static_texture_state->pot_height,
+ bld->static_sampler_state->wrap_t,
&y0, &y1, &t_fpart);
lp_build_name(y0, "tex.y0.wrapped");
lp_build_name(y1, "tex.y1.wrapped");
if (dims == 3) {
lp_build_sample_wrap_linear(bld, r, depth_vec, flt_depth_vec,
- bld->static_state->pot_depth,
- bld->static_state->wrap_r,
+ bld->static_texture_state->pot_depth,
+ bld->static_sampler_state->wrap_r,
&z0, &z1, &r_fpart);
lp_build_name(z0, "tex.z0.wrapped");
lp_build_name(z1, "tex.z1.wrapped");
}
- else if (bld->static_state->target == PIPE_TEXTURE_CUBE) {
- z0 = z1 = r; /* cube face */
- r_fpart = NULL;
- }
- else {
- z0 = z1 = NULL;
- r_fpart = NULL;
- }
}
- else {
- y0 = y1 = t_fpart = NULL;
- z0 = z1 = r_fpart = NULL;
+ if (bld->static_texture_state->target == PIPE_TEXTURE_CUBE ||
+ bld->static_texture_state->target == PIPE_TEXTURE_1D_ARRAY ||
+ bld->static_texture_state->target == PIPE_TEXTURE_2D_ARRAY) {
+ z0 = z1 = r; /* cube face or array layer */
+ lp_build_name(z0, "tex.z0.layer");
+ lp_build_name(z1, "tex.z1.layer");
}
+
/*
* Get texture colors.
*/
/* get x0/x1 texels */
- lp_build_sample_texel_soa(bld, unit,
+ lp_build_sample_texel_soa(bld, sampler_unit,
width_vec, height_vec, depth_vec,
x0, y0, z0,
row_stride_vec, img_stride_vec,
- data_ptr, neighbors[0][0]);
- lp_build_sample_texel_soa(bld, unit,
+ data_ptr, mipoffsets, neighbors[0][0]);
+ lp_build_sample_texel_soa(bld, sampler_unit,
width_vec, height_vec, depth_vec,
x1, y0, z0,
row_stride_vec, img_stride_vec,
- data_ptr, neighbors[0][1]);
+ data_ptr, mipoffsets, neighbors[0][1]);
if (dims == 1) {
/* Interpolate two samples from 1D image to produce one color */
LLVMValueRef colors0[4];
/* get x0/x1 texels at y1 */
- lp_build_sample_texel_soa(bld, unit,
+ lp_build_sample_texel_soa(bld, sampler_unit,
width_vec, height_vec, depth_vec,
x0, y1, z0,
row_stride_vec, img_stride_vec,
- data_ptr, neighbors[1][0]);
- lp_build_sample_texel_soa(bld, unit,
+ data_ptr, mipoffsets, neighbors[1][0]);
+ lp_build_sample_texel_soa(bld, sampler_unit,
width_vec, height_vec, depth_vec,
x1, y1, z0,
row_stride_vec, img_stride_vec,
- data_ptr, neighbors[1][1]);
+ data_ptr, mipoffsets, neighbors[1][1]);
/* Bilinear interpolate the four samples from the 2D image / 3D slice */
for (chan = 0; chan < 4; chan++) {
LLVMValueRef colors1[4];
/* get x0/x1/y0/y1 texels at z1 */
- lp_build_sample_texel_soa(bld, unit,
+ lp_build_sample_texel_soa(bld, sampler_unit,
width_vec, height_vec, depth_vec,
x0, y0, z1,
row_stride_vec, img_stride_vec,
- data_ptr, neighbors1[0][0]);
- lp_build_sample_texel_soa(bld, unit,
+ data_ptr, mipoffsets, neighbors1[0][0]);
+ lp_build_sample_texel_soa(bld, sampler_unit,
width_vec, height_vec, depth_vec,
x1, y0, z1,
row_stride_vec, img_stride_vec,
- data_ptr, neighbors1[0][1]);
- lp_build_sample_texel_soa(bld, unit,
+ data_ptr, mipoffsets, neighbors1[0][1]);
+ lp_build_sample_texel_soa(bld, sampler_unit,
width_vec, height_vec, depth_vec,
x0, y1, z1,
row_stride_vec, img_stride_vec,
- data_ptr, neighbors1[1][0]);
- lp_build_sample_texel_soa(bld, unit,
+ data_ptr, mipoffsets, neighbors1[1][0]);
+ lp_build_sample_texel_soa(bld, sampler_unit,
width_vec, height_vec, depth_vec,
x1, y1, z1,
row_stride_vec, img_stride_vec,
- data_ptr, neighbors1[1][1]);
+ data_ptr, mipoffsets, neighbors1[1][1]);
/* Bilinear interpolate the four samples from the second Z slice */
for (chan = 0; chan < 4; chan++) {
*/
static void
lp_build_sample_mipmap(struct lp_build_sample_context *bld,
- unsigned unit,
+ unsigned sampler_unit,
unsigned img_filter,
unsigned mip_filter,
LLVMValueRef s,
LLVMValueRef img_stride1_vec = NULL;
LLVMValueRef data_ptr0 = NULL;
LLVMValueRef data_ptr1 = NULL;
+ LLVMValueRef mipoff0 = NULL;
+ LLVMValueRef mipoff1 = NULL;
LLVMValueRef colors0[4], colors1[4];
unsigned chan;
lp_build_mipmap_level_sizes(bld, ilevel0,
&size0,
&row_stride0_vec, &img_stride0_vec);
- data_ptr0 = lp_build_get_mipmap_level(bld, ilevel0);
+ if (bld->num_lods == 1) {
+ data_ptr0 = lp_build_get_mipmap_level(bld, ilevel0);
+ }
+ else {
+ /* This path should work for num_lods 1 too but slightly less efficient */
+ data_ptr0 = bld->base_ptr;
+ mipoff0 = lp_build_get_mip_offsets(bld, ilevel0);
+ }
if (img_filter == PIPE_TEX_FILTER_NEAREST) {
- lp_build_sample_image_nearest(bld, unit,
+ lp_build_sample_image_nearest(bld, sampler_unit,
size0,
row_stride0_vec, img_stride0_vec,
- data_ptr0, s, t, r,
+ data_ptr0, mipoff0, s, t, r,
colors0);
}
else {
assert(img_filter == PIPE_TEX_FILTER_LINEAR);
- lp_build_sample_image_linear(bld, unit,
+ lp_build_sample_image_linear(bld, sampler_unit,
size0,
row_stride0_vec, img_stride0_vec,
- data_ptr0, s, t, r,
+ data_ptr0, mipoff0, s, t, r,
colors0);
}
lp_build_mipmap_level_sizes(bld, ilevel1,
&size1,
&row_stride1_vec, &img_stride1_vec);
- data_ptr1 = lp_build_get_mipmap_level(bld, ilevel1);
+ if (bld->num_lods == 1) {
+ data_ptr1 = lp_build_get_mipmap_level(bld, ilevel1);
+ }
+ else {
+ data_ptr1 = bld->base_ptr;
+ mipoff1 = lp_build_get_mip_offsets(bld, ilevel1);
+ }
if (img_filter == PIPE_TEX_FILTER_NEAREST) {
- lp_build_sample_image_nearest(bld, unit,
+ lp_build_sample_image_nearest(bld, sampler_unit,
size1,
row_stride1_vec, img_stride1_vec,
- data_ptr1, s, t, r,
+ data_ptr1, mipoff1, s, t, r,
colors1);
}
else {
- lp_build_sample_image_linear(bld, unit,
+ lp_build_sample_image_linear(bld, sampler_unit,
size1,
row_stride1_vec, img_stride1_vec,
- data_ptr1, s, t, r,
+ data_ptr1, mipoff1, s, t, r,
colors1);
}
}
}
+
+/**
+ * Clamp layer coord to valid values.
+ */
+static LLVMValueRef
+lp_build_layer_coord(struct lp_build_sample_context *bld,
+ unsigned texture_unit,
+ LLVMValueRef layer)
+{
+ LLVMValueRef maxlayer;
+
+ maxlayer = bld->dynamic_state->depth(bld->dynamic_state,
+ bld->gallivm, texture_unit);
+ maxlayer = lp_build_sub(&bld->int_bld, maxlayer, bld->int_bld.one);
+ maxlayer = lp_build_broadcast_scalar(&bld->int_coord_bld, maxlayer);
+ return lp_build_clamp(&bld->int_coord_bld, layer,
+ bld->int_coord_bld.zero, maxlayer);
+
+}
+
+
/**
* Calculate cube face, lod, mip levels.
*/
static void
lp_build_sample_common(struct lp_build_sample_context *bld,
- unsigned unit,
+ unsigned texture_index,
+ unsigned sampler_index,
LLVMValueRef *s,
LLVMValueRef *t,
LLVMValueRef *r,
LLVMValueRef *ilevel0,
LLVMValueRef *ilevel1)
{
- const unsigned mip_filter = bld->static_state->min_mip_filter;
- const unsigned min_filter = bld->static_state->min_img_filter;
- const unsigned mag_filter = bld->static_state->mag_img_filter;
+ const unsigned mip_filter = bld->static_sampler_state->min_mip_filter;
+ const unsigned min_filter = bld->static_sampler_state->min_img_filter;
+ const unsigned mag_filter = bld->static_sampler_state->mag_img_filter;
+ const unsigned target = bld->static_texture_state->target;
LLVMValueRef first_level;
struct lp_derivatives face_derivs;
/*
* Choose cube face, recompute texcoords and derivatives for the chosen face.
*/
- if (bld->static_state->target == PIPE_TEXTURE_CUBE) {
+ if (target == PIPE_TEXTURE_CUBE) {
LLVMValueRef face, face_s, face_t;
lp_build_cube_lookup(bld, *s, *t, *r, &face, &face_s, &face_t);
*s = face_s; /* vec */
face_derivs.ddx_ddy[1] = NULL;
derivs = &face_derivs;
}
+ else if (target == PIPE_TEXTURE_1D_ARRAY) {
+ *r = lp_build_iround(&bld->coord_bld, *t);
+ *r = lp_build_layer_coord(bld, texture_index, *r);
+ }
+ else if (target == PIPE_TEXTURE_2D_ARRAY) {
+ *r = lp_build_iround(&bld->coord_bld, *r);
+ *r = lp_build_layer_coord(bld, texture_index, *r);
+ }
/*
* Compute the level of detail (float).
/* Need to compute lod either to choose mipmap levels or to
* distinguish between minification/magnification with one mipmap level.
*/
- lp_build_lod_selector(bld, unit, derivs,
- lod_bias, explicit_lod,
+ lp_build_lod_selector(bld, texture_index, sampler_index,
+ derivs, lod_bias, explicit_lod,
mip_filter,
lod_ipart, lod_fpart);
} else {
/* fall-through */
case PIPE_TEX_MIPFILTER_NONE:
/* always use mip level 0 */
- if (bld->static_state->target == PIPE_TEXTURE_CUBE) {
+ if (target == PIPE_TEXTURE_CUBE) {
/* XXX this is a work-around for an apparent bug in LLVM 2.7.
* We should be able to set ilevel0 = const(0) but that causes
* bad x86 code to be emitted.
* XXX should probably disable that on other llvm versions.
*/
assert(*lod_ipart);
- lp_build_nearest_mip_level(bld, unit, *lod_ipart, ilevel0);
+ lp_build_nearest_mip_level(bld, texture_index, *lod_ipart, ilevel0);
}
else {
first_level = bld->dynamic_state->first_level(bld->dynamic_state,
- bld->gallivm, unit);
+ bld->gallivm, texture_index);
first_level = lp_build_broadcast_scalar(&bld->perquadi_bld, first_level);
*ilevel0 = first_level;
}
break;
case PIPE_TEX_MIPFILTER_NEAREST:
assert(*lod_ipart);
- lp_build_nearest_mip_level(bld, unit, *lod_ipart, ilevel0);
+ lp_build_nearest_mip_level(bld, texture_index, *lod_ipart, ilevel0);
break;
case PIPE_TEX_MIPFILTER_LINEAR:
assert(*lod_ipart);
assert(*lod_fpart);
- lp_build_linear_mip_levels(bld, unit,
+ lp_build_linear_mip_levels(bld, texture_index,
*lod_ipart, lod_fpart,
ilevel0, ilevel1);
break;
*/
static void
lp_build_sample_general(struct lp_build_sample_context *bld,
- unsigned unit,
+ unsigned sampler_unit,
LLVMValueRef s,
LLVMValueRef t,
LLVMValueRef r,
{
struct lp_build_context *int_bld = &bld->int_bld;
LLVMBuilderRef builder = bld->gallivm->builder;
- const unsigned mip_filter = bld->static_state->min_mip_filter;
- const unsigned min_filter = bld->static_state->min_img_filter;
- const unsigned mag_filter = bld->static_state->mag_img_filter;
+ const unsigned mip_filter = bld->static_sampler_state->min_mip_filter;
+ const unsigned min_filter = bld->static_sampler_state->min_img_filter;
+ const unsigned mag_filter = bld->static_sampler_state->mag_img_filter;
LLVMValueRef texels[4];
unsigned chan;
for (chan = 0; chan < 4; ++chan) {
texels[chan] = lp_build_alloca(bld->gallivm, bld->texel_bld.vec_type, "");
- lp_build_name(texels[chan], "sampler%u_texel_%c_var", unit, "xyzw"[chan]);
+ lp_build_name(texels[chan], "sampler%u_texel_%c_var", sampler_unit, "xyzw"[chan]);
}
if (min_filter == mag_filter) {
/* no need to distinguish between minification and magnification */
- lp_build_sample_mipmap(bld, unit,
+ lp_build_sample_mipmap(bld, sampler_unit,
min_filter, mip_filter,
s, t, r,
ilevel0, ilevel1, lod_fpart,
struct lp_build_if_state if_ctx;
LLVMValueRef minify;
+ /*
+ * XXX this should to all lods into account, if some are min
+ * some max probably could hack up the coords/weights in the linear
+ * path with selects to work for nearest.
+ * If that's just two quads sitting next to each other it seems
+ * quite ok to do the same filtering method on both though, at
+ * least unless we have explicit lod (and who uses different
+ * min/mag filter with that?)
+ */
+ if (bld->num_lods > 1)
+ lod_ipart = LLVMBuildExtractElement(builder, lod_ipart,
+ lp_build_const_int32(bld->gallivm, 0), "");
+
/* minify = lod >= 0.0 */
minify = LLVMBuildICmp(builder, LLVMIntSGE,
lod_ipart, int_bld->zero, "");
lp_build_if(&if_ctx, bld->gallivm, minify);
{
/* Use the minification filter */
- lp_build_sample_mipmap(bld, unit,
+ lp_build_sample_mipmap(bld, sampler_unit,
min_filter, mip_filter,
s, t, r,
ilevel0, ilevel1, lod_fpart,
lp_build_else(&if_ctx);
{
/* Use the magnification filter */
- lp_build_sample_mipmap(bld, unit,
+ lp_build_sample_mipmap(bld, sampler_unit,
mag_filter, PIPE_TEX_MIPFILTER_NONE,
s, t, r,
ilevel0, NULL, NULL,
for (chan = 0; chan < 4; ++chan) {
colors_out[chan] = LLVMBuildLoad(builder, texels[chan], "");
- lp_build_name(colors_out[chan], "sampler%u_texel_%c", unit, "xyzw"[chan]);
+ lp_build_name(colors_out[chan], "sampler%u_texel_%c", sampler_unit, "xyzw"[chan]);
+ }
+}
+
+
+/**
+ * Texel fetch function.
+ * In contrast to general sampling there is no filtering, no coord minification,
+ * lod (if any) is always explicit uint, coords are uints (in terms of texel units)
+ * directly to be applied to the selected mip level (after adding texel offsets).
+ * This function handles texel fetch for all targets where texel fetch is supported
+ * (no cube maps, but 1d, 2d, 3d are supported, arrays and buffers should be too).
+ */
+static void
+lp_build_fetch_texel(struct lp_build_sample_context *bld,
+ unsigned texture_unit,
+ const LLVMValueRef *coords,
+ LLVMValueRef explicit_lod,
+ const LLVMValueRef *offsets,
+ LLVMValueRef *colors_out)
+{
+ struct lp_build_context *perquadi_bld = &bld->perquadi_bld;
+ struct lp_build_context *int_coord_bld = &bld->int_coord_bld;
+ unsigned dims = bld->dims, chan;
+ unsigned target = bld->static_texture_state->target;
+ LLVMValueRef size, ilevel;
+ LLVMValueRef row_stride_vec = NULL, img_stride_vec = NULL;
+ LLVMValueRef x = coords[0], y = coords[1], z = coords[2];
+ LLVMValueRef width, height, depth, i, j;
+ LLVMValueRef offset, out_of_bounds, out1;
+
+ /* XXX just like ordinary sampling, we don't handle per-pixel lod (yet). */
+ if (explicit_lod && bld->static_texture_state->target != PIPE_BUFFER) {
+ ilevel = lp_build_pack_aos_scalars(bld->gallivm, int_coord_bld->type,
+ perquadi_bld->type, explicit_lod, 0);
+ lp_build_nearest_mip_level(bld, texture_unit, ilevel, &ilevel);
+ }
+ else {
+ bld->num_lods = 1;
+ ilevel = lp_build_const_int32(bld->gallivm, 0);
+ }
+ lp_build_mipmap_level_sizes(bld, ilevel,
+ &size,
+ &row_stride_vec, &img_stride_vec);
+ lp_build_extract_image_sizes(bld, &bld->int_size_bld, int_coord_bld->type,
+ size, &width, &height, &depth);
+
+ if (target == PIPE_TEXTURE_1D_ARRAY ||
+ target == PIPE_TEXTURE_2D_ARRAY) {
+ if (target == PIPE_TEXTURE_1D_ARRAY) {
+ z = lp_build_layer_coord(bld, texture_unit, y);
+ }
+ else {
+ z = lp_build_layer_coord(bld, texture_unit, z);
+ }
+ }
+
+ /* This is a lot like border sampling */
+ if (offsets[0]) {
+ /* XXX coords are really unsigned, offsets are signed */
+ x = lp_build_add(int_coord_bld, x, offsets[0]);
+ }
+ out_of_bounds = lp_build_cmp(int_coord_bld, PIPE_FUNC_LESS, x, int_coord_bld->zero);
+ out1 = lp_build_cmp(int_coord_bld, PIPE_FUNC_GEQUAL, x, width);
+ out_of_bounds = lp_build_or(int_coord_bld, out_of_bounds, out1);
+
+ if (dims >= 2) {
+ if (offsets[1]) {
+ y = lp_build_add(int_coord_bld, y, offsets[1]);
+ }
+ out1 = lp_build_cmp(int_coord_bld, PIPE_FUNC_LESS, y, int_coord_bld->zero);
+ out_of_bounds = lp_build_or(int_coord_bld, out_of_bounds, out1);
+ out1 = lp_build_cmp(int_coord_bld, PIPE_FUNC_GEQUAL, y, height);
+ out_of_bounds = lp_build_or(int_coord_bld, out_of_bounds, out1);
+
+ if (dims >= 3) {
+ if (offsets[2]) {
+ z = lp_build_add(int_coord_bld, z, offsets[2]);
+ }
+ out1 = lp_build_cmp(int_coord_bld, PIPE_FUNC_LESS, z, int_coord_bld->zero);
+ out_of_bounds = lp_build_or(int_coord_bld, out_of_bounds, out1);
+ out1 = lp_build_cmp(int_coord_bld, PIPE_FUNC_GEQUAL, z, depth);
+ out_of_bounds = lp_build_or(int_coord_bld, out_of_bounds, out1);
+ }
+ }
+
+ lp_build_sample_offset(int_coord_bld,
+ bld->format_desc,
+ x, y, z, row_stride_vec, img_stride_vec,
+ &offset, &i, &j);
+
+ if (bld->static_texture_state->target != PIPE_BUFFER) {
+ offset = lp_build_add(int_coord_bld, offset,
+ lp_build_get_mip_offsets(bld, ilevel));
+ }
+
+ offset = lp_build_andnot(int_coord_bld, offset, out_of_bounds);
+
+ lp_build_fetch_rgba_soa(bld->gallivm,
+ bld->format_desc,
+ bld->texel_type,
+ bld->base_ptr, offset,
+ i, j,
+ colors_out);
+
+ if (0) {
+ /*
+ * Not needed except for ARB_robust_buffer_access_behavior.
+ * Could use min/max above instead of out-of-bounds comparisons
+ * (in fact cast to unsigned and min only is sufficient)
+ * if we don't care about the result returned for out-of-bounds.
+ */
+ for (chan = 0; chan < 4; chan++) {
+ colors_out[chan] = lp_build_select(&bld->texel_bld, out_of_bounds,
+ bld->texel_bld.zero, colors_out[chan]);
+ }
}
}
/**
* Do shadow test/comparison.
- * \param p the texcoord Z (aka R, aka P) component
+ * \param coords incoming texcoords
* \param texel the texel to compare against (use the X channel)
+ * Ideally this should really be done per-sample.
*/
static void
lp_build_sample_compare(struct lp_build_sample_context *bld,
- LLVMValueRef p,
+ const LLVMValueRef *coords,
LLVMValueRef texel[4])
{
struct lp_build_context *texel_bld = &bld->texel_bld;
LLVMBuilderRef builder = bld->gallivm->builder;
- LLVMValueRef res;
+ LLVMValueRef res, p;
const unsigned chan = 0;
- if (bld->static_state->compare_mode == PIPE_TEX_COMPARE_NONE)
+ if (bld->static_sampler_state->compare_mode == PIPE_TEX_COMPARE_NONE)
return;
+ if (bld->static_texture_state->target == PIPE_TEXTURE_2D_ARRAY ||
+ bld->static_texture_state->target == PIPE_TEXTURE_CUBE) {
+ p = coords[3];
+ }
+ else {
+ p = coords[2];
+ }
+
/* debug code */
if (0) {
LLVMValueRef indx = lp_build_const_int32(bld->gallivm, 0);
bld->coord_bld.one);
/* result = (p FUNC texel) ? 1 : 0 */
- res = lp_build_cmp(texel_bld, bld->static_state->compare_func,
+ res = lp_build_cmp(texel_bld, bld->static_sampler_state->compare_func,
p, texel[chan]);
res = lp_build_select(texel_bld, res, texel_bld->one, texel_bld->zero);
void
lp_build_sample_nop(struct gallivm_state *gallivm,
struct lp_type type,
- unsigned num_coords,
const LLVMValueRef *coords,
LLVMValueRef texel_out[4])
{
* 'texel' will return a vector of four LLVMValueRefs corresponding to
* R, G, B, A.
* \param type vector float type to use for coords, etc.
+ * \param is_fetch if this is a texel fetch instruction.
* \param derivs partial derivatives of (s,t,r,q) with respect to x and y
*/
void
lp_build_sample_soa(struct gallivm_state *gallivm,
- const struct lp_sampler_static_state *static_state,
+ const struct lp_static_texture_state *static_texture_state,
+ const struct lp_static_sampler_state *static_sampler_state,
struct lp_sampler_dynamic_state *dynamic_state,
struct lp_type type,
- unsigned unit,
- unsigned num_coords,
+ boolean is_fetch,
+ unsigned texture_index,
+ unsigned sampler_index,
const LLVMValueRef *coords,
+ const LLVMValueRef *offsets,
const struct lp_derivatives *derivs,
LLVMValueRef lod_bias, /* optional */
LLVMValueRef explicit_lod, /* optional */
LLVMValueRef texel_out[4])
{
- unsigned dims = texture_dims(static_state->target);
+ unsigned dims = texture_dims(static_texture_state->target);
+ unsigned num_quads = type.length / 4;
+ unsigned mip_filter;
struct lp_build_sample_context bld;
+ struct lp_static_sampler_state derived_sampler_state = *static_sampler_state;
LLVMTypeRef i32t = LLVMInt32TypeInContext(gallivm->context);
LLVMBuilderRef builder = gallivm->builder;
LLVMValueRef tex_width, tex_height, tex_depth;
LLVMValueRef r;
if (0) {
- enum pipe_format fmt = static_state->format;
+ enum pipe_format fmt = static_texture_state->format;
debug_printf("Sample from %s\n", util_format_name(fmt));
}
/* Setup our build context */
memset(&bld, 0, sizeof bld);
bld.gallivm = gallivm;
- bld.static_state = static_state;
+ bld.static_sampler_state = &derived_sampler_state;
+ bld.static_texture_state = static_texture_state;
bld.dynamic_state = dynamic_state;
- bld.format_desc = util_format_description(static_state->format);
+ bld.format_desc = util_format_description(static_texture_state->format);
bld.dims = dims;
bld.vector_width = lp_type_width(type);
bld.int_type = lp_type_int(32);
bld.coord_type = type;
bld.int_coord_type = lp_int_type(type);
- bld.float_size_type = lp_type_float(32);
- bld.float_size_type.length = dims > 1 ? 4 : 1;
- bld.int_size_type = lp_int_type(bld.float_size_type);
+ bld.float_size_in_type = lp_type_float(32);
+ bld.float_size_in_type.length = dims > 1 ? 4 : 1;
+ bld.int_size_in_type = lp_int_type(bld.float_size_in_type);
bld.texel_type = type;
bld.perquadf_type = type;
/* we want native vector size to be able to use our intrinsics */
bld.perquadf_type.length = type.length > 4 ? ((type.length + 15) / 16) * 4 : 1;
bld.perquadi_type = lp_int_type(bld.perquadf_type);
+ /* always using the first channel hopefully should be safe,
+ * if not things WILL break in other places anyway.
+ */
+ if (bld.format_desc->colorspace == UTIL_FORMAT_COLORSPACE_RGB &&
+ bld.format_desc->channel[0].pure_integer) {
+ if (bld.format_desc->channel[0].type == UTIL_FORMAT_TYPE_SIGNED) {
+ bld.texel_type = lp_type_int_vec(type.width, type.width * type.length);
+ }
+ else if (bld.format_desc->channel[0].type == UTIL_FORMAT_TYPE_UNSIGNED) {
+ bld.texel_type = lp_type_uint_vec(type.width, type.width * type.length);
+ }
+ }
+
+ if (!static_texture_state->level_zero_only) {
+ derived_sampler_state.min_mip_filter = static_sampler_state->min_mip_filter;
+ } else {
+ derived_sampler_state.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
+ }
+ mip_filter = derived_sampler_state.min_mip_filter;
+
+ if (0) {
+ debug_printf(" .min_mip_filter = %u\n", derived_sampler_state.min_mip_filter);
+ }
+
+ /*
+ * There are other situations where at least the multiple int lods could be
+ * avoided like min and max lod being equal.
+ */
+ if ((is_fetch && explicit_lod && bld.static_texture_state->target != PIPE_BUFFER) ||
+ (!is_fetch && mip_filter != PIPE_TEX_MIPFILTER_NONE)) {
+ bld.num_lods = num_quads;
+ }
+ else {
+ bld.num_lods = 1;
+ }
+
+ bld.float_size_type = bld.float_size_in_type;
+ bld.float_size_type.length = bld.num_lods > 1 ? type.length :
+ bld.float_size_in_type.length;
+ bld.int_size_type = lp_int_type(bld.float_size_type);
+
lp_build_context_init(&bld.float_bld, gallivm, bld.float_type);
lp_build_context_init(&bld.float_vec_bld, gallivm, type);
lp_build_context_init(&bld.int_bld, gallivm, bld.int_type);
lp_build_context_init(&bld.coord_bld, gallivm, bld.coord_type);
lp_build_context_init(&bld.int_coord_bld, gallivm, bld.int_coord_type);
+ lp_build_context_init(&bld.int_size_in_bld, gallivm, bld.int_size_in_type);
+ lp_build_context_init(&bld.float_size_in_bld, gallivm, bld.float_size_in_type);
lp_build_context_init(&bld.int_size_bld, gallivm, bld.int_size_type);
lp_build_context_init(&bld.float_size_bld, gallivm, bld.float_size_type);
lp_build_context_init(&bld.texel_bld, gallivm, bld.texel_type);
lp_build_context_init(&bld.perquadi_bld, gallivm, bld.perquadi_type);
/* Get the dynamic state */
- tex_width = dynamic_state->width(dynamic_state, gallivm, unit);
- tex_height = dynamic_state->height(dynamic_state, gallivm, unit);
- tex_depth = dynamic_state->depth(dynamic_state, gallivm, unit);
- bld.row_stride_array = dynamic_state->row_stride(dynamic_state, gallivm, unit);
- bld.img_stride_array = dynamic_state->img_stride(dynamic_state, gallivm, unit);
- bld.data_array = dynamic_state->data_ptr(dynamic_state, gallivm, unit);
- /* Note that data_array is an array[level] of pointers to texture images */
+ tex_width = dynamic_state->width(dynamic_state, gallivm, texture_index);
+ tex_height = dynamic_state->height(dynamic_state, gallivm, texture_index);
+ tex_depth = dynamic_state->depth(dynamic_state, gallivm, texture_index);
+ bld.row_stride_array = dynamic_state->row_stride(dynamic_state, gallivm, texture_index);
+ bld.img_stride_array = dynamic_state->img_stride(dynamic_state, gallivm, texture_index);
+ bld.base_ptr = dynamic_state->base_ptr(dynamic_state, gallivm, texture_index);
+ bld.mip_offsets = dynamic_state->mip_offsets(dynamic_state, gallivm, texture_index);
+ /* Note that mip_offsets is an array[level] of offsets to texture images */
s = coords[0];
t = coords[1];
bld.int_size = tex_width;
}
else {
- bld.int_size = LLVMBuildInsertElement(builder, bld.int_size_bld.undef,
+ bld.int_size = LLVMBuildInsertElement(builder, bld.int_size_in_bld.undef,
tex_width, LLVMConstInt(i32t, 0, 0), "");
if (dims >= 2) {
bld.int_size = LLVMBuildInsertElement(builder, bld.int_size,
/* For debug: no-op texture sampling */
lp_build_sample_nop(gallivm,
bld.texel_type,
- num_coords,
coords,
texel_out);
}
+
+ else if (is_fetch) {
+ lp_build_fetch_texel(&bld, texture_index, coords,
+ explicit_lod, offsets,
+ texel_out);
+ }
+
else {
LLVMValueRef lod_ipart = NULL, lod_fpart = NULL;
LLVMValueRef ilevel0 = NULL, ilevel1 = NULL;
- unsigned num_quads = type.length / 4;
- const unsigned mip_filter = bld.static_state->min_mip_filter;
boolean use_aos = util_format_fits_8unorm(bld.format_desc) &&
- lp_is_simple_wrap_mode(static_state->wrap_s) &&
- lp_is_simple_wrap_mode(static_state->wrap_t);
+ lp_is_simple_wrap_mode(static_sampler_state->wrap_s) &&
+ lp_is_simple_wrap_mode(static_sampler_state->wrap_t);
if ((gallivm_debug & GALLIVM_DEBUG_PERF) &&
!use_aos && util_format_fits_8unorm(bld.format_desc)) {
debug_printf("%s: using floating point linear filtering for %s\n",
__FUNCTION__, bld.format_desc->short_name);
debug_printf(" min_img %d mag_img %d mip %d wraps %d wrapt %d\n",
- static_state->min_img_filter,
- static_state->mag_img_filter,
- static_state->min_mip_filter,
- static_state->wrap_s,
- static_state->wrap_t);
+ static_sampler_state->min_img_filter,
+ static_sampler_state->mag_img_filter,
+ static_sampler_state->min_mip_filter,
+ static_sampler_state->wrap_s,
+ static_sampler_state->wrap_t);
}
- lp_build_sample_common(&bld, unit,
+ lp_build_sample_common(&bld, texture_index, sampler_index,
&s, &t, &r,
derivs, lod_bias, explicit_lod,
&lod_ipart, &lod_fpart,
/*
* we only try 8-wide sampling with soa as it appears to
- * be a loss with aos with AVX.
+ * be a loss with aos with AVX (but it should work).
+ * (It should be faster if we'd support avx2)
*/
- if (num_quads == 1 || (mip_filter == PIPE_TEX_MIPFILTER_NONE &&
- !use_aos)) {
+ if (num_quads == 1 || !use_aos) {
if (num_quads > 1) {
- LLVMValueRef index0 = lp_build_const_int32(gallivm, 0);
- /* These parameters are the same for all quads */
- lod_ipart = LLVMBuildExtractElement(builder, lod_ipart, index0, "");
- ilevel0 = LLVMBuildExtractElement(builder, ilevel0, index0, "");
+ if (mip_filter == PIPE_TEX_MIPFILTER_NONE) {
+ LLVMValueRef index0 = lp_build_const_int32(gallivm, 0);
+ /*
+ * These parameters are the same for all quads,
+ * could probably simplify.
+ */
+ lod_ipart = LLVMBuildExtractElement(builder, lod_ipart, index0, "");
+ ilevel0 = LLVMBuildExtractElement(builder, ilevel0, index0, "");
+ }
}
if (use_aos) {
/* do sampling/filtering with fixed pt arithmetic */
- lp_build_sample_aos(&bld, unit,
+ lp_build_sample_aos(&bld, sampler_index,
s, t, r,
lod_ipart, lod_fpart,
ilevel0, ilevel1,
}
else {
- lp_build_sample_general(&bld, unit,
+ lp_build_sample_general(&bld, sampler_index,
s, t, r,
lod_ipart, lod_fpart,
ilevel0, ilevel1,
}
}
else {
- struct lp_build_if_state if_ctx;
- LLVMValueRef notsame_levels, notsame;
- LLVMValueRef index0 = lp_build_const_int32(gallivm, 0);
- LLVMValueRef texels[4];
- LLVMValueRef texelout[4];
unsigned j;
-
- texels[0] = lp_build_alloca(gallivm, bld.texel_bld.vec_type, "texr");
- texels[1] = lp_build_alloca(gallivm, bld.texel_bld.vec_type, "texg");
- texels[2] = lp_build_alloca(gallivm, bld.texel_bld.vec_type, "texb");
- texels[3] = lp_build_alloca(gallivm, bld.texel_bld.vec_type, "texa");
-
- /* only build the if if we MAY split, otherwise always split */
- if (!use_aos) {
- notsame = lp_build_extract_broadcast(gallivm,
- bld.perquadi_bld.type,
- bld.perquadi_bld.type,
- ilevel0, index0);
- notsame = lp_build_sub(&bld.perquadi_bld, ilevel0, notsame);
- notsame_levels = lp_build_any_true_range(&bld.perquadi_bld, num_quads,
- notsame);
- if (mip_filter == PIPE_TEX_MIPFILTER_LINEAR) {
- notsame = lp_build_extract_broadcast(gallivm,
- bld.perquadi_bld.type,
- bld.perquadi_bld.type,
- ilevel1, index0);
- notsame = lp_build_sub(&bld.perquadi_bld, ilevel1, notsame);
- notsame = lp_build_any_true_range(&bld.perquadi_bld, num_quads, notsame);
- notsame_levels = LLVMBuildOr(builder, notsame_levels, notsame, "");
- }
- lp_build_if(&if_ctx, gallivm, notsame_levels);
- }
-
- {
- struct lp_build_sample_context bld4;
- struct lp_type type4 = type;
- unsigned i;
- LLVMValueRef texelout4[4];
- LLVMValueRef texelouttmp[4][LP_MAX_VECTOR_LENGTH/16];
-
- type4.length = 4;
-
- /* Setup our build context */
- memset(&bld4, 0, sizeof bld4);
- bld4.gallivm = bld.gallivm;
- bld4.static_state = bld.static_state;
- bld4.dynamic_state = bld.dynamic_state;
- bld4.format_desc = bld.format_desc;
- bld4.dims = bld.dims;
- bld4.row_stride_array = bld.row_stride_array;
- bld4.img_stride_array = bld.img_stride_array;
- bld4.data_array = bld.data_array;
- bld4.int_size = bld.int_size;
-
- bld4.vector_width = lp_type_width(type4);
-
- bld4.float_type = lp_type_float(32);
- bld4.int_type = lp_type_int(32);
- bld4.coord_type = type4;
- bld4.int_coord_type = lp_int_type(type4);
- bld4.float_size_type = lp_type_float(32);
- bld4.float_size_type.length = dims > 1 ? 4 : 1;
- bld4.int_size_type = lp_int_type(bld4.float_size_type);
- bld4.texel_type = type4;
- bld4.perquadf_type = type4;
- /* we want native vector size to be able to use our intrinsics */
- bld4.perquadf_type.length = 1;
- bld4.perquadi_type = lp_int_type(bld4.perquadf_type);
-
- lp_build_context_init(&bld4.float_bld, gallivm, bld4.float_type);
- lp_build_context_init(&bld4.float_vec_bld, gallivm, type4);
- lp_build_context_init(&bld4.int_bld, gallivm, bld4.int_type);
- lp_build_context_init(&bld4.coord_bld, gallivm, bld4.coord_type);
- lp_build_context_init(&bld4.int_coord_bld, gallivm, bld4.int_coord_type);
- lp_build_context_init(&bld4.int_size_bld, gallivm, bld4.int_size_type);
- lp_build_context_init(&bld4.float_size_bld, gallivm, bld4.float_size_type);
- lp_build_context_init(&bld4.texel_bld, gallivm, bld4.texel_type);
- lp_build_context_init(&bld4.perquadf_bld, gallivm, bld4.perquadf_type);
- lp_build_context_init(&bld4.perquadi_bld, gallivm, bld4.perquadi_type);
-
- for (i = 0; i < num_quads; i++) {
- LLVMValueRef s4, t4, r4;
- LLVMValueRef lod_iparts, lod_fparts = NULL;
- LLVMValueRef ilevel0s, ilevel1s = NULL;
- LLVMValueRef indexi = lp_build_const_int32(gallivm, i);
-
- s4 = lp_build_extract_range(gallivm, s, 4*i, 4);
- t4 = lp_build_extract_range(gallivm, t, 4*i, 4);
- r4 = lp_build_extract_range(gallivm, r, 4*i, 4);
- lod_iparts = LLVMBuildExtractElement(builder, lod_ipart, indexi, "");
- ilevel0s = LLVMBuildExtractElement(builder, ilevel0, indexi, "");
- if (mip_filter == PIPE_TEX_MIPFILTER_LINEAR) {
- ilevel1s = LLVMBuildExtractElement(builder, ilevel1, indexi, "");
- lod_fparts = LLVMBuildExtractElement(builder, lod_fpart, indexi, "");
- }
-
- if (use_aos) {
- /* do sampling/filtering with fixed pt arithmetic */
- lp_build_sample_aos(&bld4, unit,
- s4, t4, r4,
- lod_iparts, lod_fparts,
- ilevel0s, ilevel1s,
- texelout4);
- }
-
- else {
- lp_build_sample_general(&bld4, unit,
- s4, t4, r4,
- lod_iparts, lod_fparts,
- ilevel0s, ilevel1s,
- texelout4);
- }
- for (j = 0; j < 4; j++) {
- texelouttmp[j][i] = texelout4[j];
- }
- }
- for (j = 0; j < 4; j++) {
- texelout[j] = lp_build_concat(gallivm, texelouttmp[j], type4, num_quads);
- LLVMBuildStore(builder, texelout[j], texels[j]);
- }
- }
- if (!use_aos) {
- LLVMValueRef ilevel0s, lod_iparts, ilevel1s = NULL;
-
- lp_build_else(&if_ctx);
-
- /* These parameters are the same for all quads */
- lod_iparts = LLVMBuildExtractElement(builder, lod_ipart, index0, "");
- ilevel0s = LLVMBuildExtractElement(builder, ilevel0, index0, "");
+ struct lp_build_sample_context bld4;
+ struct lp_type type4 = type;
+ unsigned i;
+ LLVMValueRef texelout4[4];
+ LLVMValueRef texelouttmp[4][LP_MAX_VECTOR_LENGTH/16];
+
+ type4.length = 4;
+
+ /* Setup our build context */
+ memset(&bld4, 0, sizeof bld4);
+ bld4.gallivm = bld.gallivm;
+ bld4.static_texture_state = bld.static_texture_state;
+ bld4.static_sampler_state = bld.static_sampler_state;
+ bld4.dynamic_state = bld.dynamic_state;
+ bld4.format_desc = bld.format_desc;
+ bld4.dims = bld.dims;
+ bld4.row_stride_array = bld.row_stride_array;
+ bld4.img_stride_array = bld.img_stride_array;
+ bld4.base_ptr = bld.base_ptr;
+ bld4.mip_offsets = bld.mip_offsets;
+ bld4.int_size = bld.int_size;
+
+ bld4.vector_width = lp_type_width(type4);
+
+ bld4.float_type = lp_type_float(32);
+ bld4.int_type = lp_type_int(32);
+ bld4.coord_type = type4;
+ bld4.int_coord_type = lp_int_type(type4);
+ bld4.float_size_in_type = lp_type_float(32);
+ bld4.float_size_in_type.length = dims > 1 ? 4 : 1;
+ bld4.int_size_in_type = lp_int_type(bld4.float_size_in_type);
+ bld4.texel_type = bld.texel_type;
+ bld4.texel_type.length = 4;
+ bld4.perquadf_type = type4;
+ /* we want native vector size to be able to use our intrinsics */
+ bld4.perquadf_type.length = 1;
+ bld4.perquadi_type = lp_int_type(bld4.perquadf_type);
+
+ bld4.num_lods = 1;
+ bld4.int_size_type = bld4.int_size_in_type;
+ bld4.float_size_type = bld4.float_size_in_type;
+
+ lp_build_context_init(&bld4.float_bld, gallivm, bld4.float_type);
+ lp_build_context_init(&bld4.float_vec_bld, gallivm, type4);
+ lp_build_context_init(&bld4.int_bld, gallivm, bld4.int_type);
+ lp_build_context_init(&bld4.coord_bld, gallivm, bld4.coord_type);
+ lp_build_context_init(&bld4.int_coord_bld, gallivm, bld4.int_coord_type);
+ lp_build_context_init(&bld4.int_size_in_bld, gallivm, bld4.int_size_in_type);
+ lp_build_context_init(&bld4.float_size_in_bld, gallivm, bld4.float_size_in_type);
+ lp_build_context_init(&bld4.int_size_bld, gallivm, bld4.int_size_type);
+ lp_build_context_init(&bld4.float_size_bld, gallivm, bld4.float_size_type);
+ lp_build_context_init(&bld4.texel_bld, gallivm, bld4.texel_type);
+ lp_build_context_init(&bld4.perquadf_bld, gallivm, bld4.perquadf_type);
+ lp_build_context_init(&bld4.perquadi_bld, gallivm, bld4.perquadi_type);
+
+ for (i = 0; i < num_quads; i++) {
+ LLVMValueRef s4, t4, r4;
+ LLVMValueRef lod_iparts, lod_fparts = NULL;
+ LLVMValueRef ilevel0s, ilevel1s = NULL;
+ LLVMValueRef indexi = lp_build_const_int32(gallivm, i);
+
+ s4 = lp_build_extract_range(gallivm, s, 4*i, 4);
+ t4 = lp_build_extract_range(gallivm, t, 4*i, 4);
+ r4 = lp_build_extract_range(gallivm, r, 4*i, 4);
+ lod_iparts = LLVMBuildExtractElement(builder, lod_ipart, indexi, "");
+ ilevel0s = LLVMBuildExtractElement(builder, ilevel0, indexi, "");
if (mip_filter == PIPE_TEX_MIPFILTER_LINEAR) {
- ilevel1s = LLVMBuildExtractElement(builder, ilevel1, index0, "");
+ ilevel1s = LLVMBuildExtractElement(builder, ilevel1, indexi, "");
+ lod_fparts = LLVMBuildExtractElement(builder, lod_fpart, indexi, "");
}
if (use_aos) {
/* do sampling/filtering with fixed pt arithmetic */
- lp_build_sample_aos(&bld, unit,
- s, t, r,
- lod_iparts, lod_fpart,
+ lp_build_sample_aos(&bld4, sampler_index,
+ s4, t4, r4,
+ lod_iparts, lod_fparts,
ilevel0s, ilevel1s,
- texelout);
+ texelout4);
}
else {
- lp_build_sample_general(&bld, unit,
- s, t, r,
- lod_iparts, lod_fpart,
+ lp_build_sample_general(&bld4, sampler_index,
+ s4, t4, r4,
+ lod_iparts, lod_fparts,
ilevel0s, ilevel1s,
- texelout);
+ texelout4);
}
for (j = 0; j < 4; j++) {
- LLVMBuildStore(builder, texelout[j], texels[j]);
+ texelouttmp[j][i] = texelout4[j];
}
-
- lp_build_endif(&if_ctx);
}
for (j = 0; j < 4; j++) {
- texel_out[j] = LLVMBuildLoad(builder, texels[j], "");
+ texel_out[j] = lp_build_concat(gallivm, texelouttmp[j], type4, num_quads);
}
}
+
+ lp_build_sample_compare(&bld, coords, texel_out);
}
- lp_build_sample_compare(&bld, r, texel_out);
+ if (static_texture_state->target != PIPE_BUFFER) {
+ apply_sampler_swizzle(&bld, texel_out);
+ }
- apply_sampler_swizzle(&bld, texel_out);
+ /*
+ * texel type can be a (32bit) int/uint (for pure int formats only),
+ * however we are expected to always return floats (storage is untyped).
+ */
+ if (!bld.texel_type.floating) {
+ unsigned chan;
+ for (chan = 0; chan < 4; chan++) {
+ texel_out[chan] = LLVMBuildBitCast(builder, texel_out[chan],
+ lp_build_vec_type(gallivm, type), "");
+ }
+ }
}
void
lp_build_size_query_soa(struct gallivm_state *gallivm,
- const struct lp_sampler_static_state *static_state,
+ const struct lp_static_texture_state *static_state,
struct lp_sampler_dynamic_state *dynamic_state,
struct lp_type int_type,
- unsigned unit,
+ unsigned texture_unit,
+ boolean need_nr_mips,
LLVMValueRef explicit_lod,
LLVMValueRef *sizes_out)
{
LLVMValueRef lod;
LLVMValueRef size;
+ LLVMValueRef first_level = NULL;
int dims, i;
+ boolean has_array;
struct lp_build_context bld_int_vec;
+ dims = texture_dims(static_state->target);
+
switch (static_state->target) {
- case PIPE_TEXTURE_1D:
- case PIPE_BUFFER:
- dims = 1;
- break;
- case PIPE_TEXTURE_2D:
- case PIPE_TEXTURE_CUBE:
- case PIPE_TEXTURE_RECT:
- dims = 2;
- break;
- case PIPE_TEXTURE_3D:
- dims = 3;
+ case PIPE_TEXTURE_1D_ARRAY:
+ case PIPE_TEXTURE_2D_ARRAY:
+ has_array = TRUE;
break;
-
default:
- assert(0);
- return;
+ has_array = FALSE;
+ break;
}
assert(!int_type.floating);
lp_build_context_init(&bld_int_vec, gallivm, lp_type_int_vec(32, 128));
if (explicit_lod) {
- LLVMValueRef first_level;
lod = LLVMBuildExtractElement(gallivm->builder, explicit_lod, lp_build_const_int32(gallivm, 0), "");
- first_level = dynamic_state->first_level(dynamic_state, gallivm, unit);
+ first_level = dynamic_state->first_level(dynamic_state, gallivm, texture_unit);
lod = lp_build_broadcast_scalar(&bld_int_vec,
LLVMBuildAdd(gallivm->builder, lod, first_level, "lod"));
lod = bld_int_vec.zero;
}
- size = bld_int_vec.undef;
+ if (need_nr_mips) {
+ size = bld_int_vec.zero;
+ }
+ else {
+ size = bld_int_vec.undef;
+ }
size = LLVMBuildInsertElement(gallivm->builder, size,
- dynamic_state->width(dynamic_state, gallivm, unit),
+ dynamic_state->width(dynamic_state, gallivm, texture_unit),
lp_build_const_int32(gallivm, 0), "");
if (dims >= 2) {
size = LLVMBuildInsertElement(gallivm->builder, size,
- dynamic_state->height(dynamic_state, gallivm, unit),
+ dynamic_state->height(dynamic_state, gallivm, texture_unit),
lp_build_const_int32(gallivm, 1), "");
}
if (dims >= 3) {
size = LLVMBuildInsertElement(gallivm->builder, size,
- dynamic_state->depth(dynamic_state, gallivm, unit),
+ dynamic_state->depth(dynamic_state, gallivm, texture_unit),
lp_build_const_int32(gallivm, 2), "");
}
size = lp_build_minify(&bld_int_vec, size, lod);
- for (i=0; i < dims; i++) {
+ if (has_array)
+ size = LLVMBuildInsertElement(gallivm->builder, size,
+ dynamic_state->depth(dynamic_state, gallivm, texture_unit),
+ lp_build_const_int32(gallivm, dims), "");
+
+ /*
+ * XXX for out-of-bounds lod, should set size to zero vector here
+ * (for dx10-style only, i.e. need_nr_mips)
+ */
+
+ for (i = 0; i < dims + (has_array ? 1 : 0); i++) {
sizes_out[i] = lp_build_extract_broadcast(gallivm, bld_int_vec.type, int_type,
size,
lp_build_const_int32(gallivm, i));
}
+
+ /*
+ * if there's no explicit_lod (buffers, rects) queries requiring nr of
+ * mips would be illegal.
+ */
+ if (need_nr_mips && explicit_lod) {
+ struct lp_build_context bld_int_scalar;
+ LLVMValueRef num_levels;
+ lp_build_context_init(&bld_int_scalar, gallivm, lp_type_int(32));
+
+ if (static_state->level_zero_only) {
+ num_levels = bld_int_scalar.one;
+ }
+ else {
+ LLVMValueRef last_level;
+
+ last_level = dynamic_state->last_level(dynamic_state, gallivm, texture_unit);
+ num_levels = lp_build_sub(&bld_int_scalar, last_level, first_level);
+ num_levels = lp_build_add(&bld_int_scalar, num_levels, bld_int_scalar.one);
+ }
+ sizes_out[3] = lp_build_broadcast(gallivm, lp_build_vec_type(gallivm, int_type),
+ num_levels);
+ }
}
projects / mesa.git / blobdiff