#include "util/u_memory.h"
#include "util/u_math.h"
#include "util/u_format.h"
-#include "util/u_cpu_detect.h"
#include "lp_bld_debug.h"
#include "lp_bld_type.h"
#include "lp_bld_const.h"
#include "lp_bld_conv.h"
#include "lp_bld_arit.h"
+#include "lp_bld_bitarit.h"
#include "lp_bld_logic.h"
+#include "lp_bld_printf.h"
#include "lp_bld_swizzle.h"
-#include "lp_bld_pack.h"
#include "lp_bld_flow.h"
#include "lp_bld_gather.h"
#include "lp_bld_format.h"
#include "lp_bld_sample.h"
-
-
-/**
- * Keep all information for sampling code generation in a single place.
- */
-struct lp_build_sample_context
-{
- LLVMBuilderRef builder;
-
- const struct lp_sampler_static_state *static_state;
-
- struct lp_sampler_dynamic_state *dynamic_state;
-
- const struct util_format_description *format_desc;
-
- /** regular scalar float type */
- struct lp_type float_type;
- struct lp_build_context float_bld;
-
- /** regular scalar float type */
- struct lp_type int_type;
- struct lp_build_context int_bld;
-
- /** Incoming coordinates type and build context */
- struct lp_type coord_type;
- struct lp_build_context coord_bld;
-
- /** Unsigned integer coordinates */
- struct lp_type uint_coord_type;
- struct lp_build_context uint_coord_bld;
-
- /** Signed integer coordinates */
- struct lp_type int_coord_type;
- struct lp_build_context int_coord_bld;
-
- /** Output texels type and build context */
- struct lp_type texel_type;
- struct lp_build_context texel_bld;
-};
-
-
-/**
- * Does the given texture wrap mode allow sampling the texture border color?
- * XXX maybe move this into gallium util code.
- */
-static boolean
-wrap_mode_uses_border_color(unsigned mode)
-{
- switch (mode) {
- case PIPE_TEX_WRAP_REPEAT:
- case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
- case PIPE_TEX_WRAP_MIRROR_REPEAT:
- case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
- return FALSE;
- case PIPE_TEX_WRAP_CLAMP:
- case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
- case PIPE_TEX_WRAP_MIRROR_CLAMP:
- case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
- return TRUE;
- default:
- assert(0 && "unexpected wrap mode");
- return FALSE;
- }
-}
-
-
-static LLVMValueRef
-lp_build_get_mipmap_level(struct lp_build_sample_context *bld,
- LLVMValueRef data_array, LLVMValueRef level)
-{
- LLVMValueRef indexes[2], data_ptr;
- indexes[0] = LLVMConstInt(LLVMInt32Type(), 0, 0);
- indexes[1] = level;
- data_ptr = LLVMBuildGEP(bld->builder, data_array, indexes, 2, "");
- data_ptr = LLVMBuildLoad(bld->builder, data_ptr, "");
- return data_ptr;
-}
-
-
-static LLVMValueRef
-lp_build_get_const_mipmap_level(struct lp_build_sample_context *bld,
- LLVMValueRef data_array, int level)
-{
- LLVMValueRef lvl = LLVMConstInt(LLVMInt32Type(), level, 0);
- return lp_build_get_mipmap_level(bld, data_array, lvl);
-}
-
-
-/**
- * Dereference stride_array[mipmap_level] array to get a stride.
- * Return stride as a vector.
- */
-static LLVMValueRef
-lp_build_get_level_stride_vec(struct lp_build_sample_context *bld,
- LLVMValueRef stride_array, LLVMValueRef level)
-{
- LLVMValueRef indexes[2], stride;
- indexes[0] = LLVMConstInt(LLVMInt32Type(), 0, 0);
- indexes[1] = level;
- stride = LLVMBuildGEP(bld->builder, stride_array, indexes, 2, "");
- stride = LLVMBuildLoad(bld->builder, stride, "");
- stride = lp_build_broadcast_scalar(&bld->int_coord_bld, stride);
- return stride;
-}
-
-
-/** Dereference stride_array[0] array to get a stride (as vector). */
-static LLVMValueRef
-lp_build_get_const_level_stride_vec(struct lp_build_sample_context *bld,
- LLVMValueRef stride_array, int level)
-{
- LLVMValueRef lvl = LLVMConstInt(LLVMInt32Type(), level, 0);
- return lp_build_get_level_stride_vec(bld, stride_array, lvl);
-}
-
-
-static int
-texture_dims(enum pipe_texture_target tex)
-{
- switch (tex) {
- case PIPE_TEXTURE_1D:
- return 1;
- case PIPE_TEXTURE_2D:
- case PIPE_TEXTURE_CUBE:
- return 2;
- case PIPE_TEXTURE_3D:
- return 3;
- default:
- assert(0 && "bad texture target in texture_dims()");
- return 2;
- }
-}
-
-
-static void
-apply_sampler_swizzle(struct lp_build_sample_context *bld,
- LLVMValueRef *texel)
-{
- unsigned char swizzles[4];
-
- swizzles[0] = bld->static_state->swizzle_r;
- swizzles[1] = bld->static_state->swizzle_g;
- swizzles[2] = bld->static_state->swizzle_b;
- swizzles[3] = bld->static_state->swizzle_a;
-
- lp_build_swizzle_soa_inplace(&bld->texel_bld, texel, swizzles);
-}
-
+#include "lp_bld_sample_aos.h"
+#include "lp_bld_struct.h"
+#include "lp_bld_quad.h"
/**
* Generate code to fetch a texel from a texture at int coords (x, y, z).
* The computation depends on whether the texture is 1D, 2D or 3D.
- * The result, texel, will be:
+ * The result, texel, will be float vectors:
* texel[0] = red values
* texel[1] = green values
* texel[2] = blue values
*/
static void
lp_build_sample_texel_soa(struct lp_build_sample_context *bld,
+ unsigned unit,
LLVMValueRef width,
LLVMValueRef height,
LLVMValueRef depth,
LLVMValueRef data_ptr,
LLVMValueRef texel_out[4])
{
- const int dims = texture_dims(bld->static_state->target);
+ const struct lp_sampler_static_state *static_state = bld->static_state;
+ const unsigned dims = bld->dims;
struct lp_build_context *int_coord_bld = &bld->int_coord_bld;
LLVMValueRef offset;
LLVMValueRef i, j;
LLVMValueRef use_border = NULL;
/* use_border = x < 0 || x >= width || y < 0 || y >= height */
- if (wrap_mode_uses_border_color(bld->static_state->wrap_s)) {
+ if (lp_sampler_wrap_mode_uses_border_color(static_state->wrap_s,
+ static_state->min_img_filter,
+ static_state->mag_img_filter)) {
LLVMValueRef b1, b2;
b1 = lp_build_cmp(int_coord_bld, PIPE_FUNC_LESS, x, int_coord_bld->zero);
b2 = lp_build_cmp(int_coord_bld, PIPE_FUNC_GEQUAL, x, width);
use_border = LLVMBuildOr(bld->builder, b1, b2, "b1_or_b2");
}
- if (dims >= 2 && wrap_mode_uses_border_color(bld->static_state->wrap_t)) {
+ if (dims >= 2 &&
+ lp_sampler_wrap_mode_uses_border_color(static_state->wrap_t,
+ static_state->min_img_filter,
+ static_state->mag_img_filter)) {
LLVMValueRef b1, b2;
b1 = lp_build_cmp(int_coord_bld, PIPE_FUNC_LESS, y, int_coord_bld->zero);
b2 = lp_build_cmp(int_coord_bld, PIPE_FUNC_GEQUAL, y, height);
}
}
- if (dims == 3 && wrap_mode_uses_border_color(bld->static_state->wrap_r)) {
+ if (dims == 3 &&
+ lp_sampler_wrap_mode_uses_border_color(static_state->wrap_r,
+ static_state->min_img_filter,
+ static_state->mag_img_filter)) {
LLVMValueRef b1, b2;
b1 = lp_build_cmp(int_coord_bld, PIPE_FUNC_LESS, z, int_coord_bld->zero);
b2 = lp_build_cmp(int_coord_bld, PIPE_FUNC_GEQUAL, z, depth);
}
/* convert x,y,z coords to linear offset from start of texture, in bytes */
- lp_build_sample_offset(&bld->uint_coord_bld,
+ lp_build_sample_offset(&bld->int_coord_bld,
bld->format_desc,
x, y, z, y_stride, z_stride,
&offset, &i, &j);
* coords which are out of bounds to become zero. Zero's guaranteed
* to be inside the texture image.
*/
- offset = lp_build_andc(&bld->uint_coord_bld, offset, use_border);
+ offset = lp_build_andnot(&bld->int_coord_bld, offset, use_border);
}
lp_build_fetch_rgba_soa(bld->builder,
i, j,
texel_out);
- apply_sampler_swizzle(bld, texel_out);
-
/*
* Note: if we find an app which frequently samples the texture border
* we might want to implement a true conditional here to avoid sampling
if (use_border) {
/* select texel color or border color depending on use_border */
+ LLVMValueRef border_color_ptr =
+ bld->dynamic_state->border_color(bld->dynamic_state,
+ bld->builder, unit);
int chan;
for (chan = 0; chan < 4; chan++) {
LLVMValueRef border_chan =
- lp_build_const_vec(bld->texel_type,
- bld->static_state->border_color[chan]);
+ lp_build_array_get(bld->builder, border_color_ptr,
+ lp_build_const_int32(chan));
+ LLVMValueRef border_chan_vec =
+ lp_build_broadcast_scalar(&bld->float_vec_bld, border_chan);
texel_out[chan] = lp_build_select(&bld->texel_bld, use_border,
- border_chan, texel_out[chan]);
+ border_chan_vec, texel_out[chan]);
}
}
-}
-
-
-/**
- * Fetch the texels as <4n x i8> in AoS form.
- */
-static LLVMValueRef
-lp_build_sample_packed(struct lp_build_sample_context *bld,
- LLVMValueRef x,
- LLVMValueRef y,
- LLVMValueRef y_stride,
- LLVMValueRef data_array)
-{
- LLVMValueRef offset, i, j;
- LLVMValueRef data_ptr;
- LLVMValueRef res;
-
- /* convert x,y,z coords to linear offset from start of texture, in bytes */
- lp_build_sample_offset(&bld->uint_coord_bld,
- bld->format_desc,
- x, y, NULL, y_stride, NULL,
- &offset, &i, &j);
-
- /* get pointer to mipmap level 0 data */
- data_ptr = lp_build_get_const_mipmap_level(bld, data_array, 0);
-
- if (util_format_is_rgba8_variant(bld->format_desc)) {
- /* Just fetch the data directly without swizzling */
- assert(bld->format_desc->block.width == 1);
- assert(bld->format_desc->block.height == 1);
- assert(bld->format_desc->block.bits <= bld->texel_type.width);
-
- res = lp_build_gather(bld->builder,
- bld->texel_type.length,
- bld->format_desc->block.bits,
- bld->texel_type.width,
- data_ptr, offset);
- }
- else {
- struct lp_type type;
-
- assert(bld->texel_type.width == 32);
-
- memset(&type, 0, sizeof type);
- type.width = 8;
- type.length = bld->texel_type.length*4;
- type.norm = TRUE;
-
- res = lp_build_fetch_rgba_aos(bld->builder, bld->format_desc, type,
- data_ptr, offset, i, j);
- }
- return res;
+ apply_sampler_swizzle(bld, texel_out);
}
struct lp_build_context *int_coord_bld = &bld->int_coord_bld;
LLVMValueRef fract, flr, isOdd;
- /* fract = coord - floor(coord) */
- fract = lp_build_sub(coord_bld, coord, lp_build_floor(coord_bld, coord));
-
- /* flr = ifloor(coord); */
- flr = lp_build_ifloor(coord_bld, coord);
+ lp_build_ifloor_fract(coord_bld, coord, &flr, &fract);
/* isOdd = flr & 1 */
isOdd = LLVMBuildAnd(bld->builder, flr, int_coord_bld->one, "");
}
-/**
- * We only support a few wrap modes in lp_build_sample_wrap_int() at this time.
- * Return whether the given mode is supported by that function.
- */
-static boolean
-is_simple_wrap_mode(unsigned mode)
-{
- switch (mode) {
- case PIPE_TEX_WRAP_REPEAT:
- case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
- return TRUE;
- default:
- return FALSE;
- }
-}
-
-
-/**
- * Build LLVM code for texture wrap mode, for scaled integer texcoords.
- * \param coord the incoming texcoord (s,t,r or q) scaled to the texture size
- * \param length the texture size along one dimension
- * \param is_pot if TRUE, length is a power of two
- * \param wrap_mode one of PIPE_TEX_WRAP_x
- */
-static LLVMValueRef
-lp_build_sample_wrap_int(struct lp_build_sample_context *bld,
- LLVMValueRef coord,
- LLVMValueRef length,
- boolean is_pot,
- unsigned wrap_mode)
-{
- struct lp_build_context *uint_coord_bld = &bld->uint_coord_bld;
- struct lp_build_context *int_coord_bld = &bld->int_coord_bld;
- LLVMValueRef length_minus_one;
-
- length_minus_one = lp_build_sub(uint_coord_bld, length, uint_coord_bld->one);
-
- switch(wrap_mode) {
- case PIPE_TEX_WRAP_REPEAT:
- if(is_pot)
- coord = LLVMBuildAnd(bld->builder, coord, length_minus_one, "");
- else
- /* Signed remainder won't give the right results for negative
- * dividends but unsigned remainder does.*/
- coord = LLVMBuildURem(bld->builder, coord, length, "");
- break;
-
- case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
- coord = lp_build_max(int_coord_bld, coord, int_coord_bld->zero);
- coord = lp_build_min(int_coord_bld, coord, length_minus_one);
- break;
-
- case PIPE_TEX_WRAP_CLAMP:
- case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
- case PIPE_TEX_WRAP_MIRROR_REPEAT:
- case PIPE_TEX_WRAP_MIRROR_CLAMP:
- case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
- case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
- default:
- assert(0);
- }
-
- return coord;
-}
-
-
/**
* Build LLVM code for texture wrap mode for linear filtering.
* \param x0_out returns first integer texcoord
lp_build_sample_wrap_linear(struct lp_build_sample_context *bld,
LLVMValueRef coord,
LLVMValueRef length,
+ LLVMValueRef length_f,
boolean is_pot,
unsigned wrap_mode,
LLVMValueRef *x0_out,
{
struct lp_build_context *coord_bld = &bld->coord_bld;
struct lp_build_context *int_coord_bld = &bld->int_coord_bld;
- struct lp_build_context *uint_coord_bld = &bld->uint_coord_bld;
LLVMValueRef half = lp_build_const_vec(coord_bld->type, 0.5);
- LLVMValueRef length_f = lp_build_int_to_float(coord_bld, length);
- LLVMValueRef length_minus_one = lp_build_sub(uint_coord_bld, length, uint_coord_bld->one);
+ LLVMValueRef length_minus_one = lp_build_sub(int_coord_bld, length, int_coord_bld->one);
LLVMValueRef coord0, coord1, weight;
switch(wrap_mode) {
/* mul by size and subtract 0.5 */
coord = lp_build_mul(coord_bld, coord, length_f);
coord = lp_build_sub(coord_bld, coord, half);
- /* convert to int */
- coord0 = lp_build_ifloor(coord_bld, coord);
- coord1 = lp_build_add(uint_coord_bld, coord0, uint_coord_bld->one);
- /* compute lerp weight */
- weight = lp_build_fract(coord_bld, coord);
+ /* convert to int, compute lerp weight */
+ lp_build_ifloor_fract(coord_bld, coord, &coord0, &weight);
/* repeat wrap */
if (is_pot) {
+ coord1 = lp_build_add(int_coord_bld, coord0, int_coord_bld->one);
coord0 = LLVMBuildAnd(bld->builder, coord0, length_minus_one, "");
coord1 = LLVMBuildAnd(bld->builder, coord1, length_minus_one, "");
}
else {
- /* Signed remainder won't give the right results for negative
- * dividends but unsigned remainder does.*/
+ /* Add a bias to the texcoord to handle negative coords */
+ LLVMValueRef bias = lp_build_mul_imm(int_coord_bld, length, 1024);
+ LLVMValueRef mask;
+ coord0 = LLVMBuildAdd(bld->builder, coord0, bias, "");
coord0 = LLVMBuildURem(bld->builder, coord0, length, "");
- coord1 = LLVMBuildURem(bld->builder, coord1, length, "");
+ mask = lp_build_compare(bld->builder, int_coord_bld->type,
+ PIPE_FUNC_NOTEQUAL, coord0, length_minus_one);
+ coord1 = LLVMBuildAnd(bld->builder,
+ lp_build_add(int_coord_bld, coord0, int_coord_bld->one),
+ mask, "");
}
break;
coord = lp_build_sub(coord_bld, coord, half);
- weight = lp_build_fract(coord_bld, coord);
- coord0 = lp_build_ifloor(coord_bld, coord);
+ /* 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_CLAMP_TO_EDGE:
- if (bld->static_state->normalized_coords) {
- /* clamp to [0,1] */
- coord = lp_build_clamp(coord_bld, coord, coord_bld->zero, coord_bld->one);
- /* mul by tex size and subtract 0.5 */
- 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) {
+ /* mul by tex size */
+ coord = lp_build_mul(coord_bld, coord, length_f);
+ }
+ /* clamp to length max */
+ coord = lp_build_min(coord_bld, coord, length_f);
+ /* subtract 0.5 */
coord = lp_build_sub(coord_bld, coord, half);
+ /* clamp to [0, length - 0.5] */
+ coord = lp_build_max(coord_bld, coord, coord_bld->zero);
+ /* convert to int, compute lerp weight */
+ lp_build_ifloor_fract(&abs_coord_bld, coord, &coord0, &weight);
+ coord1 = lp_build_add(int_coord_bld, coord0, int_coord_bld->one);
+ /* coord1 = min(coord1, length-1) */
+ coord1 = lp_build_min(int_coord_bld, coord1, length_minus_one);
+ break;
}
- else {
- LLVMValueRef min, max;
- /* clamp to [0.5, length - 0.5] */
- min = half;
- max = lp_build_sub(coord_bld, length_f, min);
- coord = lp_build_clamp(coord_bld, coord, min, max);
- }
- /* compute lerp weight */
- weight = lp_build_fract(coord_bld, coord);
- /* coord0 = floor(coord); */
- coord0 = lp_build_ifloor(coord_bld, coord);
- coord1 = lp_build_add(int_coord_bld, coord0, int_coord_bld->one);
- /* coord0 = max(coord0, 0) */
- coord0 = lp_build_max(int_coord_bld, coord0, int_coord_bld->zero);
- /* coord1 = min(coord1, length-1) */
- coord1 = lp_build_min(int_coord_bld, coord1, length_minus_one);
- break;
case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
{
- LLVMValueRef min, max;
+ LLVMValueRef min;
if (bld->static_state->normalized_coords) {
/* scale coord to length */
coord = lp_build_mul(coord_bld, coord, length_f);
}
- /* clamp to [-0.5, length + 0.5] */
- min = lp_build_const_vec(coord_bld->type, -0.5F);
- max = lp_build_sub(coord_bld, length_f, min);
- coord = lp_build_clamp(coord_bld, coord, min, max);
+ /* was: clamp to [-0.5, length + 0.5], then sub 0.5 */
coord = lp_build_sub(coord_bld, coord, half);
- /* compute lerp weight */
- weight = lp_build_fract(coord_bld, coord);
- /* convert to int */
- coord0 = lp_build_ifloor(coord_bld, coord);
+ min = lp_build_const_vec(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);
}
break;
coord = lp_build_mul(coord_bld, coord, length_f);
coord = lp_build_sub(coord_bld, coord, half);
- /* compute lerp weight */
- weight = lp_build_fract(coord_bld, coord);
-
- /* convert to int coords */
- coord0 = lp_build_ifloor(coord_bld, coord);
+ /* 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);
/* coord0 = max(coord0, 0) */
coord = lp_build_sub(coord_bld, coord, half);
- weight = lp_build_fract(coord_bld, coord);
- coord0 = lp_build_ifloor(coord_bld, coord);
+ /* 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_CLAMP_TO_EDGE:
{
LLVMValueRef min, max;
-
+ struct lp_build_context abs_coord_bld = bld->coord_bld;
+ abs_coord_bld.type.sign = FALSE;
coord = lp_build_abs(coord_bld, coord);
if (bld->static_state->normalized_coords) {
coord = lp_build_sub(coord_bld, coord, half);
- weight = lp_build_fract(coord_bld, coord);
- coord0 = lp_build_ifloor(coord_bld, coord);
+ /* convert to int, compute lerp weight */
+ lp_build_ifloor_fract(&abs_coord_bld, coord, &coord0, &weight);
coord1 = lp_build_add(int_coord_bld, coord0, int_coord_bld->one);
}
break;
case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
{
- LLVMValueRef min, max;
-
coord = lp_build_abs(coord_bld, coord);
if (bld->static_state->normalized_coords) {
coord = lp_build_mul(coord_bld, coord, length_f);
}
- /* clamp to [-0.5, length + 0.5] */
- min = lp_build_negate(coord_bld, half);
- max = lp_build_sub(coord_bld, length_f, min);
- coord = lp_build_clamp(coord_bld, coord, min, max);
-
+ /* was: clamp to [-0.5, length + 0.5] then sub 0.5 */
+ /* skip -0.5 clamp (always positive), do sub first */
coord = lp_build_sub(coord_bld, coord, half);
+ coord = lp_build_min(coord_bld, coord, length_f);
- weight = lp_build_fract(coord_bld, coord);
- coord0 = lp_build_ifloor(coord_bld, coord);
+ /* 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;
/**
* Build LLVM code for texture wrap mode for nearest filtering.
* \param coord the incoming texcoord (nominally in [0,1])
- * \param length the texture size along one dimension, as int
+ * \param length the texture size along one dimension, as int vector
* \param is_pot if TRUE, length is a power of two
* \param wrap_mode one of PIPE_TEX_WRAP_x
*/
lp_build_sample_wrap_nearest(struct lp_build_sample_context *bld,
LLVMValueRef coord,
LLVMValueRef length,
+ LLVMValueRef length_f,
boolean is_pot,
unsigned wrap_mode)
{
struct lp_build_context *coord_bld = &bld->coord_bld;
struct lp_build_context *int_coord_bld = &bld->int_coord_bld;
- struct lp_build_context *uint_coord_bld = &bld->uint_coord_bld;
- LLVMValueRef length_f = lp_build_int_to_float(coord_bld, length);
- LLVMValueRef length_minus_one = lp_build_sub(uint_coord_bld, length, uint_coord_bld->one);
+ LLVMValueRef length_minus_one = lp_build_sub(int_coord_bld, length, int_coord_bld->one);
LLVMValueRef icoord;
switch(wrap_mode) {
icoord = lp_build_ifloor(coord_bld, coord);
if (is_pot)
icoord = LLVMBuildAnd(bld->builder, icoord, length_minus_one, "");
- else
- /* Signed remainder won't give the right results for negative
- * dividends but unsigned remainder does.*/
+ else {
+ /* Add a bias to the texcoord to handle negative coords */
+ LLVMValueRef bias = lp_build_mul_imm(int_coord_bld, length, 1024);
+ icoord = LLVMBuildAdd(bld->builder, icoord, bias, "");
icoord = LLVMBuildURem(bld->builder, icoord, length, "");
+ }
break;
case PIPE_TEX_WRAP_CLAMP:
}
/* floor */
- icoord = lp_build_ifloor(coord_bld, coord);
+ /* use itrunc instead since we clamp to 0 anyway */
+ icoord = lp_build_itrunc(coord_bld, coord);
/* clamp to [0, length - 1]. */
icoord = lp_build_clamp(int_coord_bld, icoord, int_coord_bld->zero,
assert(bld->static_state->normalized_coords);
coord = lp_build_mul(coord_bld, coord, length_f);
- icoord = lp_build_ifloor(coord_bld, coord);
+ /* itrunc == ifloor here */
+ icoord = lp_build_itrunc(coord_bld, coord);
/* clamp to [0, length - 1] */
icoord = lp_build_min(int_coord_bld, icoord, length_minus_one);
coord = lp_build_mul(coord_bld, coord, length_f);
}
- icoord = lp_build_ifloor(coord_bld, coord);
+ /* itrunc == ifloor here */
+ icoord = lp_build_itrunc(coord_bld, coord);
/* clamp to [0, length - 1] */
icoord = lp_build_min(int_coord_bld, icoord, length_minus_one);
coord = lp_build_mul(coord_bld, coord, length_f);
}
- icoord = lp_build_ifloor(coord_bld, coord);
+ /* itrunc == ifloor here */
+ icoord = lp_build_itrunc(coord_bld, coord);
/* clamp to [0, length] */
icoord = lp_build_min(int_coord_bld, icoord, length);
}
-/**
- * Codegen equivalent for u_minify().
- * Return max(1, base_size >> level);
- */
-static LLVMValueRef
-lp_build_minify(struct lp_build_sample_context *bld,
- LLVMValueRef base_size,
- LLVMValueRef level)
-{
- LLVMValueRef size = LLVMBuildAShr(bld->builder, base_size, level, "minify");
- size = lp_build_max(&bld->int_coord_bld, size, bld->int_coord_bld.one);
- return size;
-}
-
-
-/**
- * Generate code to compute texture level of detail (lambda).
- * \param s vector of texcoord s values
- * \param t vector of texcoord t values
- * \param r vector of texcoord r values
- * \param lod_bias optional float vector with the shader lod bias
- * \param explicit_lod optional float vector with the explicit lod
- * \param width scalar int texture width
- * \param height scalar int texture height
- * \param depth scalar int texture depth
- *
- * XXX: The resulting lod is scalar, so ignore all but the first element of
- * derivatives, lod_bias, etc that are passed by the shader.
- */
-static LLVMValueRef
-lp_build_lod_selector(struct lp_build_sample_context *bld,
- LLVMValueRef s,
- LLVMValueRef t,
- LLVMValueRef r,
- const LLVMValueRef *ddx,
- const LLVMValueRef *ddy,
- LLVMValueRef lod_bias, /* optional */
- LLVMValueRef explicit_lod, /* optional */
- LLVMValueRef width,
- LLVMValueRef height,
- LLVMValueRef depth)
-
-{
- if (bld->static_state->min_lod == bld->static_state->max_lod) {
- /* User is forcing sampling from a particular mipmap level.
- * This is hit during mipmap generation.
- */
- return LLVMConstReal(LLVMFloatType(), bld->static_state->min_lod);
- }
- else {
- struct lp_build_context *float_bld = &bld->float_bld;
- LLVMValueRef sampler_lod_bias = LLVMConstReal(LLVMFloatType(),
- bld->static_state->lod_bias);
- LLVMValueRef min_lod = LLVMConstReal(LLVMFloatType(),
- bld->static_state->min_lod);
- LLVMValueRef max_lod = LLVMConstReal(LLVMFloatType(),
- bld->static_state->max_lod);
- LLVMValueRef index0 = LLVMConstInt(LLVMInt32Type(), 0, 0);
- LLVMValueRef lod;
-
- if (explicit_lod) {
- lod = LLVMBuildExtractElement(bld->builder, explicit_lod,
- index0, "");
- }
- else {
- const int dims = texture_dims(bld->static_state->target);
- LLVMValueRef dsdx, dsdy;
- LLVMValueRef dtdx = NULL, dtdy = NULL, drdx = NULL, drdy = NULL;
- LLVMValueRef rho;
-
- /*
- * dsdx = abs(s[1] - s[0]);
- * dsdy = abs(s[2] - s[0]);
- * dtdx = abs(t[1] - t[0]);
- * dtdy = abs(t[2] - t[0]);
- * drdx = abs(r[1] - r[0]);
- * drdy = abs(r[2] - r[0]);
- */
- dsdx = LLVMBuildExtractElement(bld->builder, ddx[0], index0, "dsdx");
- dsdx = lp_build_abs(float_bld, dsdx);
- dsdy = LLVMBuildExtractElement(bld->builder, ddy[0], index0, "dsdy");
- dsdy = lp_build_abs(float_bld, dsdy);
- if (dims > 1) {
- dtdx = LLVMBuildExtractElement(bld->builder, ddx[1], index0, "dtdx");
- dtdx = lp_build_abs(float_bld, dtdx);
- dtdy = LLVMBuildExtractElement(bld->builder, ddy[1], index0, "dtdy");
- dtdy = lp_build_abs(float_bld, dtdy);
- if (dims > 2) {
- drdx = LLVMBuildExtractElement(bld->builder, ddx[2], index0, "drdx");
- drdx = lp_build_abs(float_bld, drdx);
- drdy = LLVMBuildExtractElement(bld->builder, ddy[2], index0, "drdy");
- drdy = lp_build_abs(float_bld, drdy);
- }
- }
-
- /* Compute rho = max of all partial derivatives scaled by texture size.
- * XXX this could be vectorized somewhat
- */
- rho = LLVMBuildMul(bld->builder,
- lp_build_max(float_bld, dsdx, dsdy),
- lp_build_int_to_float(float_bld, width), "");
- if (dims > 1) {
- LLVMValueRef max;
- max = LLVMBuildMul(bld->builder,
- lp_build_max(float_bld, dtdx, dtdy),
- lp_build_int_to_float(float_bld, height), "");
- rho = lp_build_max(float_bld, rho, max);
- if (dims > 2) {
- max = LLVMBuildMul(bld->builder,
- lp_build_max(float_bld, drdx, drdy),
- lp_build_int_to_float(float_bld, depth), "");
- rho = lp_build_max(float_bld, rho, max);
- }
- }
-
- /* compute lod = log2(rho) */
- lod = lp_build_log2(float_bld, rho);
-
- /* add shader lod bias */
- if (lod_bias) {
- lod_bias = LLVMBuildExtractElement(bld->builder, lod_bias,
- index0, "");
- lod = LLVMBuildAdd(bld->builder, lod, lod_bias, "shader_lod_bias");
- }
- }
-
- /* add sampler lod bias */
- lod = LLVMBuildAdd(bld->builder, lod, sampler_lod_bias, "sampler_lod_bias");
-
- /* clamp lod */
- lod = lp_build_clamp(float_bld, lod, min_lod, max_lod);
-
- return lod;
- }
-}
-
-
-/**
- * For PIPE_TEX_MIPFILTER_NEAREST, convert float LOD to integer
- * mipmap level index.
- * Note: this is all scalar code.
- * \param lod scalar float texture level of detail
- * \param level_out returns integer
- */
-static void
-lp_build_nearest_mip_level(struct lp_build_sample_context *bld,
- unsigned unit,
- LLVMValueRef lod,
- LLVMValueRef *level_out)
-{
- struct lp_build_context *float_bld = &bld->float_bld;
- struct lp_build_context *int_bld = &bld->int_bld;
- LLVMValueRef last_level, level;
-
- LLVMValueRef zero = LLVMConstInt(LLVMInt32Type(), 0, 0);
-
- last_level = bld->dynamic_state->last_level(bld->dynamic_state,
- bld->builder, unit);
-
- /* convert float lod to integer */
- level = lp_build_itrunc(float_bld, lod);
-
- /* clamp level to legal range of levels */
- *level_out = lp_build_clamp(int_bld, level, zero, last_level);
-}
-
-
-/**
- * For PIPE_TEX_MIPFILTER_LINEAR, convert float LOD to integer to
- * two (adjacent) mipmap level indexes. Later, we'll sample from those
- * two mipmap levels and interpolate between them.
- */
-static void
-lp_build_linear_mip_levels(struct lp_build_sample_context *bld,
- unsigned unit,
- LLVMValueRef lod,
- LLVMValueRef *level0_out,
- LLVMValueRef *level1_out,
- LLVMValueRef *weight_out)
-{
- struct lp_build_context *float_bld = &bld->float_bld;
- struct lp_build_context *int_bld = &bld->int_bld;
- LLVMValueRef last_level, level;
-
- last_level = bld->dynamic_state->last_level(bld->dynamic_state,
- bld->builder, unit);
-
- /* convert float lod to integer */
- level = lp_build_ifloor(float_bld, lod);
-
- /* compute level 0 and clamp to legal range of levels */
- *level0_out = lp_build_clamp(int_bld, level,
- int_bld->zero,
- last_level);
- /* compute level 1 and clamp to legal range of levels */
- level = lp_build_add(int_bld, level, int_bld->one);
- *level1_out = lp_build_clamp(int_bld, level,
- int_bld->zero,
- last_level);
-
- *weight_out = lp_build_fract(float_bld, lod);
-}
-
-
/**
* Generate code to sample a mipmap level with nearest filtering.
* If sampling a cube texture, r = cube face in [0,5].
*/
static void
lp_build_sample_image_nearest(struct lp_build_sample_context *bld,
- LLVMValueRef width_vec,
- LLVMValueRef height_vec,
- LLVMValueRef depth_vec,
+ unsigned unit,
+ LLVMValueRef size,
LLVMValueRef row_stride_vec,
LLVMValueRef img_stride_vec,
LLVMValueRef data_ptr,
LLVMValueRef r,
LLVMValueRef colors_out[4])
{
- const int dims = texture_dims(bld->static_state->target);
+ const unsigned dims = bld->dims;
+ LLVMValueRef width_vec;
+ LLVMValueRef height_vec;
+ LLVMValueRef depth_vec;
+ LLVMValueRef flt_size;
+ LLVMValueRef flt_width_vec;
+ LLVMValueRef flt_height_vec;
+ LLVMValueRef flt_depth_vec;
LLVMValueRef x, y, z;
+ lp_build_extract_image_sizes(bld,
+ bld->int_size_type,
+ 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->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,
+ x = lp_build_sample_wrap_nearest(bld, s, width_vec, flt_width_vec,
bld->static_state->pot_width,
bld->static_state->wrap_s);
lp_build_name(x, "tex.x.wrapped");
if (dims >= 2) {
- y = lp_build_sample_wrap_nearest(bld, t, height_vec,
+ y = lp_build_sample_wrap_nearest(bld, t, height_vec, flt_height_vec,
bld->static_state->pot_height,
bld->static_state->wrap_t);
lp_build_name(y, "tex.y.wrapped");
if (dims == 3) {
- z = lp_build_sample_wrap_nearest(bld, r, depth_vec,
- bld->static_state->pot_height,
+ z = lp_build_sample_wrap_nearest(bld, r, depth_vec, flt_depth_vec,
+ bld->static_state->pot_depth,
bld->static_state->wrap_r);
lp_build_name(z, "tex.z.wrapped");
}
/*
* Get texture colors.
*/
- lp_build_sample_texel_soa(bld, width_vec, height_vec, depth_vec,
+ lp_build_sample_texel_soa(bld, unit,
+ width_vec, height_vec, depth_vec,
x, y, z,
row_stride_vec, img_stride_vec,
data_ptr, colors_out);
*/
static void
lp_build_sample_image_linear(struct lp_build_sample_context *bld,
- LLVMValueRef width_vec,
- LLVMValueRef height_vec,
- LLVMValueRef depth_vec,
+ unsigned unit,
+ LLVMValueRef size,
LLVMValueRef row_stride_vec,
LLVMValueRef img_stride_vec,
LLVMValueRef data_ptr,
LLVMValueRef r,
LLVMValueRef colors_out[4])
{
- const int dims = texture_dims(bld->static_state->target);
+ const unsigned dims = bld->dims;
+ LLVMValueRef width_vec;
+ LLVMValueRef height_vec;
+ LLVMValueRef depth_vec;
+ LLVMValueRef flt_size;
+ 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 neighbors[2][2][4];
int chan;
+ lp_build_extract_image_sizes(bld,
+ bld->int_size_type,
+ 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->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,
+ lp_build_sample_wrap_linear(bld, s, width_vec, flt_width_vec,
bld->static_state->pot_width,
bld->static_state->wrap_s,
&x0, &x1, &s_fpart);
lp_build_name(x1, "tex.x1.wrapped");
if (dims >= 2) {
- lp_build_sample_wrap_linear(bld, t, height_vec,
+ lp_build_sample_wrap_linear(bld, t, height_vec, flt_height_vec,
bld->static_state->pot_height,
bld->static_state->wrap_t,
&y0, &y1, &t_fpart);
lp_build_name(y1, "tex.y1.wrapped");
if (dims == 3) {
- lp_build_sample_wrap_linear(bld, r, depth_vec,
+ lp_build_sample_wrap_linear(bld, r, depth_vec, flt_depth_vec,
bld->static_state->pot_depth,
bld->static_state->wrap_r,
&z0, &z1, &r_fpart);
* Get texture colors.
*/
/* get x0/x1 texels */
- lp_build_sample_texel_soa(bld, width_vec, height_vec, depth_vec,
+ lp_build_sample_texel_soa(bld, 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, width_vec, height_vec, depth_vec,
+ lp_build_sample_texel_soa(bld, unit,
+ width_vec, height_vec, depth_vec,
x1, y0, z0,
row_stride_vec, img_stride_vec,
data_ptr, neighbors[0][1]);
LLVMValueRef colors0[4];
/* get x0/x1 texels at y1 */
- lp_build_sample_texel_soa(bld, width_vec, height_vec, depth_vec,
+ lp_build_sample_texel_soa(bld, 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, width_vec, height_vec, depth_vec,
+ lp_build_sample_texel_soa(bld, unit,
+ width_vec, height_vec, depth_vec,
x1, y1, z0,
row_stride_vec, img_stride_vec,
data_ptr, neighbors[1][1]);
LLVMValueRef colors1[4];
/* get x0/x1/y0/y1 texels at z1 */
- lp_build_sample_texel_soa(bld, width_vec, height_vec, depth_vec,
+ lp_build_sample_texel_soa(bld, 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, width_vec, height_vec, depth_vec,
+ lp_build_sample_texel_soa(bld, 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, width_vec, height_vec, depth_vec,
+ lp_build_sample_texel_soa(bld, 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, width_vec, height_vec, depth_vec,
+ lp_build_sample_texel_soa(bld, unit,
+ width_vec, height_vec, depth_vec,
x1, y1, z1,
row_stride_vec, img_stride_vec,
data_ptr, neighbors1[1][1]);
}
-/** Helper used by lp_build_cube_lookup() */
-static LLVMValueRef
-lp_build_cube_ima(struct lp_build_context *coord_bld, LLVMValueRef coord)
-{
- /* ima = -0.5 / abs(coord); */
- LLVMValueRef negHalf = lp_build_const_vec(coord_bld->type, -0.5);
- LLVMValueRef absCoord = lp_build_abs(coord_bld, coord);
- LLVMValueRef ima = lp_build_mul(coord_bld, negHalf,
- lp_build_rcp(coord_bld, absCoord));
- return ima;
-}
-
-
-/**
- * Helper used by lp_build_cube_lookup()
- * \param sign scalar +1 or -1
- * \param coord float vector
- * \param ima float vector
- */
-static LLVMValueRef
-lp_build_cube_coord(struct lp_build_context *coord_bld,
- LLVMValueRef sign, int negate_coord,
- LLVMValueRef coord, LLVMValueRef ima)
-{
- /* return negate(coord) * ima * sign + 0.5; */
- LLVMValueRef half = lp_build_const_vec(coord_bld->type, 0.5);
- LLVMValueRef res;
-
- assert(negate_coord == +1 || negate_coord == -1);
-
- if (negate_coord == -1) {
- coord = lp_build_negate(coord_bld, coord);
- }
-
- res = lp_build_mul(coord_bld, coord, ima);
- if (sign) {
- sign = lp_build_broadcast_scalar(coord_bld, sign);
- res = lp_build_mul(coord_bld, res, sign);
- }
- res = lp_build_add(coord_bld, res, half);
-
- return res;
-}
-
-
-/** Helper used by lp_build_cube_lookup()
- * Return (major_coord >= 0) ? pos_face : neg_face;
- */
-static LLVMValueRef
-lp_build_cube_face(struct lp_build_sample_context *bld,
- LLVMValueRef major_coord,
- unsigned pos_face, unsigned neg_face)
-{
- LLVMValueRef cmp = LLVMBuildFCmp(bld->builder, LLVMRealUGE,
- major_coord,
- bld->float_bld.zero, "");
- LLVMValueRef pos = LLVMConstInt(LLVMInt32Type(), pos_face, 0);
- LLVMValueRef neg = LLVMConstInt(LLVMInt32Type(), neg_face, 0);
- LLVMValueRef res = LLVMBuildSelect(bld->builder, cmp, pos, neg, "");
- return res;
-}
-
-
-
-/**
- * Generate code to do cube face selection and per-face texcoords.
- */
-static void
-lp_build_cube_lookup(struct lp_build_sample_context *bld,
- LLVMValueRef s,
- LLVMValueRef t,
- LLVMValueRef r,
- LLVMValueRef *face,
- LLVMValueRef *face_s,
- LLVMValueRef *face_t)
-{
- struct lp_build_context *float_bld = &bld->float_bld;
- struct lp_build_context *coord_bld = &bld->coord_bld;
- LLVMValueRef rx, ry, rz;
- LLVMValueRef arx, ary, arz;
- LLVMValueRef c25 = LLVMConstReal(LLVMFloatType(), 0.25);
- LLVMValueRef arx_ge_ary, arx_ge_arz;
- LLVMValueRef ary_ge_arx, ary_ge_arz;
- LLVMValueRef arx_ge_ary_arz, ary_ge_arx_arz;
- LLVMValueRef rx_pos, ry_pos, rz_pos;
-
- assert(bld->coord_bld.type.length == 4);
-
- /*
- * Use the average of the four pixel's texcoords to choose the face.
- */
- rx = lp_build_mul(float_bld, c25,
- lp_build_sum_vector(&bld->coord_bld, s));
- ry = lp_build_mul(float_bld, c25,
- lp_build_sum_vector(&bld->coord_bld, t));
- rz = lp_build_mul(float_bld, c25,
- lp_build_sum_vector(&bld->coord_bld, r));
-
- arx = lp_build_abs(float_bld, rx);
- ary = lp_build_abs(float_bld, ry);
- arz = lp_build_abs(float_bld, rz);
-
- /*
- * Compare sign/magnitude of rx,ry,rz to determine face
- */
- arx_ge_ary = LLVMBuildFCmp(bld->builder, LLVMRealUGE, arx, ary, "");
- arx_ge_arz = LLVMBuildFCmp(bld->builder, LLVMRealUGE, arx, arz, "");
- ary_ge_arx = LLVMBuildFCmp(bld->builder, LLVMRealUGE, ary, arx, "");
- ary_ge_arz = LLVMBuildFCmp(bld->builder, LLVMRealUGE, ary, arz, "");
-
- arx_ge_ary_arz = LLVMBuildAnd(bld->builder, arx_ge_ary, arx_ge_arz, "");
- ary_ge_arx_arz = LLVMBuildAnd(bld->builder, ary_ge_arx, ary_ge_arz, "");
-
- rx_pos = LLVMBuildFCmp(bld->builder, LLVMRealUGE, rx, float_bld->zero, "");
- ry_pos = LLVMBuildFCmp(bld->builder, LLVMRealUGE, ry, float_bld->zero, "");
- rz_pos = LLVMBuildFCmp(bld->builder, LLVMRealUGE, rz, float_bld->zero, "");
-
- {
- struct lp_build_flow_context *flow_ctx;
- struct lp_build_if_state if_ctx;
-
- flow_ctx = lp_build_flow_create(bld->builder);
- lp_build_flow_scope_begin(flow_ctx);
-
- *face_s = bld->coord_bld.undef;
- *face_t = bld->coord_bld.undef;
- *face = bld->int_bld.undef;
-
- lp_build_name(*face_s, "face_s");
- lp_build_name(*face_t, "face_t");
- lp_build_name(*face, "face");
-
- lp_build_flow_scope_declare(flow_ctx, face_s);
- lp_build_flow_scope_declare(flow_ctx, face_t);
- lp_build_flow_scope_declare(flow_ctx, face);
-
- lp_build_if(&if_ctx, flow_ctx, bld->builder, arx_ge_ary_arz);
- {
- /* +/- X face */
- LLVMValueRef sign = lp_build_sgn(float_bld, rx);
- LLVMValueRef ima = lp_build_cube_ima(coord_bld, s);
- *face_s = lp_build_cube_coord(coord_bld, sign, +1, r, ima);
- *face_t = lp_build_cube_coord(coord_bld, NULL, +1, t, ima);
- *face = lp_build_cube_face(bld, rx,
- PIPE_TEX_FACE_POS_X,
- PIPE_TEX_FACE_NEG_X);
- }
- lp_build_else(&if_ctx);
- {
- struct lp_build_flow_context *flow_ctx2;
- struct lp_build_if_state if_ctx2;
-
- LLVMValueRef face_s2 = bld->coord_bld.undef;
- LLVMValueRef face_t2 = bld->coord_bld.undef;
- LLVMValueRef face2 = bld->int_bld.undef;
-
- flow_ctx2 = lp_build_flow_create(bld->builder);
- lp_build_flow_scope_begin(flow_ctx2);
- lp_build_flow_scope_declare(flow_ctx2, &face_s2);
- lp_build_flow_scope_declare(flow_ctx2, &face_t2);
- lp_build_flow_scope_declare(flow_ctx2, &face2);
-
- ary_ge_arx_arz = LLVMBuildAnd(bld->builder, ary_ge_arx, ary_ge_arz, "");
-
- lp_build_if(&if_ctx2, flow_ctx2, bld->builder, ary_ge_arx_arz);
- {
- /* +/- Y face */
- LLVMValueRef sign = lp_build_sgn(float_bld, ry);
- LLVMValueRef ima = lp_build_cube_ima(coord_bld, t);
- face_s2 = lp_build_cube_coord(coord_bld, NULL, -1, s, ima);
- face_t2 = lp_build_cube_coord(coord_bld, sign, -1, r, ima);
- face2 = lp_build_cube_face(bld, ry,
- PIPE_TEX_FACE_POS_Y,
- PIPE_TEX_FACE_NEG_Y);
- }
- lp_build_else(&if_ctx2);
- {
- /* +/- Z face */
- LLVMValueRef sign = lp_build_sgn(float_bld, rz);
- LLVMValueRef ima = lp_build_cube_ima(coord_bld, r);
- face_s2 = lp_build_cube_coord(coord_bld, sign, -1, s, ima);
- face_t2 = lp_build_cube_coord(coord_bld, NULL, +1, t, ima);
- face2 = lp_build_cube_face(bld, rz,
- PIPE_TEX_FACE_POS_Z,
- PIPE_TEX_FACE_NEG_Z);
- }
- lp_build_endif(&if_ctx2);
- lp_build_flow_scope_end(flow_ctx2);
- lp_build_flow_destroy(flow_ctx2);
-
- *face_s = face_s2;
- *face_t = face_t2;
- *face = face2;
- }
-
- lp_build_endif(&if_ctx);
- lp_build_flow_scope_end(flow_ctx);
- lp_build_flow_destroy(flow_ctx);
- }
-}
-
-
-
/**
* Sample the texture/mipmap using given image filter and mip filter.
* data0_ptr and data1_ptr point to the two mipmap levels to sample
*/
static void
lp_build_sample_mipmap(struct lp_build_sample_context *bld,
+ unsigned unit,
unsigned img_filter,
unsigned mip_filter,
LLVMValueRef s,
LLVMValueRef t,
LLVMValueRef r,
+ LLVMValueRef ilevel0,
+ LLVMValueRef ilevel1,
LLVMValueRef lod_fpart,
- LLVMValueRef width0_vec,
- LLVMValueRef width1_vec,
- LLVMValueRef height0_vec,
- LLVMValueRef height1_vec,
- LLVMValueRef depth0_vec,
- LLVMValueRef depth1_vec,
- LLVMValueRef row_stride0_vec,
- LLVMValueRef row_stride1_vec,
- LLVMValueRef img_stride0_vec,
- LLVMValueRef img_stride1_vec,
- LLVMValueRef data_ptr0,
- LLVMValueRef data_ptr1,
LLVMValueRef *colors_out)
{
+ LLVMBuilderRef builder = bld->builder;
+ LLVMValueRef size0;
+ LLVMValueRef size1;
+ LLVMValueRef row_stride0_vec;
+ LLVMValueRef row_stride1_vec;
+ LLVMValueRef img_stride0_vec;
+ LLVMValueRef img_stride1_vec;
+ LLVMValueRef data_ptr0;
+ LLVMValueRef data_ptr1;
LLVMValueRef colors0[4], colors1[4];
- int chan;
+ unsigned chan;
+ /* sample the first mipmap level */
+ lp_build_mipmap_level_sizes(bld, ilevel0,
+ &size0,
+ &row_stride0_vec, &img_stride0_vec);
+ data_ptr0 = lp_build_get_mipmap_level(bld, ilevel0);
if (img_filter == PIPE_TEX_FILTER_NEAREST) {
- lp_build_sample_image_nearest(bld,
- width0_vec, height0_vec, depth0_vec,
+ lp_build_sample_image_nearest(bld, unit,
+ size0,
row_stride0_vec, img_stride0_vec,
- data_ptr0, s, t, r, colors0);
-
- if (mip_filter == PIPE_TEX_MIPFILTER_LINEAR) {
- /* sample the second mipmap level, and interp */
- lp_build_sample_image_nearest(bld,
- width1_vec, height1_vec, depth1_vec,
- row_stride1_vec, img_stride1_vec,
- data_ptr1, s, t, r, colors1);
- }
+ data_ptr0, s, t, r,
+ colors0);
}
else {
assert(img_filter == PIPE_TEX_FILTER_LINEAR);
-
- lp_build_sample_image_linear(bld,
- width0_vec, height0_vec, depth0_vec,
+ lp_build_sample_image_linear(bld, unit,
+ size0,
row_stride0_vec, img_stride0_vec,
- data_ptr0, s, t, r, colors0);
-
- if (mip_filter == PIPE_TEX_MIPFILTER_LINEAR) {
- /* sample the second mipmap level, and interp */
- lp_build_sample_image_linear(bld,
- width1_vec, height1_vec, depth1_vec,
- row_stride1_vec, img_stride1_vec,
- data_ptr1, s, t, r, colors1);
- }
+ data_ptr0, s, t, r,
+ colors0);
+ }
+
+ /* Store the first level's colors in the output variables */
+ for (chan = 0; chan < 4; chan++) {
+ LLVMBuildStore(builder, colors0[chan], colors_out[chan]);
}
if (mip_filter == PIPE_TEX_MIPFILTER_LINEAR) {
- /* interpolate samples from the two mipmap levels */
- for (chan = 0; chan < 4; chan++) {
- colors_out[chan] = lp_build_lerp(&bld->texel_bld, lod_fpart,
+ struct lp_build_if_state if_ctx;
+ LLVMValueRef need_lerp;
+
+ /* need_lerp = lod_fpart > 0 */
+ need_lerp = LLVMBuildFCmp(builder, LLVMRealUGT,
+ lod_fpart,
+ bld->float_bld.zero,
+ "need_lerp");
+
+ lp_build_if(&if_ctx, builder, need_lerp);
+ {
+ /* sample the second mipmap level */
+ lp_build_mipmap_level_sizes(bld, ilevel1,
+ &size1,
+ &row_stride1_vec, &img_stride1_vec);
+ data_ptr1 = lp_build_get_mipmap_level(bld, ilevel1);
+ if (img_filter == PIPE_TEX_FILTER_NEAREST) {
+ lp_build_sample_image_nearest(bld, unit,
+ size1,
+ row_stride1_vec, img_stride1_vec,
+ data_ptr1, s, t, r,
+ colors1);
+ }
+ else {
+ lp_build_sample_image_linear(bld, unit,
+ size1,
+ row_stride1_vec, img_stride1_vec,
+ data_ptr1, s, t, r,
+ colors1);
+ }
+
+ /* interpolate samples from the two mipmap levels */
+
+ lod_fpart = lp_build_broadcast_scalar(&bld->texel_bld, lod_fpart);
+
+ for (chan = 0; chan < 4; chan++) {
+ colors0[chan] = lp_build_lerp(&bld->texel_bld, lod_fpart,
colors0[chan], colors1[chan]);
+ LLVMBuildStore(builder, colors0[chan], colors_out[chan]);
+ }
}
- }
- else {
- /* use first/only level's colors */
- for (chan = 0; chan < 4; chan++) {
- colors_out[chan] = colors0[chan];
- }
+ lp_build_endif(&if_ctx);
}
}
const LLVMValueRef *ddy,
LLVMValueRef lod_bias, /* optional */
LLVMValueRef explicit_lod, /* optional */
- LLVMValueRef width,
- LLVMValueRef height,
- LLVMValueRef depth,
- LLVMValueRef width_vec,
- LLVMValueRef height_vec,
- LLVMValueRef depth_vec,
- LLVMValueRef row_stride_array,
- LLVMValueRef img_stride_array,
- LLVMValueRef data_array,
LLVMValueRef *colors_out)
{
- struct lp_build_context *float_bld = &bld->float_bld;
+ struct lp_build_context *int_bld = &bld->int_bld;
+ LLVMBuilderRef builder = bld->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 int dims = texture_dims(bld->static_state->target);
- LLVMValueRef lod = NULL, lod_fpart = NULL;
- LLVMValueRef ilevel0, ilevel1 = NULL, ilevel0_vec, ilevel1_vec = NULL;
- LLVMValueRef width0_vec = NULL, height0_vec = NULL, depth0_vec = NULL;
- LLVMValueRef width1_vec = NULL, height1_vec = NULL, depth1_vec = NULL;
- LLVMValueRef row_stride0_vec = NULL, row_stride1_vec = NULL;
- LLVMValueRef img_stride0_vec = NULL, img_stride1_vec = NULL;
- LLVMValueRef data_ptr0, data_ptr1 = NULL;
+ LLVMValueRef lod_ipart = NULL, lod_fpart = NULL;
+ LLVMValueRef ilevel0, ilevel1 = NULL;
+ LLVMValueRef face_ddx[4], face_ddy[4];
+ LLVMValueRef texels[4];
+ LLVMTypeRef i32t = LLVMInt32Type();
+ LLVMValueRef i32t_zero = LLVMConstInt(i32t, 0, 0);
+ unsigned chan;
/*
printf("%s mip %d min %d mag %d\n", __FUNCTION__,
mip_filter, min_filter, mag_filter);
*/
+ /*
+ * Choose cube face, recompute texcoords and derivatives for the chosen face.
+ */
+ if (bld->static_state->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 */
+ t = face_t; /* vec */
+ /* use 'r' to indicate cube face */
+ r = lp_build_broadcast_scalar(&bld->int_coord_bld, face); /* vec */
+
+ /* recompute ddx, ddy using the new (s,t) face texcoords */
+ face_ddx[0] = lp_build_scalar_ddx(&bld->coord_bld, s);
+ face_ddx[1] = lp_build_scalar_ddx(&bld->coord_bld, t);
+ face_ddx[2] = NULL;
+ face_ddx[3] = NULL;
+ face_ddy[0] = lp_build_scalar_ddy(&bld->coord_bld, s);
+ face_ddy[1] = lp_build_scalar_ddy(&bld->coord_bld, t);
+ face_ddy[2] = NULL;
+ face_ddy[3] = NULL;
+ ddx = face_ddx;
+ ddy = face_ddy;
+ }
+
/*
* 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.
*/
- lod = lp_build_lod_selector(bld, s, t, r, ddx, ddy,
- lod_bias, explicit_lod,
- width, height, depth);
+ lp_build_lod_selector(bld, unit, ddx, ddy,
+ lod_bias, explicit_lod,
+ mip_filter,
+ &lod_ipart, &lod_fpart);
+ } else {
+ lod_ipart = i32t_zero;
}
/*
- * Compute integer mipmap level(s) to fetch texels from.
+ * Compute integer mipmap level(s) to fetch texels from: ilevel0, ilevel1
*/
- if (mip_filter == PIPE_TEX_MIPFILTER_NONE) {
+ switch (mip_filter) {
+ default:
+ assert(0 && "bad mip_filter value in lp_build_sample_soa()");
+ /* fall-through */
+ case PIPE_TEX_MIPFILTER_NONE:
/* always use mip level 0 */
- ilevel0 = LLVMConstInt(LLVMInt32Type(), 0, 0);
- }
- else {
- if (mip_filter == PIPE_TEX_MIPFILTER_NEAREST) {
- lp_build_nearest_mip_level(bld, unit, lod, &ilevel0);
+ if (bld->static_state->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.
+ */
+ assert(lod_ipart);
+ lp_build_nearest_mip_level(bld, unit, lod_ipart, &ilevel0);
}
else {
- assert(mip_filter == PIPE_TEX_MIPFILTER_LINEAR);
- lp_build_linear_mip_levels(bld, unit, lod, &ilevel0, &ilevel1,
- &lod_fpart);
- lod_fpart = lp_build_broadcast_scalar(&bld->coord_bld, lod_fpart);
- }
- }
-
- /*
- * Convert scalar integer mipmap levels into vectors.
- */
- ilevel0_vec = lp_build_broadcast_scalar(&bld->int_coord_bld, ilevel0);
- if (mip_filter == PIPE_TEX_MIPFILTER_LINEAR)
- ilevel1_vec = lp_build_broadcast_scalar(&bld->int_coord_bld, ilevel1);
-
- /*
- * Compute width, height at mipmap level 'ilevel0'
- */
- width0_vec = lp_build_minify(bld, width_vec, ilevel0_vec);
- if (dims >= 2) {
- height0_vec = lp_build_minify(bld, height_vec, ilevel0_vec);
- row_stride0_vec = lp_build_get_level_stride_vec(bld, row_stride_array,
- ilevel0);
- if (dims == 3 || bld->static_state->target == PIPE_TEXTURE_CUBE) {
- img_stride0_vec = lp_build_get_level_stride_vec(bld,
- img_stride_array,
- ilevel0);
- if (dims == 3) {
- depth0_vec = lp_build_minify(bld, depth_vec, ilevel0_vec);
- }
- }
- }
- if (mip_filter == PIPE_TEX_MIPFILTER_LINEAR) {
- /* compute width, height, depth for second mipmap level at 'ilevel1' */
- width1_vec = lp_build_minify(bld, width_vec, ilevel1_vec);
- if (dims >= 2) {
- height1_vec = lp_build_minify(bld, height_vec, ilevel1_vec);
- row_stride1_vec = lp_build_get_level_stride_vec(bld, row_stride_array,
- ilevel1);
- if (dims == 3 || bld->static_state->target == PIPE_TEXTURE_CUBE) {
- img_stride1_vec = lp_build_get_level_stride_vec(bld,
- img_stride_array,
- ilevel1);
- if (dims ==3) {
- depth1_vec = lp_build_minify(bld, depth_vec, ilevel1_vec);
- }
- }
+ ilevel0 = i32t_zero;
}
+ break;
+ case PIPE_TEX_MIPFILTER_NEAREST:
+ assert(lod_ipart);
+ lp_build_nearest_mip_level(bld, unit, lod_ipart, &ilevel0);
+ break;
+ case PIPE_TEX_MIPFILTER_LINEAR:
+ assert(lod_ipart);
+ assert(lod_fpart);
+ lp_build_linear_mip_levels(bld, unit,
+ lod_ipart, &lod_fpart,
+ &ilevel0, &ilevel1);
+ break;
}
/*
- * Choose cube face, recompute per-face texcoords.
+ * Get/interpolate texture colors.
*/
- if (bld->static_state->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 */
- t = face_t; /* vec */
- /* use 'r' to indicate cube face */
- r = lp_build_broadcast_scalar(&bld->int_coord_bld, face); /* vec */
- }
- /*
- * Get pointer(s) to image data for mipmap level(s).
- */
- data_ptr0 = lp_build_get_mipmap_level(bld, data_array, ilevel0);
- if (mip_filter == PIPE_TEX_MIPFILTER_LINEAR) {
- data_ptr1 = lp_build_get_mipmap_level(bld, data_array, ilevel1);
+ for (chan = 0; chan < 4; ++chan) {
+ texels[chan] = lp_build_alloca(builder, bld->texel_bld.vec_type, "");
+ lp_build_name(texels[chan], "sampler%u_texel_%c_var", unit, "xyzw"[chan]);
}
- /*
- * Get/interpolate texture colors.
- */
if (min_filter == mag_filter) {
/* no need to distinquish between minification and magnification */
- lp_build_sample_mipmap(bld, min_filter, mip_filter, s, t, r, lod_fpart,
- width0_vec, width1_vec,
- height0_vec, height1_vec,
- depth0_vec, depth1_vec,
- row_stride0_vec, row_stride1_vec,
- img_stride0_vec, img_stride1_vec,
- data_ptr0, data_ptr1,
- colors_out);
+ lp_build_sample_mipmap(bld, unit,
+ min_filter, mip_filter,
+ s, t, r,
+ ilevel0, ilevel1, lod_fpart,
+ texels);
}
else {
/* Emit conditional to choose min image filter or mag image filter
- * depending on the lod being >0 or <= 0, respectively.
+ * depending on the lod being > 0 or <= 0, respectively.
*/
- struct lp_build_flow_context *flow_ctx;
struct lp_build_if_state if_ctx;
LLVMValueRef minify;
- flow_ctx = lp_build_flow_create(bld->builder);
- lp_build_flow_scope_begin(flow_ctx);
+ /* minify = lod >= 0.0 */
+ minify = LLVMBuildICmp(builder, LLVMIntSGE,
+ lod_ipart, int_bld->zero, "");
- lp_build_flow_scope_declare(flow_ctx, &colors_out[0]);
- lp_build_flow_scope_declare(flow_ctx, &colors_out[1]);
- lp_build_flow_scope_declare(flow_ctx, &colors_out[2]);
- lp_build_flow_scope_declare(flow_ctx, &colors_out[3]);
-
- /* minify = lod > 0.0 */
- minify = LLVMBuildFCmp(bld->builder, LLVMRealUGE,
- lod, float_bld->zero, "");
-
- lp_build_if(&if_ctx, flow_ctx, bld->builder, minify);
+ lp_build_if(&if_ctx, builder, minify);
{
/* Use the minification filter */
- lp_build_sample_mipmap(bld, min_filter, mip_filter,
- s, t, r, lod_fpart,
- width0_vec, width1_vec,
- height0_vec, height1_vec,
- depth0_vec, depth1_vec,
- row_stride0_vec, row_stride1_vec,
- img_stride0_vec, img_stride1_vec,
- data_ptr0, data_ptr1,
- colors_out);
+ lp_build_sample_mipmap(bld, unit,
+ min_filter, mip_filter,
+ s, t, r,
+ ilevel0, ilevel1, lod_fpart,
+ texels);
}
lp_build_else(&if_ctx);
{
/* Use the magnification filter */
- lp_build_sample_mipmap(bld, mag_filter, mip_filter,
- s, t, r, lod_fpart,
- width0_vec, width1_vec,
- height0_vec, height1_vec,
- depth0_vec, depth1_vec,
- row_stride0_vec, row_stride1_vec,
- img_stride0_vec, img_stride1_vec,
- data_ptr0, data_ptr1,
- colors_out);
+ lp_build_sample_mipmap(bld, unit,
+ mag_filter, PIPE_TEX_MIPFILTER_NONE,
+ s, t, r,
+ i32t_zero, NULL, NULL,
+ texels);
}
lp_build_endif(&if_ctx);
-
- lp_build_flow_scope_end(flow_ctx);
- lp_build_flow_destroy(flow_ctx);
- }
-}
-
-
-
-static void
-lp_build_sample_2d_linear_aos(struct lp_build_sample_context *bld,
- LLVMValueRef s,
- LLVMValueRef t,
- LLVMValueRef width,
- LLVMValueRef height,
- LLVMValueRef stride_array,
- LLVMValueRef data_array,
- LLVMValueRef texel_out[4])
-{
- LLVMBuilderRef builder = bld->builder;
- struct lp_build_context i32, h16, u8n;
- LLVMTypeRef i32_vec_type, h16_vec_type, u8n_vec_type;
- LLVMValueRef i32_c8, i32_c128, i32_c255;
- LLVMValueRef s_ipart, s_fpart, s_fpart_lo, s_fpart_hi;
- LLVMValueRef t_ipart, t_fpart, t_fpart_lo, t_fpart_hi;
- LLVMValueRef x0, x1;
- LLVMValueRef y0, y1;
- LLVMValueRef neighbors[2][2];
- LLVMValueRef neighbors_lo[2][2];
- LLVMValueRef neighbors_hi[2][2];
- LLVMValueRef packed, packed_lo, packed_hi;
- LLVMValueRef unswizzled[4];
- LLVMValueRef stride;
-
- assert(bld->static_state->target == PIPE_TEXTURE_2D);
- assert(bld->static_state->min_img_filter == PIPE_TEX_FILTER_LINEAR);
- assert(bld->static_state->mag_img_filter == PIPE_TEX_FILTER_LINEAR);
- assert(bld->static_state->min_mip_filter == PIPE_TEX_MIPFILTER_NONE);
-
- lp_build_context_init(&i32, builder, lp_type_int_vec(32));
- lp_build_context_init(&h16, builder, lp_type_ufixed(16));
- lp_build_context_init(&u8n, builder, lp_type_unorm(8));
-
- i32_vec_type = lp_build_vec_type(i32.type);
- h16_vec_type = lp_build_vec_type(h16.type);
- u8n_vec_type = lp_build_vec_type(u8n.type);
-
- if (bld->static_state->normalized_coords) {
- LLVMTypeRef coord_vec_type = lp_build_vec_type(bld->coord_type);
- LLVMValueRef fp_width = LLVMBuildSIToFP(bld->builder, width, coord_vec_type, "");
- LLVMValueRef fp_height = LLVMBuildSIToFP(bld->builder, height, coord_vec_type, "");
- s = lp_build_mul(&bld->coord_bld, s, fp_width);
- t = lp_build_mul(&bld->coord_bld, t, fp_height);
- }
-
- /* scale coords by 256 (8 fractional bits) */
- s = lp_build_mul_imm(&bld->coord_bld, s, 256);
- t = lp_build_mul_imm(&bld->coord_bld, t, 256);
-
- /* convert float to int */
- s = LLVMBuildFPToSI(builder, s, i32_vec_type, "");
- t = LLVMBuildFPToSI(builder, t, i32_vec_type, "");
-
- /* subtract 0.5 (add -128) */
- i32_c128 = lp_build_const_int_vec(i32.type, -128);
- s = LLVMBuildAdd(builder, s, i32_c128, "");
- t = LLVMBuildAdd(builder, t, i32_c128, "");
-
- /* compute floor (shift right 8) */
- i32_c8 = lp_build_const_int_vec(i32.type, 8);
- s_ipart = LLVMBuildAShr(builder, s, i32_c8, "");
- t_ipart = LLVMBuildAShr(builder, t, i32_c8, "");
-
- /* compute fractional part (AND with 0xff) */
- i32_c255 = lp_build_const_int_vec(i32.type, 255);
- s_fpart = LLVMBuildAnd(builder, s, i32_c255, "");
- t_fpart = LLVMBuildAnd(builder, t, i32_c255, "");
-
- x0 = s_ipart;
- y0 = t_ipart;
-
- x1 = lp_build_add(&bld->int_coord_bld, x0, bld->int_coord_bld.one);
- y1 = lp_build_add(&bld->int_coord_bld, y0, bld->int_coord_bld.one);
-
- x0 = lp_build_sample_wrap_int(bld, x0, width, bld->static_state->pot_width,
- bld->static_state->wrap_s);
- y0 = lp_build_sample_wrap_int(bld, y0, height, bld->static_state->pot_height,
- bld->static_state->wrap_t);
-
- x1 = lp_build_sample_wrap_int(bld, x1, width, bld->static_state->pot_width,
- bld->static_state->wrap_s);
- y1 = lp_build_sample_wrap_int(bld, y1, height, bld->static_state->pot_height,
- bld->static_state->wrap_t);
-
- /*
- * Transform 4 x i32 in
- *
- * s_fpart = {s0, s1, s2, s3}
- *
- * into 8 x i16
- *
- * s_fpart = {00, s0, 00, s1, 00, s2, 00, s3}
- *
- * into two 8 x i16
- *
- * s_fpart_lo = {s0, s0, s0, s0, s1, s1, s1, s1}
- * s_fpart_hi = {s2, s2, s2, s2, s3, s3, s3, s3}
- *
- * and likewise for t_fpart. There is no risk of loosing precision here
- * since the fractional parts only use the lower 8bits.
- */
-
- s_fpart = LLVMBuildBitCast(builder, s_fpart, h16_vec_type, "");
- t_fpart = LLVMBuildBitCast(builder, t_fpart, h16_vec_type, "");
-
- {
- LLVMTypeRef elem_type = LLVMInt32Type();
- LLVMValueRef shuffles_lo[LP_MAX_VECTOR_LENGTH];
- LLVMValueRef shuffles_hi[LP_MAX_VECTOR_LENGTH];
- LLVMValueRef shuffle_lo;
- LLVMValueRef shuffle_hi;
- unsigned i, j;
-
- for(j = 0; j < h16.type.length; j += 4) {
- unsigned subindex = util_cpu_caps.little_endian ? 0 : 1;
- LLVMValueRef index;
-
- index = LLVMConstInt(elem_type, j/2 + subindex, 0);
- for(i = 0; i < 4; ++i)
- shuffles_lo[j + i] = index;
-
- index = LLVMConstInt(elem_type, h16.type.length/2 + j/2 + subindex, 0);
- for(i = 0; i < 4; ++i)
- shuffles_hi[j + i] = index;
- }
-
- shuffle_lo = LLVMConstVector(shuffles_lo, h16.type.length);
- shuffle_hi = LLVMConstVector(shuffles_hi, h16.type.length);
-
- s_fpart_lo = LLVMBuildShuffleVector(builder, s_fpart, h16.undef, shuffle_lo, "");
- t_fpart_lo = LLVMBuildShuffleVector(builder, t_fpart, h16.undef, shuffle_lo, "");
- s_fpart_hi = LLVMBuildShuffleVector(builder, s_fpart, h16.undef, shuffle_hi, "");
- t_fpart_hi = LLVMBuildShuffleVector(builder, t_fpart, h16.undef, shuffle_hi, "");
}
- stride = lp_build_get_const_level_stride_vec(bld, stride_array, 0);
-
- /*
- * Fetch the pixels as 4 x 32bit (rgba order might differ):
- *
- * rgba0 rgba1 rgba2 rgba3
- *
- * bit cast them into 16 x u8
- *
- * r0 g0 b0 a0 r1 g1 b1 a1 r2 g2 b2 a2 r3 g3 b3 a3
- *
- * unpack them into two 8 x i16:
- *
- * r0 g0 b0 a0 r1 g1 b1 a1
- * r2 g2 b2 a2 r3 g3 b3 a3
- *
- * The higher 8 bits of the resulting elements will be zero.
- */
-
- neighbors[0][0] = lp_build_sample_packed(bld, x0, y0, stride, data_array);
- neighbors[0][1] = lp_build_sample_packed(bld, x1, y0, stride, data_array);
- neighbors[1][0] = lp_build_sample_packed(bld, x0, y1, stride, data_array);
- neighbors[1][1] = lp_build_sample_packed(bld, x1, y1, stride, data_array);
-
- neighbors[0][0] = LLVMBuildBitCast(builder, neighbors[0][0], u8n_vec_type, "");
- neighbors[0][1] = LLVMBuildBitCast(builder, neighbors[0][1], u8n_vec_type, "");
- neighbors[1][0] = LLVMBuildBitCast(builder, neighbors[1][0], u8n_vec_type, "");
- neighbors[1][1] = LLVMBuildBitCast(builder, neighbors[1][1], u8n_vec_type, "");
-
- lp_build_unpack2(builder, u8n.type, h16.type, neighbors[0][0], &neighbors_lo[0][0], &neighbors_hi[0][0]);
- lp_build_unpack2(builder, u8n.type, h16.type, neighbors[0][1], &neighbors_lo[0][1], &neighbors_hi[0][1]);
- lp_build_unpack2(builder, u8n.type, h16.type, neighbors[1][0], &neighbors_lo[1][0], &neighbors_hi[1][0]);
- lp_build_unpack2(builder, u8n.type, h16.type, neighbors[1][1], &neighbors_lo[1][1], &neighbors_hi[1][1]);
-
- /*
- * Linear interpolate with 8.8 fixed point.
- */
-
- packed_lo = lp_build_lerp_2d(&h16,
- s_fpart_lo, t_fpart_lo,
- neighbors_lo[0][0],
- neighbors_lo[0][1],
- neighbors_lo[1][0],
- neighbors_lo[1][1]);
-
- packed_hi = lp_build_lerp_2d(&h16,
- s_fpart_hi, t_fpart_hi,
- neighbors_hi[0][0],
- neighbors_hi[0][1],
- neighbors_hi[1][0],
- neighbors_hi[1][1]);
-
- packed = lp_build_pack2(builder, h16.type, u8n.type, packed_lo, packed_hi);
-
- /*
- * Convert to SoA and swizzle.
- */
-
- lp_build_rgba8_to_f32_soa(bld->builder,
- bld->texel_type,
- packed, unswizzled);
-
- if (util_format_is_rgba8_variant(bld->format_desc)) {
- lp_build_format_swizzle_soa(bld->format_desc,
- &bld->texel_bld,
- unswizzled, texel_out);
- } else {
- texel_out[0] = unswizzled[0];
- texel_out[1] = unswizzled[1];
- texel_out[2] = unswizzled[2];
- texel_out[3] = unswizzled[3];
+ 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]);
}
-
- apply_sampler_swizzle(bld, texel_out);
}
+/**
+ * Do shadow test/comparison.
+ * \param p the texcoord Z (aka R, aka P) component
+ * \param texel the texel to compare against (use the X channel)
+ */
static void
lp_build_sample_compare(struct lp_build_sample_context *bld,
LLVMValueRef p,
{
struct lp_build_context *texel_bld = &bld->texel_bld;
LLVMValueRef res;
- unsigned chan;
+ const unsigned chan = 0;
- if(bld->static_state->compare_mode == PIPE_TEX_COMPARE_NONE)
+ if (bld->static_state->compare_mode == PIPE_TEX_COMPARE_NONE)
return;
- /* TODO: Compare before swizzling, to avoid redundant computations */
- res = NULL;
- for(chan = 0; chan < 4; ++chan) {
- LLVMValueRef cmp;
- cmp = lp_build_cmp(texel_bld, bld->static_state->compare_func, p, texel[chan]);
- cmp = lp_build_select(texel_bld, cmp, texel_bld->one, texel_bld->zero);
-
- if(res)
- res = lp_build_add(texel_bld, res, cmp);
- else
- res = cmp;
+ /* debug code */
+ if (0) {
+ LLVMValueRef indx = lp_build_const_int32(0);
+ LLVMValueRef coord = LLVMBuildExtractElement(bld->builder, p, indx, "");
+ LLVMValueRef tex = LLVMBuildExtractElement(bld->builder,
+ texel[chan], indx, "");
+ lp_build_printf(bld->builder, "shadow compare coord %f to texture %f\n",
+ coord, tex);
}
- assert(res);
- res = lp_build_mul(texel_bld, res, lp_build_const_vec(texel_bld->type, 0.25));
+ /* result = (p FUNC texel) ? 1 : 0 */
+ res = lp_build_cmp(texel_bld, bld->static_state->compare_func,
+ p, texel[chan]);
+ res = lp_build_select(texel_bld, res, texel_bld->one, texel_bld->zero);
/* XXX returning result for default GL_DEPTH_TEXTURE_MODE = GL_LUMINANCE */
- for(chan = 0; chan < 3; ++chan)
- texel[chan] = res;
+ texel[0] =
+ texel[1] =
+ texel[2] = res;
texel[3] = texel_bld->one;
}
* Just set texels to white instead of actually sampling the texture.
* For debugging.
*/
-static void
-lp_build_sample_nop(struct lp_build_sample_context *bld,
+void
+lp_build_sample_nop(struct lp_type type,
LLVMValueRef texel_out[4])
{
- struct lp_build_context *texel_bld = &bld->texel_bld;
+ LLVMValueRef one = lp_build_one(type);
unsigned chan;
for (chan = 0; chan < 4; chan++) {
- /*lp_bld_mov(texel_bld, texel, texel_bld->one);*/
- texel_out[chan] = texel_bld->one;
+ texel_out[chan] = one;
}
}
* '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 ddx partial derivatives of (s,t,r,q) with respect to x
+ * \param ddy partial derivatives of (s,t,r,q) with respect to y
*/
void
lp_build_sample_soa(LLVMBuilderRef builder,
unsigned unit,
unsigned num_coords,
const LLVMValueRef *coords,
- const LLVMValueRef *ddx,
- const LLVMValueRef *ddy,
+ const LLVMValueRef ddx[4],
+ const LLVMValueRef ddy[4],
LLVMValueRef lod_bias, /* optional */
LLVMValueRef explicit_lod, /* optional */
LLVMValueRef texel_out[4])
{
+ unsigned dims = texture_dims(static_state->target);
struct lp_build_sample_context bld;
- LLVMValueRef width, width_vec;
- LLVMValueRef height, height_vec;
- LLVMValueRef depth, depth_vec;
- LLVMValueRef row_stride_array, img_stride_array;
- LLVMValueRef data_array;
+ LLVMTypeRef i32t = LLVMInt32Type();
+
LLVMValueRef s;
LLVMValueRef t;
LLVMValueRef r;
+ struct lp_type float_vec_type;
if (0) {
enum pipe_format fmt = static_state->format;
debug_printf("Sample from %s\n", util_format_name(fmt));
}
+ assert(type.floating);
+
/* Setup our build context */
memset(&bld, 0, sizeof bld);
bld.builder = builder;
bld.static_state = static_state;
bld.dynamic_state = dynamic_state;
bld.format_desc = util_format_description(static_state->format);
+ bld.dims = dims;
bld.float_type = lp_type_float(32);
bld.int_type = lp_type_int(32);
bld.coord_type = type;
- bld.uint_coord_type = lp_uint_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.texel_type = type;
+ float_vec_type = lp_type_float_vec(32);
+
lp_build_context_init(&bld.float_bld, builder, bld.float_type);
+ lp_build_context_init(&bld.float_vec_bld, builder, float_vec_type);
lp_build_context_init(&bld.int_bld, builder, bld.int_type);
lp_build_context_init(&bld.coord_bld, builder, bld.coord_type);
- lp_build_context_init(&bld.uint_coord_bld, builder, bld.uint_coord_type);
lp_build_context_init(&bld.int_coord_bld, builder, bld.int_coord_type);
+ lp_build_context_init(&bld.int_size_bld, builder, bld.int_size_type);
+ lp_build_context_init(&bld.float_size_bld, builder, bld.float_size_type);
lp_build_context_init(&bld.texel_bld, builder, bld.texel_type);
/* Get the dynamic state */
- width = dynamic_state->width(dynamic_state, builder, unit);
- height = dynamic_state->height(dynamic_state, builder, unit);
- depth = dynamic_state->depth(dynamic_state, builder, unit);
- row_stride_array = dynamic_state->row_stride(dynamic_state, builder, unit);
- img_stride_array = dynamic_state->img_stride(dynamic_state, builder, unit);
- data_array = dynamic_state->data_ptr(dynamic_state, builder, unit);
+ bld.width = dynamic_state->width(dynamic_state, builder, unit);
+ bld.height = dynamic_state->height(dynamic_state, builder, unit);
+ bld.depth = dynamic_state->depth(dynamic_state, builder, unit);
+ bld.row_stride_array = dynamic_state->row_stride(dynamic_state, builder, unit);
+ bld.img_stride_array = dynamic_state->img_stride(dynamic_state, builder, unit);
+ bld.data_array = dynamic_state->data_ptr(dynamic_state, builder, unit);
/* Note that data_array is an array[level] of pointers to texture images */
s = coords[0];
t = coords[1];
r = coords[2];
- width_vec = lp_build_broadcast_scalar(&bld.uint_coord_bld, width);
- height_vec = lp_build_broadcast_scalar(&bld.uint_coord_bld, height);
- depth_vec = lp_build_broadcast_scalar(&bld.uint_coord_bld, depth);
+ /* width, height, depth as single int vector */
+ if (dims <= 1) {
+ bld.int_size = bld.width;
+ }
+ else {
+ bld.int_size = LLVMBuildInsertElement(builder, bld.int_size_bld.undef,
+ bld.width, LLVMConstInt(i32t, 0, 0), "");
+ if (dims >= 2) {
+ bld.int_size = LLVMBuildInsertElement(builder, bld.int_size,
+ bld.height, LLVMConstInt(i32t, 1, 0), "");
+ if (dims >= 3) {
+ bld.int_size = LLVMBuildInsertElement(builder, bld.int_size,
+ bld.depth, LLVMConstInt(i32t, 2, 0), "");
+ }
+ }
+ }
if (0) {
/* For debug: no-op texture sampling */
- lp_build_sample_nop(&bld, texel_out);
+ lp_build_sample_nop(bld.texel_type, texel_out);
}
else if (util_format_fits_8unorm(bld.format_desc) &&
- bld.format_desc->nr_channels > 1 &&
- static_state->target == PIPE_TEXTURE_2D &&
- static_state->min_img_filter == PIPE_TEX_FILTER_LINEAR &&
- static_state->mag_img_filter == PIPE_TEX_FILTER_LINEAR &&
- static_state->min_mip_filter == PIPE_TEX_MIPFILTER_NONE &&
- is_simple_wrap_mode(static_state->wrap_s) &&
- is_simple_wrap_mode(static_state->wrap_t)) {
- /* special case */
- lp_build_sample_2d_linear_aos(&bld, s, t, width_vec, height_vec,
- row_stride_array, data_array, texel_out);
+ lp_is_simple_wrap_mode(static_state->wrap_s) &&
+ lp_is_simple_wrap_mode(static_state->wrap_t)) {
+ /* do sampling/filtering with fixed pt arithmetic */
+ lp_build_sample_aos(&bld, unit, s, t, r, ddx, ddy,
+ lod_bias, explicit_lod,
+ texel_out);
}
+
else {
+ if ((gallivm_debug & GALLIVM_DEBUG_PERF) &&
+ 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);
+ }
+
lp_build_sample_general(&bld, unit, s, t, r, ddx, ddy,
lod_bias, explicit_lod,
- width, height, depth,
- width_vec, height_vec, depth_vec,
- row_stride_array, img_stride_array,
- data_array,
texel_out);
}