Mesa (master): gallivm: Eliminate unsigned integer arithmetic from texture coordinates.
Jose Fonseca
jrfonseca at kemper.freedesktop.org
Mon Oct 11 07:14:24 UTC 2010
Module: Mesa
Branch: master
Commit: e1003336f0dcd9248c0127fbdc173522e35c5bdb
URL: http://cgit.freedesktop.org/mesa/mesa/commit/?id=e1003336f0dcd9248c0127fbdc173522e35c5bdb
Author: José Fonseca <jfonseca at vmware.com>
Date: Sun Oct 10 23:55:24 2010 +0100
gallivm: Eliminate unsigned integer arithmetic from texture coordinates.
SSE support for 32bit and 16bit unsigned arithmetic is not complete, and
can easily result in inefficient code.
In most cases signed/unsigned doesn't make a difference, such as for
integer texture coordinates.
So remove uint_coord_type and uint_coord_bld to avoid inefficient
operations to sneak in the future.
---
src/gallium/auxiliary/gallivm/lp_bld_sample.h | 4 --
src/gallium/auxiliary/gallivm/lp_bld_sample_aos.c | 50 ++++++++++-----------
src/gallium/auxiliary/gallivm/lp_bld_sample_soa.c | 20 +++-----
3 files changed, 32 insertions(+), 42 deletions(-)
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_sample.h b/src/gallium/auxiliary/gallivm/lp_bld_sample.h
index ce22854..ffed27c 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_sample.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_sample.h
@@ -197,10 +197,6 @@ struct lp_build_sample_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;
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_sample_aos.c b/src/gallium/auxiliary/gallivm/lp_bld_sample_aos.c
index be5d9a2..641d24b 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_sample_aos.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_sample_aos.c
@@ -81,11 +81,10 @@ lp_build_sample_wrap_nearest_int(struct lp_build_sample_context *bld,
LLVMValueRef *out_offset,
LLVMValueRef *out_i)
{
- 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);
+ length_minus_one = lp_build_sub(int_coord_bld, length, int_coord_bld->one);
switch(wrap_mode) {
case PIPE_TEX_WRAP_REPEAT:
@@ -93,7 +92,7 @@ lp_build_sample_wrap_nearest_int(struct lp_build_sample_context *bld,
coord = LLVMBuildAnd(bld->builder, coord, length_minus_one, "");
else {
/* Add a bias to the texcoord to handle negative coords */
- LLVMValueRef bias = lp_build_mul_imm(uint_coord_bld, length, 1024);
+ LLVMValueRef bias = lp_build_mul_imm(int_coord_bld, length, 1024);
coord = LLVMBuildAdd(bld->builder, coord, bias, "");
coord = LLVMBuildURem(bld->builder, coord, length, "");
}
@@ -114,7 +113,7 @@ lp_build_sample_wrap_nearest_int(struct lp_build_sample_context *bld,
assert(0);
}
- lp_build_sample_partial_offset(uint_coord_bld, block_length, coord, stride,
+ lp_build_sample_partial_offset(int_coord_bld, block_length, coord, stride,
out_offset, out_i);
}
@@ -147,7 +146,6 @@ lp_build_sample_wrap_linear_int(struct lp_build_sample_context *bld,
LLVMValueRef *i0,
LLVMValueRef *i1)
{
- 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;
LLVMValueRef lmask, umask, mask;
@@ -189,8 +187,8 @@ lp_build_sample_wrap_linear_int(struct lp_build_sample_context *bld,
* multiplication.
*/
- *i0 = uint_coord_bld->zero;
- *i1 = uint_coord_bld->zero;
+ *i0 = int_coord_bld->zero;
+ *i1 = int_coord_bld->zero;
length_minus_one = lp_build_sub(int_coord_bld, length, int_coord_bld->one);
@@ -201,7 +199,7 @@ lp_build_sample_wrap_linear_int(struct lp_build_sample_context *bld,
}
else {
/* Add a bias to the texcoord to handle negative coords */
- LLVMValueRef bias = lp_build_mul_imm(uint_coord_bld, length, 1024);
+ LLVMValueRef bias = lp_build_mul_imm(int_coord_bld, length, 1024);
coord0 = LLVMBuildAdd(bld->builder, coord0, bias, "");
coord0 = LLVMBuildURem(bld->builder, coord0, length, "");
}
@@ -209,9 +207,9 @@ lp_build_sample_wrap_linear_int(struct lp_build_sample_context *bld,
mask = lp_build_compare(bld->builder, int_coord_bld->type,
PIPE_FUNC_NOTEQUAL, coord0, length_minus_one);
- *offset0 = lp_build_mul(uint_coord_bld, coord0, stride);
+ *offset0 = lp_build_mul(int_coord_bld, coord0, stride);
*offset1 = LLVMBuildAnd(bld->builder,
- lp_build_add(uint_coord_bld, *offset0, stride),
+ lp_build_add(int_coord_bld, *offset0, stride),
mask, "");
break;
@@ -226,8 +224,8 @@ lp_build_sample_wrap_linear_int(struct lp_build_sample_context *bld,
mask = LLVMBuildAnd(bld->builder, lmask, umask, "");
- *offset0 = lp_build_mul(uint_coord_bld, coord0, stride);
- *offset1 = lp_build_add(uint_coord_bld,
+ *offset0 = lp_build_mul(int_coord_bld, coord0, stride);
+ *offset1 = lp_build_add(int_coord_bld,
*offset0,
LLVMBuildAnd(bld->builder, stride, mask, ""));
break;
@@ -240,8 +238,8 @@ lp_build_sample_wrap_linear_int(struct lp_build_sample_context *bld,
case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
default:
assert(0);
- *offset0 = uint_coord_bld->zero;
- *offset1 = uint_coord_bld->zero;
+ *offset0 = int_coord_bld->zero;
+ *offset1 = int_coord_bld->zero;
break;
}
}
@@ -327,7 +325,7 @@ lp_build_sample_image_nearest(struct lp_build_sample_context *bld,
r_ipart = LLVMBuildAShr(builder, r, i32_c8, "");
/* get pixel, row, image strides */
- x_stride = lp_build_const_vec(bld->uint_coord_bld.type,
+ x_stride = lp_build_const_vec(bld->int_coord_bld.type,
bld->format_desc->block.bits/8);
/* Do texcoord wrapping, compute texel offset */
@@ -346,7 +344,7 @@ lp_build_sample_image_nearest(struct lp_build_sample_context *bld,
bld->static_state->pot_height,
bld->static_state->wrap_t,
&y_offset, &y_subcoord);
- offset = lp_build_add(&bld->uint_coord_bld, offset, y_offset);
+ offset = lp_build_add(&bld->int_coord_bld, offset, y_offset);
if (dims >= 3) {
LLVMValueRef z_offset;
lp_build_sample_wrap_nearest_int(bld,
@@ -355,13 +353,13 @@ lp_build_sample_image_nearest(struct lp_build_sample_context *bld,
bld->static_state->pot_height,
bld->static_state->wrap_r,
&z_offset, &z_subcoord);
- offset = lp_build_add(&bld->uint_coord_bld, offset, z_offset);
+ offset = lp_build_add(&bld->int_coord_bld, offset, z_offset);
}
else if (bld->static_state->target == PIPE_TEXTURE_CUBE) {
LLVMValueRef z_offset;
/* The r coord is the cube face in [0,5] */
- z_offset = lp_build_mul(&bld->uint_coord_bld, r, img_stride_vec);
- offset = lp_build_add(&bld->uint_coord_bld, offset, z_offset);
+ z_offset = lp_build_mul(&bld->int_coord_bld, r, img_stride_vec);
+ offset = lp_build_add(&bld->int_coord_bld, offset, z_offset);
}
}
@@ -522,7 +520,7 @@ lp_build_sample_image_linear(struct lp_build_sample_context *bld,
r_fpart = LLVMBuildAnd(builder, r, i32_c255, "");
/* get pixel, row and image strides */
- x_stride = lp_build_const_vec(bld->uint_coord_bld.type,
+ x_stride = lp_build_const_vec(bld->int_coord_bld.type,
bld->format_desc->block.bits/8);
y_stride = row_stride_vec;
z_stride = img_stride_vec;
@@ -553,9 +551,9 @@ lp_build_sample_image_linear(struct lp_build_sample_context *bld,
for (z = 0; z < 2; z++) {
for (x = 0; x < 2; x++) {
- offset[z][0][x] = lp_build_add(&bld->uint_coord_bld,
+ offset[z][0][x] = lp_build_add(&bld->int_coord_bld,
offset[z][0][x], y_offset0);
- offset[z][1][x] = lp_build_add(&bld->uint_coord_bld,
+ offset[z][1][x] = lp_build_add(&bld->int_coord_bld,
offset[z][1][x], y_offset1);
}
}
@@ -571,20 +569,20 @@ lp_build_sample_image_linear(struct lp_build_sample_context *bld,
&z_subcoord[0], &z_subcoord[1]);
for (y = 0; y < 2; y++) {
for (x = 0; x < 2; x++) {
- offset[0][y][x] = lp_build_add(&bld->uint_coord_bld,
+ offset[0][y][x] = lp_build_add(&bld->int_coord_bld,
offset[0][y][x], z_offset0);
- offset[1][y][x] = lp_build_add(&bld->uint_coord_bld,
+ offset[1][y][x] = lp_build_add(&bld->int_coord_bld,
offset[1][y][x], z_offset1);
}
}
}
else if (bld->static_state->target == PIPE_TEXTURE_CUBE) {
LLVMValueRef z_offset;
- z_offset = lp_build_mul(&bld->uint_coord_bld, r, img_stride_vec);
+ z_offset = lp_build_mul(&bld->int_coord_bld, r, img_stride_vec);
for (y = 0; y < 2; y++) {
for (x = 0; x < 2; x++) {
/* The r coord is the cube face in [0,5] */
- offset[0][y][x] = lp_build_add(&bld->uint_coord_bld,
+ offset[0][y][x] = lp_build_add(&bld->int_coord_bld,
offset[0][y][x], z_offset);
}
}
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_sample_soa.c b/src/gallium/auxiliary/gallivm/lp_bld_sample_soa.c
index 75130e1..af3f468 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_sample_soa.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_sample_soa.c
@@ -131,7 +131,7 @@ lp_build_sample_texel_soa(struct lp_build_sample_context *bld,
}
/* 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);
@@ -145,7 +145,7 @@ lp_build_sample_texel_soa(struct lp_build_sample_context *bld,
* coords which are out of bounds to become zero. Zero's guaranteed
* to be inside the texture image.
*/
- offset = lp_build_andnot(&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,
@@ -239,9 +239,8 @@ lp_build_sample_wrap_linear(struct lp_build_sample_context *bld,
{
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_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) {
@@ -253,20 +252,20 @@ lp_build_sample_wrap_linear(struct lp_build_sample_context *bld,
lp_build_ifloor_fract(coord_bld, coord, &coord0, &weight);
/* repeat wrap */
if (is_pot) {
- coord1 = lp_build_add(uint_coord_bld, coord0, uint_coord_bld->one);
+ 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 {
/* Add a bias to the texcoord to handle negative coords */
- LLVMValueRef bias = lp_build_mul_imm(uint_coord_bld, length, 1024);
+ 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, "");
mask = lp_build_compare(bld->builder, int_coord_bld->type,
PIPE_FUNC_NOTEQUAL, coord0, length_minus_one);
coord1 = LLVMBuildAnd(bld->builder,
- lp_build_add(uint_coord_bld, coord0, uint_coord_bld->one),
+ lp_build_add(int_coord_bld, coord0, int_coord_bld->one),
mask, "");
}
break;
@@ -448,8 +447,7 @@ lp_build_sample_wrap_nearest(struct lp_build_sample_context *bld,
{
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_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) {
@@ -460,7 +458,7 @@ lp_build_sample_wrap_nearest(struct lp_build_sample_context *bld,
icoord = LLVMBuildAnd(bld->builder, icoord, length_minus_one, "");
else {
/* Add a bias to the texcoord to handle negative coords */
- LLVMValueRef bias = lp_build_mul_imm(uint_coord_bld, length, 1024);
+ LLVMValueRef bias = lp_build_mul_imm(int_coord_bld, length, 1024);
icoord = LLVMBuildAdd(bld->builder, icoord, bias, "");
icoord = LLVMBuildURem(bld->builder, icoord, length, "");
}
@@ -1199,7 +1197,6 @@ lp_build_sample_soa(LLVMBuilderRef builder,
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;
@@ -1212,7 +1209,6 @@ lp_build_sample_soa(LLVMBuilderRef builder,
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);
More information about the mesa-commit
mailing list