AlexIndustrial/mesa
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

gallivm: specialized x8z24 depthtest path

Browse Source 
Avoid unnecessary masking of non-existant stencil component.
This commit is contained in:
Keith Whitwell
2010-10-06 18:21:56 +01:00
parent 954965366f
commit d2cf757f44
2 changed files with 94 additions and 31 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/gallium/drivers/llvmpipe/lp_bld_depth.c
+93 -2
View File
@@ -71,6 +71,7 @@
#include "gallivm/lp_bld_arit.h"
#include "gallivm/lp_bld_bitarit.h"
#include "gallivm/lp_bld_const.h"
#include "gallivm/lp_bld_conv.h"
#include "gallivm/lp_bld_logic.h"
#include "gallivm/lp_bld_flow.h"
#include "gallivm/lp_bld_intr.h"
@@ -446,7 +447,7 @@ lp_build_occlusion_count(LLVMBuilderRef builder,
* \param format_desc description of the depth/stencil surface
* \param mask the alive/dead pixel mask for the quad (vector)
* \param stencil_refs the front/back stencil ref values (scalar)
* \param z_src the incoming depth/stencil values (a 2x2 quad)
* \param z_src the incoming depth/stencil values (a 2x2 quad, float32)
* \param zs_dst_ptr pointer to depth/stencil values in framebuffer
* \param facing contains float value indicating front/back facing polygon
*/
@@ -454,7 +455,7 @@ void
lp_build_depth_stencil_test(LLVMBuilderRef builder,
const struct pipe_depth_state *depth,
const struct pipe_stencil_state stencil[2],
struct lp_type type,
struct lp_type z_src_type,
const struct util_format_description *format_desc,
struct lp_build_mask_context *mask,
LLVMValueRef stencil_refs[2],
@@ -463,6 +464,7 @@ lp_build_depth_stencil_test(LLVMBuilderRef builder,
LLVMValueRef face,
LLVMValueRef counter)
{
struct lp_type type;
struct lp_build_context bld;
struct lp_build_context sbld;
struct lp_type s_type;
@@ -473,6 +475,95 @@ lp_build_depth_stencil_test(LLVMBuilderRef builder,
LLVMValueRef orig_mask = mask->value;
LLVMValueRef front_facing = NULL;
/* Prototype a simpler path:
*/
if (z_src_type.floating &&
format_desc->format == PIPE_FORMAT_X8Z24_UNORM &&
depth->enabled)
{
LLVMValueRef zscaled;
LLVMValueRef const_ffffff_float;
LLVMValueRef const_8_int;
LLVMTypeRef int32_vec_type;
/* We know the values in z_dst are all >= 0, so allow
* lp_build_compare to use signed compare intrinsics:
*/
type.floating = 0;
type.fixed = 0;
type.sign = 1;
type.norm = 1;
type.width = 32;
type.length = z_src_type.length;
int32_vec_type = LLVMVectorType(LLVMInt32Type(), z_src_type.length);
const_8_int = lp_build_const_int_vec(type, 8);
const_ffffff_float = lp_build_const_vec(z_src_type, (float)0xffffff);
zscaled = LLVMBuildFMul(builder, z_src, const_ffffff_float, "zscaled");
z_src = LLVMBuildFPToSI(builder, zscaled, int32_vec_type, "z_src");
/* Load current z/stencil value from z/stencil buffer */
z_dst = LLVMBuildLoad(builder, zs_dst_ptr, "zsbufval");
z_dst = LLVMBuildLShr(builder, z_dst, const_8_int, "z_dst");
/* compare src Z to dst Z, returning 'pass' mask */
z_pass = lp_build_compare(builder,
type,
depth->func, z_src, z_dst);
lp_build_mask_update(mask, z_pass);
/* No need to worry about old stencil contents, just blend the
* old and new values and shift into the correct position for
* storage.
*/
if (depth->writemask) {
type.sign = 0;
lp_build_context_init(&bld, builder, type);
z_dst = lp_build_select(&bld, mask->value, z_src, z_dst);
z_dst = LLVMBuildShl(builder, z_dst, const_8_int, "z_dst");
LLVMBuildStore(builder, z_dst, zs_dst_ptr);
}
if (counter)
lp_build_occlusion_count(builder, type, mask->value, counter);
return;
}
/*
* Depths are expected to be between 0 and 1, even if they are stored in
* floats. Setting these bits here will ensure that the lp_build_conv() call
* below won't try to unnecessarily clamp the incoming values.
*/
if(z_src_type.floating) {
z_src_type.sign = FALSE;
z_src_type.norm = TRUE;
}
else {
assert(!z_src_type.sign);
assert(z_src_type.norm);
}
/* Pick the depth type. */
type = lp_depth_type(format_desc, z_src_type.width*z_src_type.length);
/* FIXME: Cope with a depth test type with a different bit width. */
assert(type.width == z_src_type.width);
assert(type.length == z_src_type.length);
/* Convert fragment Z from float to integer */
lp_build_conv(builder, z_src_type, type, &z_src, 1, &z_src, 1);
zs_dst_ptr = LLVMBuildBitCast(builder,
zs_dst_ptr,
LLVMPointerType(lp_build_vec_type(type), 0), "");
/* Sanity checking */
{
const unsigned z_swizzle = format_desc->swizzle[0];
src/gallium/drivers/llvmpipe/lp_state_fs.c
+1 -29
View File
@@ -119,7 +119,6 @@ generate_depth_stencil(LLVMBuilderRef builder,
LLVMValueRef counter)
{
const struct util_format_description *format_desc;
struct lp_type dst_type;
if (!key->depth.enabled && !key->stencil[0].enabled && !key->stencil[1].enabled)
return;
@@ -127,37 +126,10 @@ generate_depth_stencil(LLVMBuilderRef builder,
format_desc = util_format_description(key->zsbuf_format);
assert(format_desc);
/*
* Depths are expected to be between 0 and 1, even if they are stored in
* floats. Setting these bits here will ensure that the lp_build_conv() call
* below won't try to unnecessarily clamp the incoming values.
*/
if(src_type.floating) {
src_type.sign = FALSE;
src_type.norm = TRUE;
}
else {
assert(!src_type.sign);
assert(src_type.norm);
}
/* Pick the depth type. */
dst_type = lp_depth_type(format_desc, src_type.width*src_type.length);
/* FIXME: Cope with a depth test type with a different bit width. */
assert(dst_type.width == src_type.width);
assert(dst_type.length == src_type.length);
/* Convert fragment Z from float to integer */
lp_build_conv(builder, src_type, dst_type, &src, 1, &src, 1);
dst_ptr = LLVMBuildBitCast(builder,
dst_ptr,
LLVMPointerType(lp_build_vec_type(dst_type), 0), "");
lp_build_depth_stencil_test(builder,
&key->depth,
key->stencil,
dst_type,
src_type,
format_desc,
mask,
stencil_refs,