summaryrefslogtreecommitdiff
path: root/src/gallium/auxiliary/gallivm
diff options
context:
space:
mode:
Diffstat (limited to 'src/gallium/auxiliary/gallivm')
-rw-r--r--src/gallium/auxiliary/gallivm/lp_bld.h47
-rw-r--r--src/gallium/auxiliary/gallivm/lp_bld_alpha.h2
-rw-r--r--src/gallium/auxiliary/gallivm/lp_bld_arit.c69
-rw-r--r--src/gallium/auxiliary/gallivm/lp_bld_arit.h2
-rw-r--r--src/gallium/auxiliary/gallivm/lp_bld_blend.h2
-rw-r--r--src/gallium/auxiliary/gallivm/lp_bld_const.c8
-rw-r--r--src/gallium/auxiliary/gallivm/lp_bld_const.h10
-rw-r--r--src/gallium/auxiliary/gallivm/lp_bld_conv.c34
-rw-r--r--src/gallium/auxiliary/gallivm/lp_bld_conv.h2
-rw-r--r--src/gallium/auxiliary/gallivm/lp_bld_debug.h2
-rw-r--r--src/gallium/auxiliary/gallivm/lp_bld_depth.c608
-rw-r--r--src/gallium/auxiliary/gallivm/lp_bld_depth.h19
-rw-r--r--src/gallium/auxiliary/gallivm/lp_bld_flow.h2
-rw-r--r--src/gallium/auxiliary/gallivm/lp_bld_format.h2
-rw-r--r--src/gallium/auxiliary/gallivm/lp_bld_format_soa.c4
-rw-r--r--src/gallium/auxiliary/gallivm/lp_bld_interp.c6
-rw-r--r--src/gallium/auxiliary/gallivm/lp_bld_interp.h2
-rw-r--r--src/gallium/auxiliary/gallivm/lp_bld_intr.h2
-rw-r--r--src/gallium/auxiliary/gallivm/lp_bld_logic.c12
-rw-r--r--src/gallium/auxiliary/gallivm/lp_bld_logic.h7
-rw-r--r--src/gallium/auxiliary/gallivm/lp_bld_pack.c4
-rw-r--r--src/gallium/auxiliary/gallivm/lp_bld_pack.h2
-rw-r--r--src/gallium/auxiliary/gallivm/lp_bld_sample.c6
-rw-r--r--src/gallium/auxiliary/gallivm/lp_bld_sample.h2
-rw-r--r--src/gallium/auxiliary/gallivm/lp_bld_sample_soa.c401
-rw-r--r--src/gallium/auxiliary/gallivm/lp_bld_struct.h2
-rw-r--r--src/gallium/auxiliary/gallivm/lp_bld_swizzle.c4
-rw-r--r--src/gallium/auxiliary/gallivm/lp_bld_swizzle.h2
-rw-r--r--src/gallium/auxiliary/gallivm/lp_bld_tgsi.h2
-rw-r--r--src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c6
-rw-r--r--src/gallium/auxiliary/gallivm/lp_bld_type.h4
31 files changed, 952 insertions, 325 deletions
diff --git a/src/gallium/auxiliary/gallivm/lp_bld.h b/src/gallium/auxiliary/gallivm/lp_bld.h
new file mode 100644
index 0000000000..70a4960f91
--- /dev/null
+++ b/src/gallium/auxiliary/gallivm/lp_bld.h
@@ -0,0 +1,47 @@
+/**************************************************************************
+ *
+ * Copyright 2010 VMware, Inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * @file
+ * Wrapper for LLVM header file #includes.
+ */
+
+
+#ifndef LP_BLD_H
+#define LP_BLD_H
+
+
+#include <llvm-c/Core.h>
+
+
+/** Set version to 0 if missing to avoid #ifdef HAVE_LLVM everywhere */
+#ifndef HAVE_LLVM
+#define HAVE_LLVM 0x0207
+#endif
+
+
+#endif /* LP_BLD_H */
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_alpha.h b/src/gallium/auxiliary/gallivm/lp_bld_alpha.h
index fe3cedcc48..0f99fec65e 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_alpha.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_alpha.h
@@ -35,7 +35,7 @@
#define LP_BLD_ALPHA_H
-#include "os/os_llvm.h"
+#include "gallivm/lp_bld.h"
struct pipe_alpha_state;
struct lp_type;
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_arit.c b/src/gallium/auxiliary/gallivm/lp_bld_arit.c
index aa47338b32..8e8fcccf56 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_arit.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_arit.c
@@ -361,12 +361,12 @@ lp_build_mul_u8n(LLVMBuilderRef builder,
LLVMValueRef c8;
LLVMValueRef ab;
- c8 = lp_build_int_const_scalar(i16_type, 8);
+ c8 = lp_build_const_int_vec(i16_type, 8);
#if 0
/* a*b/255 ~= (a*(b + 1)) >> 256 */
- b = LLVMBuildAdd(builder, b, lp_build_int_const_scalar(i16_type, 1), "");
+ b = LLVMBuildAdd(builder, b, lp_build_const_int_vec(i16_type, 1), "");
ab = LLVMBuildMul(builder, a, b, "");
#else
@@ -374,7 +374,7 @@ lp_build_mul_u8n(LLVMBuilderRef builder,
/* ab/255 ~= (ab + (ab >> 8) + 0x80) >> 8 */
ab = LLVMBuildMul(builder, a, b, "");
ab = LLVMBuildAdd(builder, ab, LLVMBuildLShr(builder, ab, c8, ""), "");
- ab = LLVMBuildAdd(builder, ab, lp_build_int_const_scalar(i16_type, 0x80), "");
+ ab = LLVMBuildAdd(builder, ab, lp_build_const_int_vec(i16_type, 0x80), "");
#endif
@@ -429,7 +429,7 @@ lp_build_mul(struct lp_build_context *bld,
}
if(type.fixed)
- shift = lp_build_int_const_scalar(type, type.width/2);
+ shift = lp_build_const_int_vec(type, type.width/2);
else
shift = NULL;
@@ -491,7 +491,7 @@ lp_build_mul_imm(struct lp_build_context *bld,
* for Inf and NaN.
*/
unsigned mantissa = lp_mantissa(bld->type);
- factor = lp_build_int_const_scalar(bld->type, (unsigned long long)shift << mantissa);
+ factor = lp_build_const_int_vec(bld->type, (unsigned long long)shift << mantissa);
a = LLVMBuildBitCast(bld->builder, a, lp_build_int_vec_type(bld->type), "");
a = LLVMBuildAdd(bld->builder, a, factor, "");
a = LLVMBuildBitCast(bld->builder, a, lp_build_vec_type(bld->type), "");
@@ -499,12 +499,12 @@ lp_build_mul_imm(struct lp_build_context *bld,
#endif
}
else {
- factor = lp_build_const_scalar(bld->type, shift);
+ factor = lp_build_const_vec(bld->type, shift);
return LLVMBuildShl(bld->builder, a, factor, "");
}
}
- factor = lp_build_const_scalar(bld->type, (double)b);
+ factor = lp_build_const_vec(bld->type, (double)b);
return lp_build_mul(bld, a, factor);
}
@@ -567,7 +567,7 @@ lp_build_lerp(struct lp_build_context *bld,
* but it will be wrong for other uses. Basically we need a more
* powerful lp_type, capable of further distinguishing the values
* interpretation from the value storage. */
- res = LLVMBuildAnd(bld->builder, res, lp_build_int_const_scalar(bld->type, (1 << bld->type.width/2) - 1), "");
+ res = LLVMBuildAnd(bld->builder, res, lp_build_const_int_vec(bld->type, (1 << bld->type.width/2) - 1), "");
return res;
}
@@ -689,7 +689,7 @@ lp_build_abs(struct lp_build_context *bld,
/* vector of floats */
LLVMTypeRef int_vec_type = lp_build_int_vec_type(type);
unsigned long long absMask = ~(1ULL << (type.width - 1));
- LLVMValueRef mask = lp_build_int_const_scalar(type, ((unsigned long long) absMask));
+ LLVMValueRef mask = lp_build_const_int_vec(type, ((unsigned long long) absMask));
a = LLVMBuildBitCast(bld->builder, a, int_vec_type, "");
a = LLVMBuildAnd(bld->builder, a, mask, "");
a = LLVMBuildBitCast(bld->builder, a, vec_type, "");
@@ -751,7 +751,7 @@ lp_build_sgn(struct lp_build_context *bld,
/* vector */
int_type = lp_build_int_vec_type(type);
vec_type = lp_build_vec_type(type);
- mask = lp_build_int_const_scalar(type, maskBit);
+ mask = lp_build_const_int_vec(type, maskBit);
}
/* Take the sign bit and add it to 1 constant */
@@ -763,14 +763,14 @@ lp_build_sgn(struct lp_build_context *bld,
}
else
{
- LLVMValueRef minus_one = lp_build_const_scalar(type, -1.0);
+ LLVMValueRef minus_one = lp_build_const_vec(type, -1.0);
cond = lp_build_cmp(bld, PIPE_FUNC_GREATER, a, bld->zero);
res = lp_build_select(bld, cond, bld->one, minus_one);
}
/* Handle zero */
cond = lp_build_cmp(bld, PIPE_FUNC_EQUAL, a, bld->zero);
- res = lp_build_select(bld, cond, bld->zero, bld->one);
+ res = lp_build_select(bld, cond, bld->zero, res);
return res;
}
@@ -789,8 +789,8 @@ lp_build_set_sign(struct lp_build_context *bld,
const struct lp_type type = bld->type;
LLVMTypeRef int_vec_type = lp_build_int_vec_type(type);
LLVMTypeRef vec_type = lp_build_vec_type(type);
- LLVMValueRef shift = lp_build_int_const_scalar(type, type.width - 1);
- LLVMValueRef mask = lp_build_int_const_scalar(type,
+ LLVMValueRef shift = lp_build_const_int_vec(type, type.width - 1);
+ LLVMValueRef mask = lp_build_const_int_vec(type,
~((unsigned long long) 1 << (type.width - 1)));
LLVMValueRef val, res;
@@ -930,7 +930,10 @@ lp_build_floor(struct lp_build_context *bld,
assert(type.floating);
if (type.length == 1) {
- return LLVMBuildFPTrunc(bld->builder, a, LLVMFloatType(), "");
+ LLVMValueRef res;
+ res = lp_build_ifloor(bld, a);
+ res = LLVMBuildSIToFP(bld->builder, res, LLVMFloatType(), "");
+ return res;
}
if(util_cpu_caps.has_sse4_1)
@@ -993,7 +996,7 @@ lp_build_itrunc(struct lp_build_context *bld,
if (type.length == 1) {
LLVMTypeRef int_type = LLVMIntType(type.width);
- return LLVMBuildFPTrunc(bld->builder, a, int_type, "");
+ return LLVMBuildFPToSI(bld->builder, a, int_type, "");
}
else {
LLVMTypeRef int_vec_type = lp_build_int_vec_type(type);
@@ -1031,7 +1034,7 @@ lp_build_iround(struct lp_build_context *bld,
}
else {
LLVMTypeRef vec_type = lp_build_vec_type(type);
- LLVMValueRef mask = lp_build_int_const_scalar(type, (unsigned long long)1 << (type.width - 1));
+ LLVMValueRef mask = lp_build_const_int_vec(type, (unsigned long long)1 << (type.width - 1));
LLVMValueRef sign;
LLVMValueRef half;
@@ -1040,7 +1043,7 @@ lp_build_iround(struct lp_build_context *bld,
sign = LLVMBuildAnd(bld->builder, sign, mask, "");
/* sign * 0.5 */
- half = lp_build_const_scalar(type, 0.5);
+ half = lp_build_const_vec(type, 0.5);
half = LLVMBuildBitCast(bld->builder, half, int_vec_type, "");
half = LLVMBuildOr(bld->builder, sign, half, "");
half = LLVMBuildBitCast(bld->builder, half, vec_type, "");
@@ -1083,18 +1086,18 @@ lp_build_ifloor(struct lp_build_context *bld,
/* Take the sign bit and add it to 1 constant */
LLVMTypeRef vec_type = lp_build_vec_type(type);
unsigned mantissa = lp_mantissa(type);
- LLVMValueRef mask = lp_build_int_const_scalar(type, (unsigned long long)1 << (type.width - 1));
+ LLVMValueRef mask = lp_build_const_int_vec(type, (unsigned long long)1 << (type.width - 1));
LLVMValueRef sign;
LLVMValueRef offset;
/* sign = a < 0 ? ~0 : 0 */
sign = LLVMBuildBitCast(bld->builder, a, int_vec_type, "");
sign = LLVMBuildAnd(bld->builder, sign, mask, "");
- sign = LLVMBuildAShr(bld->builder, sign, lp_build_int_const_scalar(type, type.width - 1), "");
+ sign = LLVMBuildAShr(bld->builder, sign, lp_build_const_int_vec(type, type.width - 1), "");
lp_build_name(sign, "floor.sign");
/* offset = -0.99999(9)f */
- offset = lp_build_const_scalar(type, -(double)(((unsigned long long)1 << mantissa) - 1)/((unsigned long long)1 << mantissa));
+ offset = lp_build_const_vec(type, -(double)(((unsigned long long)1 << mantissa) - 1)/((unsigned long long)1 << mantissa));
offset = LLVMConstBitCast(offset, int_vec_type);
/* offset = a < 0 ? -0.99999(9)f : 0.0f */
@@ -1265,7 +1268,7 @@ lp_build_exp(struct lp_build_context *bld,
LLVMValueRef x)
{
/* log2(e) = 1/log(2) */
- LLVMValueRef log2e = lp_build_const_scalar(bld->type, 1.4426950408889634);
+ LLVMValueRef log2e = lp_build_const_vec(bld->type, 1.4426950408889634);
return lp_build_mul(bld, log2e, lp_build_exp2(bld, x));
}
@@ -1279,7 +1282,7 @@ lp_build_log(struct lp_build_context *bld,
LLVMValueRef x)
{
/* log(2) */
- LLVMValueRef log2 = lp_build_const_scalar(bld->type, 0.69314718055994529);
+ LLVMValueRef log2 = lp_build_const_vec(bld->type, 0.69314718055994529);
return lp_build_mul(bld, log2, lp_build_exp2(bld, x));
}
@@ -1315,7 +1318,7 @@ lp_build_polynomial(struct lp_build_context *bld,
if (type.length == 1)
coeff = LLVMConstReal(float_type, coeffs[i]);
else
- coeff = lp_build_const_scalar(type, coeffs[i]);
+ coeff = lp_build_const_vec(type, coeffs[i]);
if(res)
res = lp_build_add(bld, coeff, lp_build_mul(bld, x, res));
@@ -1372,11 +1375,11 @@ lp_build_exp2_approx(struct lp_build_context *bld,
assert(type.floating && type.width == 32);
- x = lp_build_min(bld, x, lp_build_const_scalar(type, 129.0));
- x = lp_build_max(bld, x, lp_build_const_scalar(type, -126.99999));
+ x = lp_build_min(bld, x, lp_build_const_vec(type, 129.0));
+ x = lp_build_max(bld, x, lp_build_const_vec(type, -126.99999));
/* ipart = int(x - 0.5) */
- ipart = LLVMBuildSub(bld->builder, x, lp_build_const_scalar(type, 0.5f), "");
+ ipart = LLVMBuildSub(bld->builder, x, lp_build_const_vec(type, 0.5f), "");
ipart = LLVMBuildFPToSI(bld->builder, ipart, int_vec_type, "");
/* fpart = x - ipart */
@@ -1386,8 +1389,8 @@ lp_build_exp2_approx(struct lp_build_context *bld,
if(p_exp2_int_part || p_exp2) {
/* expipart = (float) (1 << ipart) */
- expipart = LLVMBuildAdd(bld->builder, ipart, lp_build_int_const_scalar(type, 127), "");
- expipart = LLVMBuildShl(bld->builder, expipart, lp_build_int_const_scalar(type, 23), "");
+ expipart = LLVMBuildAdd(bld->builder, ipart, lp_build_const_int_vec(type, 127), "");
+ expipart = LLVMBuildShl(bld->builder, expipart, lp_build_const_int_vec(type, 23), "");
expipart = LLVMBuildBitCast(bld->builder, expipart, vec_type, "");
}
@@ -1453,8 +1456,8 @@ lp_build_log2_approx(struct lp_build_context *bld,
LLVMTypeRef vec_type = lp_build_vec_type(type);
LLVMTypeRef int_vec_type = lp_build_int_vec_type(type);
- LLVMValueRef expmask = lp_build_int_const_scalar(type, 0x7f800000);
- LLVMValueRef mantmask = lp_build_int_const_scalar(type, 0x007fffff);
+ LLVMValueRef expmask = lp_build_const_int_vec(type, 0x7f800000);
+ LLVMValueRef mantmask = lp_build_const_int_vec(type, 0x007fffff);
LLVMValueRef one = LLVMConstBitCast(bld->one, int_vec_type);
LLVMValueRef i = NULL;
@@ -1479,8 +1482,8 @@ lp_build_log2_approx(struct lp_build_context *bld,
}
if(p_floor_log2 || p_log2) {
- logexp = LLVMBuildLShr(bld->builder, exp, lp_build_int_const_scalar(type, 23), "");
- logexp = LLVMBuildSub(bld->builder, logexp, lp_build_int_const_scalar(type, 127), "");
+ logexp = LLVMBuildLShr(bld->builder, exp, lp_build_const_int_vec(type, 23), "");
+ logexp = LLVMBuildSub(bld->builder, logexp, lp_build_const_int_vec(type, 127), "");
logexp = LLVMBuildSIToFP(bld->builder, logexp, vec_type, "");
}
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_arit.h b/src/gallium/auxiliary/gallivm/lp_bld_arit.h
index 7a10fe1220..31efa9921c 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_arit.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_arit.h
@@ -37,7 +37,7 @@
#define LP_BLD_ARIT_H
-#include "os/os_llvm.h"
+#include "gallivm/lp_bld.h"
struct lp_type;
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_blend.h b/src/gallium/auxiliary/gallivm/lp_bld_blend.h
index 5a9e1c1fb2..ebbdb1a604 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_blend.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_blend.h
@@ -40,7 +40,7 @@
* for a standalone example.
*/
-#include "os/os_llvm.h"
+#include "gallivm/lp_bld.h"
#include "pipe/p_format.h"
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_const.c b/src/gallium/auxiliary/gallivm/lp_bld_const.c
index 8a275fa72f..57843e9a60 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_const.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_const.c
@@ -263,7 +263,7 @@ lp_build_one(struct lp_type type)
if(type.sign)
/* TODO: Unfortunately this caused "Tried to create a shift operation
* on a non-integer type!" */
- vec = LLVMConstLShr(vec, lp_build_int_const_scalar(type, 1));
+ vec = LLVMConstLShr(vec, lp_build_const_int_vec(type, 1));
#endif
return vec;
@@ -283,8 +283,8 @@ lp_build_one(struct lp_type type)
* Build constant-valued vector from a scalar value.
*/
LLVMValueRef
-lp_build_const_scalar(struct lp_type type,
- double val)
+lp_build_const_vec(struct lp_type type,
+ double val)
{
LLVMTypeRef elem_type = lp_build_elem_type(type);
LLVMValueRef elems[LP_MAX_VECTOR_LENGTH];
@@ -309,7 +309,7 @@ lp_build_const_scalar(struct lp_type type,
LLVMValueRef
-lp_build_int_const_scalar(struct lp_type type,
+lp_build_const_int_vec(struct lp_type type,
long long val)
{
LLVMTypeRef elem_type = lp_build_int_elem_type(type);
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_const.h b/src/gallium/auxiliary/gallivm/lp_bld_const.h
index 4078636103..9ca2f0664e 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_const.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_const.h
@@ -37,9 +37,9 @@
#define LP_BLD_CONST_H
-#include "os/os_llvm.h"
+#include "pipe/p_compiler.h"
+#include "gallivm/lp_bld.h"
-#include <pipe/p_compiler.h>
struct lp_type;
@@ -85,13 +85,11 @@ lp_build_one(struct lp_type type);
LLVMValueRef
-lp_build_const_scalar(struct lp_type type,
- double val);
+lp_build_const_vec(struct lp_type type, double val);
LLVMValueRef
-lp_build_int_const_scalar(struct lp_type type,
- long long val);
+lp_build_const_int_vec(struct lp_type type, long long val);
LLVMValueRef
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_conv.c b/src/gallium/auxiliary/gallivm/lp_bld_conv.c
index f77cf78721..3f7f2ebde9 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_conv.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_conv.c
@@ -114,13 +114,13 @@ lp_build_clamped_float_to_unsigned_norm(LLVMBuilderRef builder,
scale = (double)mask/ubound;
bias = (double)((unsigned long long)1 << (mantissa - n));
- res = LLVMBuildMul(builder, src, lp_build_const_scalar(src_type, scale), "");
- res = LLVMBuildAdd(builder, res, lp_build_const_scalar(src_type, bias), "");
+ res = LLVMBuildMul(builder, src, lp_build_const_vec(src_type, scale), "");
+ res = LLVMBuildAdd(builder, res, lp_build_const_vec(src_type, bias), "");
res = LLVMBuildBitCast(builder, res, int_vec_type, "");
if(dst_width > n) {
int shift = dst_width - n;
- res = LLVMBuildShl(builder, res, lp_build_int_const_scalar(src_type, shift), "");
+ res = LLVMBuildShl(builder, res, lp_build_const_int_vec(src_type, shift), "");
/* TODO: Fill in the empty lower bits for additional precision? */
/* YES: this fixes progs/trivial/tri-z-eq.c.
@@ -130,21 +130,21 @@ lp_build_clamped_float_to_unsigned_norm(LLVMBuilderRef builder,
#if 0
{
LLVMValueRef msb;
- msb = LLVMBuildLShr(builder, res, lp_build_int_const_scalar(src_type, dst_width - 1), "");
- msb = LLVMBuildShl(builder, msb, lp_build_int_const_scalar(src_type, shift), "");
- msb = LLVMBuildSub(builder, msb, lp_build_int_const_scalar(src_type, 1), "");
+ msb = LLVMBuildLShr(builder, res, lp_build_const_int_vec(src_type, dst_width - 1), "");
+ msb = LLVMBuildShl(builder, msb, lp_build_const_int_vec(src_type, shift), "");
+ msb = LLVMBuildSub(builder, msb, lp_build_const_int_vec(src_type, 1), "");
res = LLVMBuildOr(builder, res, msb, "");
}
#elif 0
while(shift > 0) {
- res = LLVMBuildOr(builder, res, LLVMBuildLShr(builder, res, lp_build_int_const_scalar(src_type, n), ""), "");
+ res = LLVMBuildOr(builder, res, LLVMBuildLShr(builder, res, lp_build_const_int_vec(src_type, n), ""), "");
shift -= n;
n *= 2;
}
#endif
}
else
- res = LLVMBuildAnd(builder, res, lp_build_int_const_scalar(src_type, mask), "");
+ res = LLVMBuildAnd(builder, res, lp_build_const_int_vec(src_type, mask), "");
return res;
}
@@ -183,10 +183,10 @@ lp_build_unsigned_norm_to_float(LLVMBuilderRef builder,
if(src_width > mantissa) {
int shift = src_width - mantissa;
- res = LLVMBuildLShr(builder, res, lp_build_int_const_scalar(dst_type, shift), "");
+ res = LLVMBuildLShr(builder, res, lp_build_const_int_vec(dst_type, shift), "");
}
- bias_ = lp_build_const_scalar(dst_type, bias);
+ bias_ = lp_build_const_vec(dst_type, bias);
res = LLVMBuildOr(builder,
res,
@@ -195,7 +195,7 @@ lp_build_unsigned_norm_to_float(LLVMBuilderRef builder,
res = LLVMBuildBitCast(builder, res, vec_type, "");
res = LLVMBuildSub(builder, res, bias_, "");
- res = LLVMBuildMul(builder, res, lp_build_const_scalar(dst_type, scale), "");
+ res = LLVMBuildMul(builder, res, lp_build_const_vec(dst_type, scale), "");
return res;
}
@@ -251,7 +251,7 @@ lp_build_conv(LLVMBuilderRef builder,
if(dst_min == 0.0)
thres = bld.zero;
else
- thres = lp_build_const_scalar(src_type, dst_min);
+ thres = lp_build_const_vec(src_type, dst_min);
for(i = 0; i < num_tmps; ++i)
tmp[i] = lp_build_max(&bld, tmp[i], thres);
}
@@ -260,7 +260,7 @@ lp_build_conv(LLVMBuilderRef builder,
if(dst_max == 1.0)
thres = bld.one;
else
- thres = lp_build_const_scalar(src_type, dst_max);
+ thres = lp_build_const_vec(src_type, dst_max);
for(i = 0; i < num_tmps; ++i)
tmp[i] = lp_build_min(&bld, tmp[i], thres);
}
@@ -288,7 +288,7 @@ lp_build_conv(LLVMBuilderRef builder,
LLVMTypeRef tmp_vec_type;
if (dst_scale != 1.0) {
- LLVMValueRef scale = lp_build_const_scalar(tmp_type, dst_scale);
+ LLVMValueRef scale = lp_build_const_vec(tmp_type, dst_scale);
for(i = 0; i < num_tmps; ++i)
tmp[i] = LLVMBuildMul(builder, tmp[i], scale, "");
}
@@ -315,7 +315,7 @@ lp_build_conv(LLVMBuilderRef builder,
/* FIXME: compensate different offsets too */
if(src_shift > dst_shift) {
- LLVMValueRef shift = lp_build_int_const_scalar(tmp_type, src_shift - dst_shift);
+ LLVMValueRef shift = lp_build_const_int_vec(tmp_type, src_shift - dst_shift);
for(i = 0; i < num_tmps; ++i)
if(src_type.sign)
tmp[i] = LLVMBuildAShr(builder, tmp[i], shift, "");
@@ -388,7 +388,7 @@ lp_build_conv(LLVMBuilderRef builder,
}
if (src_scale != 1.0) {
- LLVMValueRef scale = lp_build_const_scalar(tmp_type, 1.0/src_scale);
+ LLVMValueRef scale = lp_build_const_vec(tmp_type, 1.0/src_scale);
for(i = 0; i < num_tmps; ++i)
tmp[i] = LLVMBuildMul(builder, tmp[i], scale, "");
}
@@ -400,7 +400,7 @@ lp_build_conv(LLVMBuilderRef builder,
/* FIXME: compensate different offsets too */
if(src_shift < dst_shift) {
- LLVMValueRef shift = lp_build_int_const_scalar(tmp_type, dst_shift - src_shift);
+ LLVMValueRef shift = lp_build_const_int_vec(tmp_type, dst_shift - src_shift);
for(i = 0; i < num_tmps; ++i)
tmp[i] = LLVMBuildShl(builder, tmp[i], shift, "");
}
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_conv.h b/src/gallium/auxiliary/gallivm/lp_bld_conv.h
index 78e8155ff7..628831c3ad 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_conv.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_conv.h
@@ -37,7 +37,7 @@
#define LP_BLD_CONV_H
-#include "os/os_llvm.h"
+#include "gallivm/lp_bld.h"
struct lp_type;
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_debug.h b/src/gallium/auxiliary/gallivm/lp_bld_debug.h
index 441ad94786..7b010cbdb0 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_debug.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_debug.h
@@ -30,7 +30,7 @@
#define LP_BLD_DEBUG_H
-#include "os/os_llvm.h"
+#include "gallivm/lp_bld.h"
#include "pipe/p_compiler.h"
#include "util/u_string.h"
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_depth.c b/src/gallium/auxiliary/gallivm/lp_bld_depth.c
index f08f8eb6d8..4ce1a27a06 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_depth.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_depth.c
@@ -52,7 +52,14 @@
* Z31 Z32 Z41 Z42 Z33 Z34 Z43 Z44 ...
* ... ... ... ... ... ... ... ... ...
*
- * FIXME: Code generate stencil test
+ *
+ * Stencil test:
+ * Two-sided stencil test is supported but probably not as efficient as
+ * it could be. Currently, we use if/then/else constructs to do the
+ * operations for front vs. back-facing polygons. We could probably do
+ * both the front and back arithmetic then use a Select() instruction to
+ * choose the result depending on polyon orientation. We'd have to
+ * measure performance both ways and see which is better.
*
* @author Jose Fonseca <jfonseca@vmware.com>
*/
@@ -61,11 +68,264 @@
#include "util/u_format.h"
#include "lp_bld_type.h"
+#include "lp_bld_arit.h"
#include "lp_bld_const.h"
#include "lp_bld_logic.h"
#include "lp_bld_flow.h"
#include "lp_bld_debug.h"
#include "lp_bld_depth.h"
+#include "lp_bld_swizzle.h"
+
+
+/** Used to select fields from pipe_stencil_state */
+enum stencil_op {
+ S_FAIL_OP,
+ Z_FAIL_OP,
+ Z_PASS_OP
+};
+
+
+
+/**
+ * Do the stencil test comparison (compare FB stencil values against ref value).
+ * This will be used twice when generating two-sided stencil code.
+ * \param stencil the front/back stencil state
+ * \param stencilRef the stencil reference value, replicated as a vector
+ * \param stencilVals vector of stencil values from framebuffer
+ * \return vector mask of pass/fail values (~0 or 0)
+ */
+static LLVMValueRef
+lp_build_stencil_test_single(struct lp_build_context *bld,
+ const struct pipe_stencil_state *stencil,
+ LLVMValueRef stencilRef,
+ LLVMValueRef stencilVals)
+{
+ const unsigned stencilMax = 255; /* XXX fix */
+ struct lp_type type = bld->type;
+ LLVMValueRef res;
+
+ assert(type.sign);
+
+ assert(stencil->enabled);
+
+ if (stencil->valuemask != stencilMax) {
+ /* compute stencilRef = stencilRef & valuemask */
+ LLVMValueRef valuemask = lp_build_const_int_vec(type, stencil->valuemask);
+ stencilRef = LLVMBuildAnd(bld->builder, stencilRef, valuemask, "");
+ /* compute stencilVals = stencilVals & valuemask */
+ stencilVals = LLVMBuildAnd(bld->builder, stencilVals, valuemask, "");
+ }
+
+ res = lp_build_cmp(bld, stencil->func, stencilVals, stencilRef);
+
+ return res;
+}
+
+
+/**
+ * Do the one or two-sided stencil test comparison.
+ * \sa lp_build_stencil_test_single
+ * \param face an integer indicating front (+) or back (-) facing polygon.
+ * If NULL, assume front-facing.
+ */
+static LLVMValueRef
+lp_build_stencil_test(struct lp_build_context *bld,
+ const struct pipe_stencil_state stencil[2],
+ LLVMValueRef stencilRefs[2],
+ LLVMValueRef stencilVals,
+ LLVMValueRef face)
+{
+ LLVMValueRef res;
+
+ assert(stencil[0].enabled);
+
+ if (stencil[1].enabled && face) {
+ /* do two-sided test */
+ struct lp_build_flow_context *flow_ctx;
+ struct lp_build_if_state if_ctx;
+ LLVMValueRef front_facing;
+ LLVMValueRef zero = LLVMConstReal(LLVMFloatType(), 0.0);
+ LLVMValueRef result = bld->undef;
+
+ flow_ctx = lp_build_flow_create(bld->builder);
+ lp_build_flow_scope_begin(flow_ctx);
+
+ lp_build_flow_scope_declare(flow_ctx, &result);
+
+ /* front_facing = face > 0.0 */
+ front_facing = LLVMBuildFCmp(bld->builder, LLVMRealUGT, face, zero, "");
+
+ lp_build_if(&if_ctx, flow_ctx, bld->builder, front_facing);
+ {
+ result = lp_build_stencil_test_single(bld, &stencil[0],
+ stencilRefs[0], stencilVals);
+ }
+ lp_build_else(&if_ctx);
+ {
+ result = lp_build_stencil_test_single(bld, &stencil[1],
+ stencilRefs[1], stencilVals);
+ }
+ lp_build_endif(&if_ctx);
+
+ lp_build_flow_scope_end(flow_ctx);
+ lp_build_flow_destroy(flow_ctx);
+
+ res = result;
+ }
+ else {
+ /* do single-side test */
+ res = lp_build_stencil_test_single(bld, &stencil[0],
+ stencilRefs[0], stencilVals);
+ }
+
+ return res;
+}
+
+
+/**
+ * Apply the stencil operator (add/sub/keep/etc) to the given vector
+ * of stencil values.
+ * \return new stencil values vector
+ */
+static LLVMValueRef
+lp_build_stencil_op_single(struct lp_build_context *bld,
+ const struct pipe_stencil_state *stencil,
+ enum stencil_op op,
+ LLVMValueRef stencilRef,
+ LLVMValueRef stencilVals,
+ LLVMValueRef mask)
+
+{
+ const unsigned stencilMax = 255; /* XXX fix */
+ struct lp_type type = bld->type;
+ LLVMValueRef res;
+ LLVMValueRef max = lp_build_const_int_vec(type, stencilMax);
+ unsigned stencil_op;
+
+ assert(type.sign);
+
+ switch (op) {
+ case S_FAIL_OP:
+ stencil_op = stencil->fail_op;
+ break;
+ case Z_FAIL_OP:
+ stencil_op = stencil->zfail_op;
+ break;
+ case Z_PASS_OP:
+ stencil_op = stencil->zpass_op;
+ break;
+ default:
+ assert(0 && "Invalid stencil_op mode");
+ stencil_op = PIPE_STENCIL_OP_KEEP;
+ }
+
+ switch (stencil_op) {
+ case PIPE_STENCIL_OP_KEEP:
+ res = stencilVals;
+ /* we can return early for this case */
+ return res;
+ case PIPE_STENCIL_OP_ZERO:
+ res = bld->zero;
+ break;
+ case PIPE_STENCIL_OP_REPLACE:
+ res = stencilRef;
+ break;
+ case PIPE_STENCIL_OP_INCR:
+ res = lp_build_add(bld, stencilVals, bld->one);
+ res = lp_build_min(bld, res, max);
+ break;
+ case PIPE_STENCIL_OP_DECR:
+ res = lp_build_sub(bld, stencilVals, bld->one);
+ res = lp_build_max(bld, res, bld->zero);
+ break;
+ case PIPE_STENCIL_OP_INCR_WRAP:
+ res = lp_build_add(bld, stencilVals, bld->one);
+ res = LLVMBuildAnd(bld->builder, res, max, "");
+ break;
+ case PIPE_STENCIL_OP_DECR_WRAP:
+ res = lp_build_sub(bld, stencilVals, bld->one);
+ res = LLVMBuildAnd(bld->builder, res, max, "");
+ break;
+ case PIPE_STENCIL_OP_INVERT:
+ res = LLVMBuildNot(bld->builder, stencilVals, "");
+ res = LLVMBuildAnd(bld->builder, res, max, "");
+ break;
+ default:
+ assert(0 && "bad stencil op mode");
+ res = NULL;
+ }
+
+ if (stencil->writemask != stencilMax) {
+ /* compute res = (res & mask) | (stencilVals & ~mask) */
+ LLVMValueRef mask = lp_build_const_int_vec(type, stencil->writemask);
+ LLVMValueRef cmask = LLVMBuildNot(bld->builder, mask, "notWritemask");
+ LLVMValueRef t1 = LLVMBuildAnd(bld->builder, res, mask, "t1");
+ LLVMValueRef t2 = LLVMBuildAnd(bld->builder, stencilVals, cmask, "t2");
+ res = LLVMBuildOr(bld->builder, t1, t2, "t1_or_t2");
+ }
+
+ /* only the update the vector elements enabled by 'mask' */
+ res = lp_build_select(bld, mask, res, stencilVals);
+
+ return res;
+}
+
+
+/**
+ * Do the one or two-sided stencil test op/update.
+ */
+static LLVMValueRef
+lp_build_stencil_op(struct lp_build_context *bld,
+ const struct pipe_stencil_state stencil[2],
+ enum stencil_op op,
+ LLVMValueRef stencilRefs[2],
+ LLVMValueRef stencilVals,
+ LLVMValueRef mask,
+ LLVMValueRef face)
+
+{
+ assert(stencil[0].enabled);
+
+ if (stencil[1].enabled && face) {
+ /* do two-sided op */
+ struct lp_build_flow_context *flow_ctx;
+ struct lp_build_if_state if_ctx;
+ LLVMValueRef front_facing;
+ LLVMValueRef zero = LLVMConstReal(LLVMFloatType(), 0.0);
+ LLVMValueRef result = bld->undef;
+
+ flow_ctx = lp_build_flow_create(bld->builder);
+ lp_build_flow_scope_begin(flow_ctx);
+
+ lp_build_flow_scope_declare(flow_ctx, &result);
+
+ /* front_facing = face > 0.0 */
+ front_facing = LLVMBuildFCmp(bld->builder, LLVMRealUGT, face, zero, "");
+
+ lp_build_if(&if_ctx, flow_ctx, bld->builder, front_facing);
+ {
+ result = lp_build_stencil_op_single(bld, &stencil[0], op,
+ stencilRefs[0], stencilVals, mask);
+ }
+ lp_build_else(&if_ctx);
+ {
+ result = lp_build_stencil_op_single(bld, &stencil[1], op,
+ stencilRefs[1], stencilVals, mask);
+ }
+ lp_build_endif(&if_ctx);
+
+ lp_build_flow_scope_end(flow_ctx);
+ lp_build_flow_destroy(flow_ctx);
+
+ return result;
+ }
+ else {
+ /* do single-sided op */
+ return lp_build_stencil_op_single(bld, &stencil[0], op,
+ stencilRefs[0], stencilVals, mask);
+ }
+}
+
/**
@@ -109,105 +369,303 @@ lp_depth_type(const struct util_format_description *format_desc,
/**
- * Depth test.
+ * Compute bitmask and bit shift to apply to the incoming fragment Z values
+ * and the Z buffer values needed before doing the Z comparison.
+ *
+ * Note that we leave the Z bits in the position that we find them
+ * in the Z buffer (typically 0xffffff00 or 0x00ffffff). That lets us
+ * get by with fewer bit twiddling steps.
*/
-void
-lp_build_depth_test(LLVMBuilderRef builder,
- const struct pipe_depth_state *state,
- struct lp_type type,
- const struct util_format_description *format_desc,
- struct lp_build_mask_context *mask,
- LLVMValueRef src,
- LLVMValueRef dst_ptr)
+static boolean
+get_z_shift_and_mask(const struct util_format_description *format_desc,
+ unsigned *shift, unsigned *mask)
{
- struct lp_build_context bld;
+ const unsigned total_bits = format_desc->block.bits;
unsigned z_swizzle;
- LLVMValueRef dst;
- LLVMValueRef z_bitmask = NULL;
- LLVMValueRef test;
-
- if(!state->enabled)
- return;
-
+ int chan;
+ unsigned padding_left, padding_right;
+
assert(format_desc->colorspace == UTIL_FORMAT_COLORSPACE_ZS);
assert(format_desc->block.width == 1);
assert(format_desc->block.height == 1);
z_swizzle = format_desc->swizzle[0];
- if(z_swizzle == UTIL_FORMAT_SWIZZLE_NONE)
- return;
- /* Sanity checking */
- assert(z_swizzle < 4);
- assert(format_desc->block.bits == type.width);
- if(type.floating) {
- assert(z_swizzle == 0);
- assert(format_desc->channel[z_swizzle].type == UTIL_FORMAT_TYPE_FLOAT);
- assert(format_desc->channel[z_swizzle].size == format_desc->block.bits);
+ if (z_swizzle == UTIL_FORMAT_SWIZZLE_NONE)
+ return FALSE;
+
+ padding_right = 0;
+ for (chan = 0; chan < z_swizzle; ++chan)
+ padding_right += format_desc->channel[chan].size;
+
+ padding_left =
+ total_bits - (padding_right + format_desc->channel[z_swizzle].size);
+
+ if (padding_left || padding_right) {
+ unsigned long long mask_left = (1ULL << (total_bits - padding_left)) - 1;
+ unsigned long long mask_right = (1ULL << (padding_right)) - 1;
+ *mask = mask_left ^ mask_right;
}
else {
- assert(format_desc->channel[z_swizzle].type == UTIL_FORMAT_TYPE_UNSIGNED);
- assert(format_desc->channel[z_swizzle].normalized);
- assert(!type.fixed);
- assert(!type.sign);
- assert(type.norm);
+ *mask = 0xffffffff;
+ }
+
+ *shift = padding_left;
+
+ return TRUE;
+}
+
+
+/**
+ * Compute bitmask and bit shift to apply to the framebuffer pixel values
+ * to put the stencil bits in the least significant position.
+ * (i.e. 0x000000ff)
+ */
+static boolean
+get_s_shift_and_mask(const struct util_format_description *format_desc,
+ unsigned *shift, unsigned *mask)
+{
+ unsigned s_swizzle;
+ int chan, sz;
+
+ s_swizzle = format_desc->swizzle[1];
+
+ if (s_swizzle == UTIL_FORMAT_SWIZZLE_NONE)
+ return FALSE;
+
+ *shift = 0;
+ for (chan = 0; chan < s_swizzle; chan++)
+ *shift += format_desc->channel[chan].size;
+
+ sz = format_desc->channel[s_swizzle].size;
+ *mask = (1U << sz) - 1U;
+
+ return TRUE;
+}
+
+
+
+/**
+ * Generate code for performing depth and/or stencil tests.
+ * We operate on a vector of values (typically a 2x2 quad).
+ *
+ * \param depth the depth test state
+ * \param stencil the front/back stencil state
+ * \param type the data type of the fragment depth/stencil values
+ * \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 zs_dst_ptr pointer to depth/stencil values in framebuffer
+ * \param facing contains float value indicating front/back facing polygon
+ */
+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,
+ const struct util_format_description *format_desc,
+ struct lp_build_mask_context *mask,
+ LLVMValueRef stencil_refs[2],
+ LLVMValueRef z_src,
+ LLVMValueRef zs_dst_ptr,
+ LLVMValueRef face)
+{
+ struct lp_build_context bld;
+ struct lp_build_context sbld;
+ struct lp_type s_type;
+ LLVMValueRef zs_dst, z_dst = NULL;
+ LLVMValueRef stencil_vals = NULL;
+ LLVMValueRef z_bitmask = NULL, stencil_shift = NULL;
+ LLVMValueRef z_pass = NULL, s_pass_mask = NULL;
+ LLVMValueRef orig_mask = mask->value;
+
+ /* Sanity checking */
+ {
+ const unsigned z_swizzle = format_desc->swizzle[0];
+ const unsigned s_swizzle = format_desc->swizzle[1];
+
+ assert(z_swizzle != UTIL_FORMAT_SWIZZLE_NONE ||
+ s_swizzle != UTIL_FORMAT_SWIZZLE_NONE);
+
+ assert(depth->enabled || stencil[0].enabled);
+
+ assert(format_desc->colorspace == UTIL_FORMAT_COLORSPACE_ZS);
+ assert(format_desc->block.width == 1);
+ assert(format_desc->block.height == 1);
+
+ if (stencil[0].enabled) {
+ assert(format_desc->format == PIPE_FORMAT_Z24S8_UNORM ||
+ format_desc->format == PIPE_FORMAT_S8Z24_UNORM);
+ }
+
+ assert(z_swizzle < 4);
+ assert(format_desc->block.bits == type.width);
+ if (type.floating) {
+ assert(z_swizzle == 0);
+ assert(format_desc->channel[z_swizzle].type ==
+ UTIL_FORMAT_TYPE_FLOAT);
+ assert(format_desc->channel[z_swizzle].size ==
+ format_desc->block.bits);
+ }
+ else {
+ assert(format_desc->channel[z_swizzle].type ==
+ UTIL_FORMAT_TYPE_UNSIGNED);
+ assert(format_desc->channel[z_swizzle].normalized);
+ assert(!type.fixed);
+ assert(!type.sign);
+ assert(type.norm);
+ }
}
- /* Setup build context */
+
+ /* Setup build context for Z vals */
lp_build_context_init(&bld, builder, type);
- dst = LLVMBuildLoad(builder, dst_ptr, "");
+ /* Setup build context for stencil vals */
+ s_type = lp_type_int_vec(type.width);
+ lp_build_context_init(&sbld, builder, s_type);
+
+ /* Load current z/stencil value from z/stencil buffer */
+ zs_dst = LLVMBuildLoad(builder, zs_dst_ptr, "");
+
+ lp_build_name(zs_dst, "zsbufval");
- lp_build_name(dst, "zsbuf");
- /* Align the source depth bits with the destination's, and mask out any
- * stencil or padding bits from both */
- if(format_desc->channel[z_swizzle].size == format_desc->block.bits) {
- assert(z_swizzle == 0);
- /* nothing to do */
+ /* Compute and apply the Z/stencil bitmasks and shifts.
+ */
+ {
+ unsigned z_shift, z_mask;
+ unsigned s_shift, s_mask;
+
+ if (get_z_shift_and_mask(format_desc, &z_shift, &z_mask)) {
+ if (z_shift) {
+ LLVMValueRef shift = lp_build_const_int_vec(type, z_shift);
+ z_src = LLVMBuildLShr(builder, z_src, shift, "");
+ }
+
+ if (z_mask != 0xffffffff) {
+ LLVMValueRef mask = lp_build_const_int_vec(type, z_mask);
+ z_src = LLVMBuildAnd(builder, z_src, mask, "");
+ z_dst = LLVMBuildAnd(builder, zs_dst, mask, "");
+ z_bitmask = mask; /* used below */
+ }
+ else {
+ z_dst = zs_dst;
+ }
+
+ lp_build_name(z_dst, "zsbuf.z");
+ }
+
+ if (get_s_shift_and_mask(format_desc, &s_shift, &s_mask)) {
+ if (s_shift) {
+ LLVMValueRef shift = lp_build_const_int_vec(type, s_shift);
+ stencil_vals = LLVMBuildLShr(builder, zs_dst, shift, "");
+ stencil_shift = shift; /* used below */
+ }
+ else {
+ stencil_vals = zs_dst;
+ }
+
+ if (s_mask != 0xffffffff) {
+ LLVMValueRef mask = lp_build_const_int_vec(type, s_mask);
+ stencil_vals = LLVMBuildAnd(builder, stencil_vals, mask, "");
+ }
+
+ lp_build_name(stencil_vals, "stencil");
+ }
}
- else {
- unsigned padding_left;
- unsigned padding_right;
- unsigned chan;
-
- assert(format_desc->layout == UTIL_FORMAT_LAYOUT_PLAIN);
- assert(format_desc->channel[z_swizzle].type == UTIL_FORMAT_TYPE_UNSIGNED);
- assert(format_desc->channel[z_swizzle].size <= format_desc->block.bits);
- assert(format_desc->channel[z_swizzle].normalized);
-
- padding_right = 0;
- for(chan = 0; chan < z_swizzle; ++chan)
- padding_right += format_desc->channel[chan].size;
- padding_left = format_desc->block.bits -
- (padding_right + format_desc->channel[z_swizzle].size);
-
- if(padding_left || padding_right) {
- const unsigned long long mask_left = ((unsigned long long)1 << (format_desc->block.bits - padding_left)) - 1;
- const unsigned long long mask_right = ((unsigned long long)1 << (padding_right)) - 1;
- z_bitmask = lp_build_int_const_scalar(type, mask_left ^ mask_right);
+
+
+ if (stencil[0].enabled) {
+ /* convert scalar stencil refs into vectors */
+ stencil_refs[0] = lp_build_broadcast_scalar(&bld, stencil_refs[0]);
+ stencil_refs[1] = lp_build_broadcast_scalar(&bld, stencil_refs[1]);
+
+ s_pass_mask = lp_build_stencil_test(&sbld, stencil,
+ stencil_refs, stencil_vals, face);
+
+ /* apply stencil-fail operator */
+ {
+ LLVMValueRef s_fail_mask = lp_build_andc(&bld, orig_mask, s_pass_mask);
+ stencil_vals = lp_build_stencil_op(&sbld, stencil, S_FAIL_OP,
+ stencil_refs, stencil_vals,
+ s_fail_mask, face);
+ }
+ }
+
+ if (depth->enabled) {
+ /* compare src Z to dst Z, returning 'pass' mask */
+ z_pass = lp_build_cmp(&bld, depth->func, z_src, z_dst);
+
+ if (!stencil[0].enabled) {
+ /* We can potentially skip all remaining operations here, but only
+ * if stencil is disabled because we still need to update the stencil
+ * buffer values. Don't need to update Z buffer values.
+ */
+ lp_build_mask_update(mask, z_pass);
+ }
+
+ if (depth->writemask) {
+ if(z_bitmask)
+ z_bitmask = LLVMBuildAnd(builder, mask->value, z_bitmask, "");
+ else
+ z_bitmask = mask->value;
+
+ z_dst = lp_build_select(&bld, z_bitmask, z_src, z_dst);
}
- if(padding_left)
- src = LLVMBuildLShr(builder, src, lp_build_int_const_scalar(type, padding_left), "");
- if(padding_right)
- src = LLVMBuildAnd(builder, src, z_bitmask, "");
- if(padding_left || padding_right)
- dst = LLVMBuildAnd(builder, dst, z_bitmask, "");
+ if (stencil[0].enabled) {
+ /* update stencil buffer values according to z pass/fail result */
+ LLVMValueRef z_fail_mask, z_pass_mask;
+
+ /* apply Z-fail operator */
+ z_fail_mask = lp_build_andc(&bld, orig_mask, z_pass);
+ stencil_vals = lp_build_stencil_op(&sbld, stencil, Z_FAIL_OP,
+ stencil_refs, stencil_vals,
+ z_fail_mask, face);
+
+ /* apply Z-pass operator */
+ z_pass_mask = LLVMBuildAnd(bld.builder, orig_mask, z_pass, "");
+ stencil_vals = lp_build_stencil_op(&sbld, stencil, Z_PASS_OP,
+ stencil_refs, stencil_vals,
+ z_pass_mask, face);
+ }
+ }
+ else {
+ /* No depth test: apply Z-pass operator to stencil buffer values which
+ * passed the stencil test.
+ */
+ s_pass_mask = LLVMBuildAnd(bld.builder, orig_mask, s_pass_mask, "");
+ stencil_vals = lp_build_stencil_op(&sbld, stencil, Z_PASS_OP,
+ stencil_refs, stencil_vals,
+ s_pass_mask, face);
}
- lp_build_name(dst, "zsbuf.z");
+ /* The Z bits are already in the right place but we may need to shift the
+ * stencil bits before ORing Z with Stencil to make the final pixel value.
+ */
+ if (stencil_vals && stencil_shift)
+ stencil_vals = LLVMBuildShl(bld.builder, stencil_vals,
+ stencil_shift, "");
- test = lp_build_cmp(&bld, state->func, src, dst);
- lp_build_mask_update(mask, test);
+ /* Finally, merge/store the z/stencil values */
+ if ((depth->enabled && depth->writemask) ||
+ (stencil[0].enabled && stencil[0].writemask)) {
- if(state->writemask) {
- if(z_bitmask)
- z_bitmask = LLVMBuildAnd(builder, mask->value, z_bitmask, "");
+ if (z_dst && stencil_vals)
+ zs_dst = LLVMBuildOr(bld.builder, z_dst, stencil_vals, "");
+ else if (z_dst)
+ zs_dst = z_dst;
else
- z_bitmask = mask->value;
+ zs_dst = stencil_vals;
- dst = lp_build_select(&bld, z_bitmask, src, dst);
- LLVMBuildStore(builder, dst, dst_ptr);
+ LLVMBuildStore(builder, zs_dst, zs_dst_ptr);
}
+
+ if (s_pass_mask)
+ lp_build_mask_update(mask, s_pass_mask);
+
+ if (depth->enabled && stencil[0].enabled)
+ lp_build_mask_update(mask, z_pass);
}
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_depth.h b/src/gallium/auxiliary/gallivm/lp_bld_depth.h
index 8be80024ae..27dd46b625 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_depth.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_depth.h
@@ -36,7 +36,7 @@
#define LP_BLD_DEPTH_H
-#include "os/os_llvm.h"
+#include "gallivm/lp_bld.h"
struct pipe_depth_state;
@@ -51,13 +51,16 @@ lp_depth_type(const struct util_format_description *format_desc,
void
-lp_build_depth_test(LLVMBuilderRef builder,
- const struct pipe_depth_state *state,
- struct lp_type type,
- const struct util_format_description *format_desc,
- struct lp_build_mask_context *mask,
- LLVMValueRef src,
- LLVMValueRef dst_ptr);
+lp_build_depth_stencil_test(LLVMBuilderRef builder,
+ const struct pipe_depth_state *depth,
+ const struct pipe_stencil_state stencil[2],
+ struct lp_type type,
+ const struct util_format_description *format_desc,
+ struct lp_build_mask_context *mask,
+ LLVMValueRef stencil_refs[2],
+ LLVMValueRef zs_src,
+ LLVMValueRef zs_dst_ptr,
+ LLVMValueRef facing);
#endif /* !LP_BLD_DEPTH_H */
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_flow.h b/src/gallium/auxiliary/gallivm/lp_bld_flow.h
index e158836549..c2b50e1b60 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_flow.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_flow.h
@@ -35,7 +35,7 @@
#define LP_BLD_FLOW_H
-#include "os/os_llvm.h"
+#include "gallivm/lp_bld.h"
struct lp_type;
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_format.h b/src/gallium/auxiliary/gallivm/lp_bld_format.h
index 8972c0dc17..73ab6de3f2 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_format.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_format.h
@@ -34,7 +34,7 @@
* Pixel format helpers.
*/
-#include "os/os_llvm.h"
+#include "gallivm/lp_bld.h"
#include "pipe/p_format.h"
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_format_soa.c b/src/gallium/auxiliary/gallivm/lp_bld_format_soa.c
index abb27e4c32..45ee4b12ce 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_format_soa.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_format_soa.c
@@ -114,10 +114,10 @@ lp_build_unpack_rgba_soa(LLVMBuilderRef builder,
case UTIL_FORMAT_TYPE_UNSIGNED:
if(type.floating) {
if(start)
- input = LLVMBuildLShr(builder, input, lp_build_int_const_scalar(type, start), "");
+ input = LLVMBuildLShr(builder, input, lp_build_const_int_vec(type, start), "");
if(stop < format_desc->block.bits) {
unsigned mask = ((unsigned long long)1 << width) - 1;
- input = LLVMBuildAnd(builder, input, lp_build_int_const_scalar(type, mask), "");
+ input = LLVMBuildAnd(builder, input, lp_build_const_int_vec(type, mask), "");
}
if(format_desc->channel[chan].normalized)
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_interp.c b/src/gallium/auxiliary/gallivm/lp_bld_interp.c
index 2fc894017d..09efb16121 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_interp.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_interp.c
@@ -289,17 +289,17 @@ pos_update(struct lp_build_interp_soa_context *bld, int quad_index)
/* top-right or bottom-right quad in block */
/* build x += xstep */
x = lp_build_add(&bld->base, x,
- lp_build_const_scalar(bld->base.type, xstep));
+ lp_build_const_vec(bld->base.type, xstep));
}
if (quad_index == 2) {
/* bottom-left quad in block */
/* build y += ystep */
y = lp_build_add(&bld->base, y,
- lp_build_const_scalar(bld->base.type, ystep));
+ lp_build_const_vec(bld->base.type, ystep));
/* build x -= xstep */
x = lp_build_sub(&bld->base, x,
- lp_build_const_scalar(bld->base.type, xstep));
+ lp_build_const_vec(bld->base.type, xstep));
}
lp_build_name(x, "pos.x");
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_interp.h b/src/gallium/auxiliary/gallivm/lp_bld_interp.h
index 177b5e943e..a4937bbb04 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_interp.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_interp.h
@@ -41,7 +41,7 @@
#define LP_BLD_INTERP_H
-#include "os/os_llvm.h"
+#include "gallivm/lp_bld.h"
#include "tgsi/tgsi_exec.h"
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_intr.h b/src/gallium/auxiliary/gallivm/lp_bld_intr.h
index 7d5506c733..977f767322 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_intr.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_intr.h
@@ -37,7 +37,7 @@
#define LP_BLD_INTR_H
-#include "os/os_llvm.h"
+#include "gallivm/lp_bld.h"
/**
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_logic.c b/src/gallium/auxiliary/gallivm/lp_bld_logic.c
index f3df3dd138..a3b6970116 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_logic.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_logic.c
@@ -193,7 +193,7 @@ lp_build_compare(LLVMBuilderRef builder,
if(table[func].gt &&
((type.width == 8 && type.sign) ||
(type.width != 8 && !type.sign))) {
- LLVMValueRef msb = lp_build_int_const_scalar(type, (unsigned long long)1 << (type.width - 1));
+ LLVMValueRef msb = lp_build_const_int_vec(type, (unsigned long long)1 << (type.width - 1));
a = LLVMBuildXor(builder, a, msb, "");
b = LLVMBuildXor(builder, b, msb, "");
}
@@ -483,3 +483,13 @@ lp_build_alloca(struct lp_build_context *bld)
return LLVMBuildAlloca(bld->builder, lp_build_elem_type(type), "");
}
}
+
+
+/** Return (a & ~b) */
+LLVMValueRef
+lp_build_andc(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b)
+{
+ b = LLVMBuildNot(bld->builder, b, "");
+ b = LLVMBuildAnd(bld->builder, a, b, "");
+ return b;
+}
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_logic.h b/src/gallium/auxiliary/gallivm/lp_bld_logic.h
index b54ec13b70..00a8c75019 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_logic.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_logic.h
@@ -37,7 +37,7 @@
#define LP_BLD_LOGIC_H
-#include "os/os_llvm.h"
+#include "gallivm/lp_bld.h"
#include "pipe/p_defines.h" /* For PIPE_FUNC_xxx */
@@ -79,4 +79,9 @@ lp_build_select_aos(struct lp_build_context *bld,
LLVMValueRef
lp_build_alloca(struct lp_build_context *bld);
+
+LLVMValueRef
+lp_build_andc(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b);
+
+
#endif /* !LP_BLD_LOGIC_H */
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_pack.c b/src/gallium/auxiliary/gallivm/lp_bld_pack.c
index 23398f41f9..2daa8a3b58 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_pack.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_pack.c
@@ -164,7 +164,7 @@ lp_build_unpack2(LLVMBuilderRef builder,
if(dst_type.sign && src_type.sign) {
/* Replicate the sign bit in the most significant bits */
- msb = LLVMBuildAShr(builder, src, lp_build_int_const_scalar(src_type, src_type.width - 1), "");
+ msb = LLVMBuildAShr(builder, src, lp_build_const_int_vec(src_type, src_type.width - 1), "");
}
else
/* Most significant bits always zero */
@@ -361,7 +361,7 @@ lp_build_packs2(LLVMBuilderRef builder,
if(clamp) {
struct lp_build_context bld;
unsigned dst_bits = dst_type.sign ? dst_type.width - 1 : dst_type.width;
- LLVMValueRef dst_max = lp_build_int_const_scalar(src_type, ((unsigned long long)1 << dst_bits) - 1);
+ LLVMValueRef dst_max = lp_build_const_int_vec(src_type, ((unsigned long long)1 << dst_bits) - 1);
lp_build_context_init(&bld, builder, src_type);
lo = lp_build_min(&bld, lo, dst_max);
hi = lp_build_min(&bld, hi, dst_max);
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_pack.h b/src/gallium/auxiliary/gallivm/lp_bld_pack.h
index 346a17d580..41adeed220 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_pack.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_pack.h
@@ -37,7 +37,7 @@
#define LP_BLD_PACK_H
-#include "os/os_llvm.h"
+#include "gallivm/lp_bld.h"
struct lp_type;
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_sample.c b/src/gallium/auxiliary/gallivm/lp_bld_sample.c
index 672e06c992..0761e35f8e 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_sample.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_sample.c
@@ -173,7 +173,7 @@ lp_build_sample_offset(struct lp_build_context *bld,
LLVMValueRef x_stride;
LLVMValueRef offset;
- x_stride = lp_build_const_scalar(bld->type, format_desc->block.bits/8);
+ x_stride = lp_build_const_vec(bld->type, format_desc->block.bits/8);
if(format_desc->colorspace == UTIL_FORMAT_COLORSPACE_ZS) {
LLVMValueRef x_lo, x_hi;
@@ -195,9 +195,9 @@ lp_build_sample_offset(struct lp_build_context *bld,
y_hi = LLVMBuildLShr(bld->builder, y, bld->one, "");
x_stride_lo = x_stride;
- y_stride_lo = lp_build_const_scalar(bld->type, 2*format_desc->block.bits/8);
+ y_stride_lo = lp_build_const_vec(bld->type, 2*format_desc->block.bits/8);
- x_stride_hi = lp_build_const_scalar(bld->type, 4*format_desc->block.bits/8);
+ x_stride_hi = lp_build_const_vec(bld->type, 4*format_desc->block.bits/8);
y_stride_hi = LLVMBuildShl(bld->builder, y_stride, bld->one, "");
x_offset_lo = lp_build_mul(bld, x_lo, x_stride_lo);
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_sample.h b/src/gallium/auxiliary/gallivm/lp_bld_sample.h
index d35fdbb78d..fcbf084baf 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_sample.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_sample.h
@@ -36,7 +36,7 @@
#define LP_BLD_SAMPLE_H
-#include "os/os_llvm.h"
+#include "gallivm/lp_bld.h"
struct pipe_resource;
struct pipe_sampler_state;
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_sample_soa.c b/src/gallium/auxiliary/gallivm/lp_bld_sample_soa.c
index 72018737a8..995c016b9d 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_sample_soa.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_sample_soa.c
@@ -292,7 +292,7 @@ lp_build_sample_texel_soa(struct lp_build_sample_context *bld,
int chan;
for (chan = 0; chan < 4; chan++) {
LLVMValueRef border_chan =
- lp_build_const_scalar(bld->texel_type,
+ lp_build_const_vec(bld->texel_type,
bld->static_state->border_color[chan]);
texel[chan] = lp_build_select(&bld->texel_bld, use_border,
border_chan, texel[chan]);
@@ -457,8 +457,8 @@ 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 two = lp_build_const_scalar(coord_bld->type, 2.0);
- LLVMValueRef half = lp_build_const_scalar(coord_bld->type, 0.5);
+ LLVMValueRef two = lp_build_const_vec(coord_bld->type, 2.0);
+ 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_f_minus_one = lp_build_sub(coord_bld, length_f, coord_bld->one);
@@ -512,7 +512,7 @@ lp_build_sample_wrap_linear(struct lp_build_sample_context *bld,
else {
LLVMValueRef min, max;
/* clamp to [0.5, length - 0.5] */
- min = lp_build_const_scalar(coord_bld->type, 0.5F);
+ 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);
}
@@ -533,7 +533,7 @@ lp_build_sample_wrap_linear(struct lp_build_sample_context *bld,
if (bld->static_state->normalized_coords) {
/* min = -1.0 / (2 * length) = -0.5 / length */
min = lp_build_mul(coord_bld,
- lp_build_const_scalar(coord_bld->type, -0.5F),
+ lp_build_const_vec(coord_bld->type, -0.5F),
lp_build_rcp(coord_bld, length_f));
/* max = 1.0 - min */
max = lp_build_sub(coord_bld, coord_bld->one, min);
@@ -545,7 +545,7 @@ lp_build_sample_wrap_linear(struct lp_build_sample_context *bld,
}
else {
/* clamp to [-0.5, length + 0.5] */
- min = lp_build_const_scalar(coord_bld->type, -0.5F);
+ 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);
coord = lp_build_sub(coord_bld, coord, half);
@@ -620,7 +620,7 @@ lp_build_sample_wrap_linear(struct lp_build_sample_context *bld,
LLVMValueRef min, max;
/* min = -1.0 / (2 * length) = -0.5 / length */
min = lp_build_mul(coord_bld,
- lp_build_const_scalar(coord_bld->type, -0.5F),
+ lp_build_const_vec(coord_bld->type, -0.5F),
lp_build_rcp(coord_bld, length_f));
/* max = 1.0 - min */
max = lp_build_sub(coord_bld, coord_bld->one, min);
@@ -665,7 +665,7 @@ 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 two = lp_build_const_scalar(coord_bld->type, 2.0);
+ LLVMValueRef two = lp_build_const_vec(coord_bld->type, 2.0);
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_f_minus_one = lp_build_sub(coord_bld, length_f, coord_bld->one);
@@ -708,7 +708,7 @@ lp_build_sample_wrap_nearest(struct lp_build_sample_context *bld,
}
else {
/* clamp to [0.5, length - 0.5] */
- min = lp_build_const_scalar(coord_bld->type, 0.5F);
+ min = lp_build_const_vec(coord_bld->type, 0.5F);
max = lp_build_sub(coord_bld, length_f, min);
}
/* coord = clamp(coord, min, max) */
@@ -724,7 +724,7 @@ lp_build_sample_wrap_nearest(struct lp_build_sample_context *bld,
if (bld->static_state->normalized_coords) {
/* min = -1.0 / (2 * length) = -0.5 / length */
min = lp_build_mul(coord_bld,
- lp_build_const_scalar(coord_bld->type, -0.5F),
+ lp_build_const_vec(coord_bld->type, -0.5F),
lp_build_rcp(coord_bld, length_f));
/* max = length - min */
max = lp_build_sub(coord_bld, length_f, min);
@@ -733,7 +733,7 @@ lp_build_sample_wrap_nearest(struct lp_build_sample_context *bld,
}
else {
/* clamp to [-0.5, length + 0.5] */
- min = lp_build_const_scalar(coord_bld->type, -0.5F);
+ min = lp_build_const_vec(coord_bld->type, -0.5F);
max = lp_build_sub(coord_bld, length_f, min);
}
/* coord = clamp(coord, min, max) */
@@ -843,87 +843,98 @@ lp_build_lod_selector(struct lp_build_sample_context *bld,
LLVMValueRef depth)
{
- const int dims = texture_dims(bld->static_state->target);
- struct lp_build_context *float_bld = &bld->float_bld;
- LLVMValueRef 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 index1 = LLVMConstInt(LLVMInt32Type(), 1, 0);
- LLVMValueRef index2 = LLVMConstInt(LLVMInt32Type(), 2, 0);
-
- LLVMValueRef s0, s1, s2;
- LLVMValueRef t0, t1, t2;
- LLVMValueRef r0, r1, r2;
- LLVMValueRef dsdx, dsdy, dtdx, dtdy, drdx, drdy;
- LLVMValueRef rho, lod;
-
- /*
- * 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]);
- * XXX we're assuming a four-element quad in 2x2 layout here.
- */
- s0 = LLVMBuildExtractElement(bld->builder, s, index0, "s0");
- s1 = LLVMBuildExtractElement(bld->builder, s, index1, "s1");
- s2 = LLVMBuildExtractElement(bld->builder, s, index2, "s2");
- dsdx = LLVMBuildSub(bld->builder, s1, s0, "");
- dsdx = lp_build_abs(float_bld, dsdx);
- dsdy = LLVMBuildSub(bld->builder, s2, s0, "");
- dsdy = lp_build_abs(float_bld, dsdy);
- if (dims > 1) {
- t0 = LLVMBuildExtractElement(bld->builder, t, index0, "t0");
- t1 = LLVMBuildExtractElement(bld->builder, t, index1, "t1");
- t2 = LLVMBuildExtractElement(bld->builder, t, index2, "t2");
- dtdx = LLVMBuildSub(bld->builder, t1, t0, "");
- dtdx = lp_build_abs(float_bld, dtdx);
- dtdy = LLVMBuildSub(bld->builder, t2, t0, "");
- dtdy = lp_build_abs(float_bld, dtdy);
- if (dims > 2) {
- r0 = LLVMBuildExtractElement(bld->builder, r, index0, "r0");
- r1 = LLVMBuildExtractElement(bld->builder, r, index1, "r1");
- r2 = LLVMBuildExtractElement(bld->builder, r, index2, "r2");
- drdx = LLVMBuildSub(bld->builder, r1, r0, "");
- drdx = lp_build_abs(float_bld, drdx);
- drdy = LLVMBuildSub(bld->builder, r2, r0, "");
- drdy = lp_build_abs(float_bld, drdy);
- }
+ 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 {
+ const int dims = texture_dims(bld->static_state->target);
+ struct lp_build_context *float_bld = &bld->float_bld;
+ LLVMValueRef 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 index1 = LLVMConstInt(LLVMInt32Type(), 1, 0);
+ LLVMValueRef index2 = LLVMConstInt(LLVMInt32Type(), 2, 0);
+
+ LLVMValueRef s0, s1, s2;
+ LLVMValueRef t0, t1, t2;
+ LLVMValueRef r0, r1, r2;
+ LLVMValueRef dsdx, dsdy, dtdx, dtdy, drdx, drdy;
+ LLVMValueRef rho, lod;
+
+ /*
+ * 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]);
+ * XXX we're assuming a four-element quad in 2x2 layout here.
+ */
+ s0 = LLVMBuildExtractElement(bld->builder, s, index0, "s0");
+ s1 = LLVMBuildExtractElement(bld->builder, s, index1, "s1");
+ s2 = LLVMBuildExtractElement(bld->builder, s, index2, "s2");
+ dsdx = LLVMBuildSub(bld->builder, s1, s0, "");
+ dsdx = lp_build_abs(float_bld, dsdx);
+ dsdy = LLVMBuildSub(bld->builder, s2, s0, "");
+ dsdy = lp_build_abs(float_bld, dsdy);
+ if (dims > 1) {
+ t0 = LLVMBuildExtractElement(bld->builder, t, index0, "t0");
+ t1 = LLVMBuildExtractElement(bld->builder, t, index1, "t1");
+ t2 = LLVMBuildExtractElement(bld->builder, t, index2, "t2");
+ dtdx = LLVMBuildSub(bld->builder, t1, t0, "");
+ dtdx = lp_build_abs(float_bld, dtdx);
+ dtdy = LLVMBuildSub(bld->builder, t2, t0, "");
+ dtdy = lp_build_abs(float_bld, dtdy);
+ if (dims > 2) {
+ r0 = LLVMBuildExtractElement(bld->builder, r, index0, "r0");
+ r1 = LLVMBuildExtractElement(bld->builder, r, index1, "r1");
+ r2 = LLVMBuildExtractElement(bld->builder, r, index2, "r2");
+ drdx = LLVMBuildSub(bld->builder, r1, r0, "");
+ drdx = lp_build_abs(float_bld, drdx);
+ drdy = LLVMBuildSub(bld->builder, r2, r0, "");
+ 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) {
+ /* 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, drdx, drdy),
- lp_build_int_to_float(float_bld, depth), "");
+ 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);
+ /* compute lod = log2(rho) */
+ lod = lp_build_log2(float_bld, rho);
- /* add lod bias */
- lod = LLVMBuildAdd(bld->builder, lod, lod_bias, "LOD bias");
+ /* add lod bias */
+ lod = LLVMBuildAdd(bld->builder, lod, lod_bias, "LOD bias");
- /* clamp lod */
- lod = lp_build_clamp(float_bld, lod, min_lod, max_lod);
+ /* clamp lod */
+ lod = lp_build_clamp(float_bld, lod, min_lod, max_lod);
- return lod;
+ return lod;
+ }
}
@@ -986,7 +997,7 @@ lp_build_linear_mip_levels(struct lp_build_sample_context *bld,
last_level);
/* compute level 1 and clamp to legal range of levels */
*level1_out = lp_build_add(int_bld, *level0_out, int_bld->one);
- *level1_out = lp_build_min(int_bld, *level1_out, int_bld->zero);
+ *level1_out = lp_build_min(int_bld, *level1_out, last_level);
*weight_out = lp_build_fract(float_bld, lod);
}
@@ -1215,7 +1226,7 @@ static LLVMValueRef
lp_build_cube_ima(struct lp_build_context *coord_bld, LLVMValueRef coord)
{
/* ima = -0.5 / abs(coord); */
- LLVMValueRef negHalf = lp_build_const_scalar(coord_bld->type, -0.5);
+ 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));
@@ -1235,7 +1246,7 @@ lp_build_cube_coord(struct lp_build_context *coord_bld,
LLVMValueRef coord, LLVMValueRef ima)
{
/* return negate(coord) * ima * sign + 0.5; */
- LLVMValueRef half = lp_build_const_scalar(coord_bld->type, 0.5);
+ LLVMValueRef half = lp_build_const_vec(coord_bld->type, 0.5);
LLVMValueRef res;
assert(negate_coord == +1 || negate_coord == -1);
@@ -1414,6 +1425,85 @@ lp_build_cube_lookup(struct lp_build_sample_context *bld,
/**
+ * Sample the texture/mipmap using given image filter and mip filter.
+ * data0_ptr and data1_ptr point to the two mipmap levels to sample
+ * from. width0/1_vec, height0/1_vec, depth0/1_vec indicate their sizes.
+ * If we're using nearest miplevel sampling the '1' values will be null/unused.
+ */
+static void
+lp_build_sample_mipmap(struct lp_build_sample_context *bld,
+ unsigned img_filter,
+ unsigned mip_filter,
+ LLVMValueRef s,
+ LLVMValueRef t,
+ LLVMValueRef r,
+ 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)
+{
+ LLVMValueRef colors0[4], colors1[4];
+ int chan;
+
+ if (img_filter == PIPE_TEX_FILTER_NEAREST) {
+ lp_build_sample_image_nearest(bld,
+ width0_vec, height0_vec, depth0_vec,
+ 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);
+ }
+ }
+ else {
+ assert(img_filter == PIPE_TEX_FILTER_LINEAR);
+
+ lp_build_sample_image_linear(bld,
+ width0_vec, height0_vec, depth0_vec,
+ 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);
+ }
+ }
+
+ 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,
+ colors0[chan], colors1[chan]);
+ }
+ }
+ else {
+ /* use first/only level's colors */
+ for (chan = 0; chan < 4; chan++) {
+ colors_out[chan] = colors0[chan];
+ }
+ }
+}
+
+
+
+/**
* General texture sampling codegen.
* This function handles texture sampling for all texture targets (1D,
* 2D, 3D, cube) and all filtering modes.
@@ -1435,6 +1525,7 @@ lp_build_sample_general(struct lp_build_sample_context *bld,
LLVMValueRef data_array,
LLVMValueRef *colors_out)
{
+ struct lp_build_context *float_bld = &bld->float_bld;
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;
@@ -1446,7 +1537,6 @@ lp_build_sample_general(struct lp_build_sample_context *bld,
LLVMValueRef row_stride0_vec = NULL, row_stride1_vec = NULL;
LLVMValueRef img_stride0_vec = NULL, img_stride1_vec = NULL;
LLVMValueRef data_ptr0, data_ptr1;
- int chan;
/*
printf("%s mip %d min %d mag %d\n", __FUNCTION__,
@@ -1454,16 +1544,24 @@ lp_build_sample_general(struct lp_build_sample_context *bld,
*/
/*
- * Compute the level of detail (mipmap level index(es)).
+ * Compute the level of detail (float).
+ */
+ if (min_filter != mag_filter ||
+ mip_filter != PIPE_TEX_MIPFILTER_NONE) {
+ /* 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, width, height, depth);
+ }
+
+ /*
+ * Compute integer mipmap level(s) to fetch texels from.
*/
if (mip_filter == PIPE_TEX_MIPFILTER_NONE) {
/* always use mip level 0 */
ilevel0 = LLVMConstInt(LLVMInt32Type(), 0, 0);
}
else {
- /* compute float LOD */
- lod = lp_build_lod_selector(bld, s, t, r, width, height, depth);
-
if (mip_filter == PIPE_TEX_MIPFILTER_NEAREST) {
lp_build_nearest_mip_level(bld, unit, lod, &ilevel0);
}
@@ -1499,7 +1597,7 @@ lp_build_sample_general(struct lp_build_sample_context *bld,
}
}
if (mip_filter == PIPE_TEX_MIPFILTER_LINEAR) {
- /* compute width, height, depth for second mipmap level at ilevel1 */
+ /* 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);
@@ -1516,7 +1614,7 @@ lp_build_sample_general(struct lp_build_sample_context *bld,
}
/*
- * Choose cube face, recompute texcoords.
+ * Choose cube face, recompute per-face texcoords.
*/
if (bld->static_state->target == PIPE_TEXTURE_CUBE) {
LLVMValueRef face, face_s, face_t;
@@ -1538,62 +1636,67 @@ lp_build_sample_general(struct lp_build_sample_context *bld,
/*
* Get/interpolate texture colors.
*/
- /* XXX temporarily force this path: */
- if (1 /*min_filter == mag_filter*/) {
- /* same filter for minification or magnification */
- LLVMValueRef colors0[4], colors1[4];
-
- if (min_filter == PIPE_TEX_FILTER_NEAREST) {
- lp_build_sample_image_nearest(bld,
- width0_vec, height0_vec, depth0_vec,
- 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);
- }
- }
- else {
- assert(min_filter == PIPE_TEX_FILTER_LINEAR);
+ 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);
+ }
+ else {
+ /* Emit conditional to choose min image filter or mag image filter
+ * 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;
- lp_build_sample_image_linear(bld,
- width0_vec, height0_vec, depth0_vec,
- row_stride0_vec, img_stride0_vec,
- data_ptr0, s, t, r, colors0);
+ flow_ctx = lp_build_flow_create(bld->builder);
+ lp_build_flow_scope_begin(flow_ctx);
+ 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]);
- 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);
- }
- }
+ /* minify = lod > 0.0 */
+ minify = LLVMBuildFCmp(bld->builder, LLVMRealUGE,
+ lod, float_bld->zero, "");
- 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,
- colors0[chan], colors1[chan]);
- }
+ lp_build_if(&if_ctx, flow_ctx, bld->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);
}
- else {
- /* use first/only level's colors */
- for (chan = 0; chan < 4; chan++) {
- colors_out[chan] = colors0[chan];
- }
+ 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);
}
- }
- else {
- /* emit conditional to choose min image filter or mag image filter
- * depending on the lod being >0 or <= 0, respectively.
- */
- abort();
+ lp_build_endif(&if_ctx);
+
+ lp_build_flow_scope_end(flow_ctx);
+ lp_build_flow_destroy(flow_ctx);
}
}
@@ -1605,7 +1708,7 @@ lp_build_rgba8_to_f32_soa(LLVMBuilderRef builder,
LLVMValueRef packed,
LLVMValueRef *rgba)
{
- LLVMValueRef mask = lp_build_int_const_scalar(dst_type, 0xff);
+ LLVMValueRef mask = lp_build_const_int_vec(dst_type, 0xff);
unsigned chan;
/* Decode the input vector components */
@@ -1617,7 +1720,7 @@ lp_build_rgba8_to_f32_soa(LLVMBuilderRef builder,
input = packed;
if(start)
- input = LLVMBuildLShr(builder, input, lp_build_int_const_scalar(dst_type, start), "");
+ input = LLVMBuildLShr(builder, input, lp_build_const_int_vec(dst_type, start), "");
if(stop < 32)
input = LLVMBuildAnd(builder, input, mask, "");
@@ -1679,17 +1782,17 @@ lp_build_sample_2d_linear_aos(struct lp_build_sample_context *bld,
t = LLVMBuildFPToSI(builder, t, i32_vec_type, "");
/* subtract 0.5 (add -128) */
- i32_c128 = lp_build_int_const_scalar(i32.type, -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_int_const_scalar(i32.type, 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_int_const_scalar(i32.type, 255);
+ i32_c255 = lp_build_const_int_vec(i32.type, 255);
s_fpart = LLVMBuildAnd(builder, s, i32_c255, "");
t_fpart = LLVMBuildAnd(builder, t, i32_c255, "");
@@ -1856,7 +1959,7 @@ lp_build_sample_compare(struct lp_build_sample_context *bld,
}
assert(res);
- res = lp_build_mul(texel_bld, res, lp_build_const_scalar(texel_bld->type, 0.25));
+ res = lp_build_mul(texel_bld, res, lp_build_const_vec(texel_bld->type, 0.25));
/* XXX returning result for default GL_DEPTH_TEXTURE_MODE = GL_LUMINANCE */
for(chan = 0; chan < 3; ++chan)
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_struct.h b/src/gallium/auxiliary/gallivm/lp_bld_struct.h
index 34478c10f5..147336edb4 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_struct.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_struct.h
@@ -37,7 +37,7 @@
#define LP_BLD_STRUCT_H
-#include "os/os_llvm.h"
+#include "gallivm/lp_bld.h"
#include <llvm-c/Target.h>
#include "util/u_debug.h"
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_swizzle.c b/src/gallium/auxiliary/gallivm/lp_bld_swizzle.c
index 64e81f7b1f..278c838eac 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_swizzle.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_swizzle.c
@@ -144,9 +144,9 @@ lp_build_broadcast_aos(struct lp_build_context *bld,
#endif
if(shift > 0)
- tmp = LLVMBuildLShr(bld->builder, a, lp_build_int_const_scalar(type4, shift*type.width), "");
+ tmp = LLVMBuildLShr(bld->builder, a, lp_build_const_int_vec(type4, shift*type.width), "");
if(shift < 0)
- tmp = LLVMBuildShl(bld->builder, a, lp_build_int_const_scalar(type4, -shift*type.width), "");
+ tmp = LLVMBuildShl(bld->builder, a, lp_build_const_int_vec(type4, -shift*type.width), "");
assert(tmp);
if(tmp)
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_swizzle.h b/src/gallium/auxiliary/gallivm/lp_bld_swizzle.h
index 57b5cc079f..138ca620e6 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_swizzle.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_swizzle.h
@@ -37,7 +37,7 @@
#define LP_BLD_SWIZZLE_H
-#include "os/os_llvm.h"
+#include "gallivm/lp_bld.h"
struct lp_type;
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_tgsi.h b/src/gallium/auxiliary/gallivm/lp_bld_tgsi.h
index 0f2f8a65b1..63b938bfa9 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_tgsi.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_tgsi.h
@@ -35,7 +35,7 @@
#ifndef LP_BLD_TGSI_H
#define LP_BLD_TGSI_H
-#include "os/os_llvm.h"
+#include "gallivm/lp_bld.h"
struct tgsi_token;
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c
index 5ec59d636c..f160be878f 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c
@@ -475,7 +475,7 @@ emit_store(
break;
case TGSI_SAT_MINUS_PLUS_ONE:
- value = lp_build_max(&bld->base, value, lp_build_const_scalar(bld->base.type, -1.0));
+ value = lp_build_max(&bld->base, value, lp_build_const_vec(bld->base.type, -1.0));
value = lp_build_min(&bld->base, value, bld->base.one);
break;
@@ -996,7 +996,7 @@ emit_instruction(
src0 = emit_fetch( bld, inst, 0, chan_index );
src1 = emit_fetch( bld, inst, 1, chan_index );
src2 = emit_fetch( bld, inst, 2, chan_index );
- tmp1 = lp_build_const_scalar(bld->base.type, 0.5);
+ tmp1 = lp_build_const_vec(bld->base.type, 0.5);
tmp0 = lp_build_cmp( &bld->base, PIPE_FUNC_GREATER, src2, tmp1);
dst0[chan_index] = lp_build_select( &bld->base, tmp0, src0, src1 );
}
@@ -1713,7 +1713,7 @@ lp_build_tgsi_soa(LLVMBuilderRef builder,
assert(num_immediates < LP_MAX_IMMEDIATES);
for( i = 0; i < size; ++i )
bld.immediates[num_immediates][i] =
- lp_build_const_scalar(type, parse.FullToken.FullImmediate.u[i].Float);
+ lp_build_const_vec(type, parse.FullToken.FullImmediate.u[i].Float);
for( i = size; i < 4; ++i )
bld.immediates[num_immediates][i] = bld.base.undef;
num_immediates++;
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_type.h b/src/gallium/auxiliary/gallivm/lp_bld_type.h
index 5b351476ac..cd59d2faa6 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_type.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_type.h
@@ -37,9 +37,9 @@
#define LP_BLD_TYPE_H
-#include "os/os_llvm.h"
+#include "pipe/p_compiler.h"
+#include "gallivm/lp_bld.h"
-#include <pipe/p_compiler.h>
/**