1/**************************************************************************
2 *
3 * Copyright 2010 VMware, Inc.
4 * All Rights Reserved.
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sub license, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
13 *
14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
17 * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
18 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
19 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
20 * USE OR OTHER DEALINGS IN THE SOFTWARE.
21 *
22 * The above copyright notice and this permission notice (including the
23 * next paragraph) shall be included in all copies or substantial portions
24 * of the Software.
25 *
26 **************************************************************************/
27
28
29#include "lp_bld_type.h"
30#include "lp_bld_arit.h"
31#include "lp_bld_const.h"
32#include "lp_bld_swizzle.h"
33#include "lp_bld_quad.h"
34#include "lp_bld_pack.h"
35
36
37static const unsigned char
38swizzle_left[4] = {
39   LP_BLD_QUAD_TOP_LEFT,     LP_BLD_QUAD_TOP_LEFT,
40   LP_BLD_QUAD_BOTTOM_LEFT,  LP_BLD_QUAD_BOTTOM_LEFT
41};
42
43static const unsigned char
44swizzle_right[4] = {
45   LP_BLD_QUAD_TOP_RIGHT,    LP_BLD_QUAD_TOP_RIGHT,
46   LP_BLD_QUAD_BOTTOM_RIGHT, LP_BLD_QUAD_BOTTOM_RIGHT
47};
48
49static const unsigned char
50swizzle_top[4] = {
51   LP_BLD_QUAD_TOP_LEFT,     LP_BLD_QUAD_TOP_RIGHT,
52   LP_BLD_QUAD_TOP_LEFT,     LP_BLD_QUAD_TOP_RIGHT
53};
54
55static const unsigned char
56swizzle_bottom[4] = {
57   LP_BLD_QUAD_BOTTOM_LEFT,  LP_BLD_QUAD_BOTTOM_RIGHT,
58   LP_BLD_QUAD_BOTTOM_LEFT,  LP_BLD_QUAD_BOTTOM_RIGHT
59};
60
61
62LLVMValueRef
63lp_build_ddx(struct lp_build_context *bld,
64             LLVMValueRef a)
65{
66   LLVMValueRef a_left  = lp_build_swizzle_aos(bld, a, swizzle_left);
67   LLVMValueRef a_right = lp_build_swizzle_aos(bld, a, swizzle_right);
68   return lp_build_sub(bld, a_right, a_left);
69}
70
71
72LLVMValueRef
73lp_build_ddy(struct lp_build_context *bld,
74             LLVMValueRef a)
75{
76   LLVMValueRef a_top    = lp_build_swizzle_aos(bld, a, swizzle_top);
77   LLVMValueRef a_bottom = lp_build_swizzle_aos(bld, a, swizzle_bottom);
78   return lp_build_sub(bld, a_bottom, a_top);
79}
80
81/*
82 * Helper for building packed ddx/ddy vector for one coord (scalar per quad
83 * values). The vector will look like this (8-wide):
84 * dr1dx _____ -dr1dy _____ dr2dx _____ -dr2dy _____
85 * This only requires one shuffle instead of two for more straightforward packing.
86 */
87LLVMValueRef
88lp_build_packed_ddx_ddy_onecoord(struct lp_build_context *bld,
89                                 LLVMValueRef a)
90{
91   struct gallivm_state *gallivm = bld->gallivm;
92   LLVMBuilderRef builder = gallivm->builder;
93   LLVMValueRef vec1, vec2;
94
95   /* use aos swizzle helper */
96
97   static const unsigned char swizzle1[] = { /* no-op swizzle */
98      LP_BLD_QUAD_TOP_LEFT, LP_BLD_SWIZZLE_DONTCARE,
99      LP_BLD_QUAD_BOTTOM_LEFT, LP_BLD_SWIZZLE_DONTCARE
100   };
101   static const unsigned char swizzle2[] = {
102      LP_BLD_QUAD_TOP_RIGHT, LP_BLD_SWIZZLE_DONTCARE,
103      LP_BLD_QUAD_TOP_LEFT, LP_BLD_SWIZZLE_DONTCARE
104   };
105
106   vec1 = lp_build_swizzle_aos(bld, a, swizzle1);
107   vec2 = lp_build_swizzle_aos(bld, a, swizzle2);
108
109   if (bld->type.floating)
110      return LLVMBuildFSub(builder, vec2, vec1, "ddxddy");
111   else
112      return LLVMBuildSub(builder, vec2, vec1, "ddxddy");
113}
114
115
116/*
117 * Helper for building packed ddx/ddy vector for one coord (scalar per quad
118 * values). The vector will look like this (8-wide):
119 * ds1dx ds1dy dt1dx dt1dy ds2dx ds2dy dt2dx dt2dy
120 * This only needs 2 (v)shufps.
121 */
122LLVMValueRef
123lp_build_packed_ddx_ddy_twocoord(struct lp_build_context *bld,
124                                 LLVMValueRef a, LLVMValueRef b)
125{
126   struct gallivm_state *gallivm = bld->gallivm;
127   LLVMBuilderRef builder = gallivm->builder;
128   LLVMValueRef shuffles1[LP_MAX_VECTOR_LENGTH/4];
129   LLVMValueRef shuffles2[LP_MAX_VECTOR_LENGTH/4];
130   LLVMValueRef vec1, vec2;
131   unsigned length, num_quads, i;
132
133   /* XXX: do hsub version */
134   length = bld->type.length;
135   num_quads = length / 4;
136   for (i = 0; i < num_quads; i++) {
137      unsigned s1 = 4 * i;
138      unsigned s2 = 4 * i + length;
139      shuffles1[4*i + 0] = lp_build_const_int32(gallivm, LP_BLD_QUAD_TOP_LEFT + s1);
140      shuffles1[4*i + 1] = lp_build_const_int32(gallivm, LP_BLD_QUAD_TOP_LEFT + s1);
141      shuffles1[4*i + 2] = lp_build_const_int32(gallivm, LP_BLD_QUAD_TOP_LEFT + s2);
142      shuffles1[4*i + 3] = lp_build_const_int32(gallivm, LP_BLD_QUAD_TOP_LEFT + s2);
143      shuffles2[4*i + 0] = lp_build_const_int32(gallivm, LP_BLD_QUAD_TOP_RIGHT + s1);
144      shuffles2[4*i + 1] = lp_build_const_int32(gallivm, LP_BLD_QUAD_BOTTOM_LEFT + s1);
145      shuffles2[4*i + 2] = lp_build_const_int32(gallivm, LP_BLD_QUAD_TOP_RIGHT + s2);
146      shuffles2[4*i + 3] = lp_build_const_int32(gallivm, LP_BLD_QUAD_BOTTOM_LEFT + s2);
147   }
148   vec1 = LLVMBuildShuffleVector(builder, a, b,
149                                 LLVMConstVector(shuffles1, length), "");
150   vec2 = LLVMBuildShuffleVector(builder, a, b,
151                                 LLVMConstVector(shuffles2, length), "");
152   if (bld->type.floating)
153      return LLVMBuildFSub(builder, vec2, vec1, "ddxddyddxddy");
154   else
155      return LLVMBuildSub(builder, vec2, vec1, "ddxddyddxddy");
156}
157
158
159/**
160 * Twiddle from quad format to row format
161 *
162 *   src0      src1
163 * ######### #########      #################
164 * # 0 | 1 # # 4 | 5 #      # 0 | 1 | 4 | 5 # src0
165 * #---+---# #---+---#  ->  #################
166 * # 2 | 3 # # 6 | 7 #      # 2 | 3 | 6 | 7 # src1
167 * ######### #########      #################
168 *
169 */
170void
171lp_bld_quad_twiddle(struct gallivm_state *gallivm,
172                    struct lp_type lp_dst_type,
173                    const LLVMValueRef* src,
174                    unsigned src_count,
175                    LLVMValueRef* dst)
176{
177   LLVMBuilderRef builder = gallivm->builder;
178   LLVMTypeRef dst_type_ref;
179   LLVMTypeRef type2_ref;
180   struct lp_type type2;
181   unsigned i;
182
183   assert((src_count % 2) == 0);
184
185   /* Create a type with only 2 elements */
186   type2 = lp_dst_type;
187   type2.width = (lp_dst_type.width * lp_dst_type.length) / 2;
188   type2.length = 2;
189   type2.floating = 0;
190
191   type2_ref = lp_build_vec_type(gallivm, type2);
192   dst_type_ref = lp_build_vec_type(gallivm, lp_dst_type);
193
194   for (i = 0; i < src_count; i += 2) {
195      LLVMValueRef src0, src1;
196
197      src0 = LLVMBuildBitCast(builder, src[i + 0], type2_ref, "");
198      src1 = LLVMBuildBitCast(builder, src[i + 1], type2_ref, "");
199
200      dst[i + 0] = lp_build_interleave2(gallivm, type2, src0, src1, 0);
201      dst[i + 1] = lp_build_interleave2(gallivm, type2, src0, src1, 1);
202
203      dst[i + 0] = LLVMBuildBitCast(builder, dst[i + 0], dst_type_ref, "");
204      dst[i + 1] = LLVMBuildBitCast(builder, dst[i + 1], dst_type_ref, "");
205   }
206}
207