summaryrefslogtreecommitdiff
path: root/src/gallium/drivers/llvmpipe/lp_bld_swizzle.c
blob: 3182573db29cc425d04ddd97c1ece51fea5f781f (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
/**************************************************************************
 *
 * Copyright 2009 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.
 *
 **************************************************************************/


#include "util/u_debug.h"

#include "lp_bld_type.h"
#include "lp_bld_const.h"
#include "lp_bld_swizzle.h"


LLVMValueRef
lp_build_broadcast_scalar(struct lp_build_context *bld,
                          LLVMValueRef scalar)
{
   const union lp_type type = bld->type;
   LLVMValueRef res;
   unsigned i;

   res = bld->undef;
   for(i = 0; i < type.length; ++i) {
      LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
      res = LLVMBuildInsertElement(bld->builder, res, scalar, index, "");
   }

   return res;
}


LLVMValueRef
lp_build_broadcast_aos(struct lp_build_context *bld,
                       LLVMValueRef a,
                       unsigned channel)
{
   const union lp_type type = bld->type;
   const unsigned n = type.length;
   unsigned i, j;

   if(a == bld->undef || a == bld->zero || a == bld->one)
      return a;

   /* XXX: SSE3 has PSHUFB which should be better than bitmasks, but forcing
    * using shuffles here actually causes worst results. More investigation is
    * needed. */
   if (n <= 4) {
      /*
       * Shuffle.
       */
      LLVMTypeRef elem_type = LLVMInt32Type();
      LLVMValueRef shuffles[LP_MAX_VECTOR_LENGTH];

      for(j = 0; j < n; j += 4)
         for(i = 0; i < 4; ++i)
            shuffles[j + i] = LLVMConstInt(elem_type, j + channel, 0);

      return LLVMBuildShuffleVector(bld->builder, a, bld->undef, LLVMConstVector(shuffles, n), "");
   }
   else {
      /*
       * Bit mask and recursive shifts
       *
       *   XYZW XYZW .... XYZW  <= input
       *   0Y00 0Y00 .... 0Y00
       *   YY00 YY00 .... YY00
       *   YYYY YYYY .... YYYY  <= output
       */
      union lp_type type4 = type;
      const char shifts[4][2] = {
         { 1,  2},
         {-1,  2},
         { 1, -2},
         {-1, -2}
      };
      boolean cond[4];
      unsigned i;

      memset(cond, 0, sizeof cond);
      cond[channel] = 1;

      a = LLVMBuildAnd(bld->builder, a, lp_build_const_mask_aos(type, cond), "");

      type4.width *= 4;
      type4.length /= 4;

      a = LLVMBuildBitCast(bld->builder, a, lp_build_vec_type(type4), "");

      for(i = 0; i < 2; ++i) {
         LLVMValueRef tmp = NULL;
         int shift = shifts[channel][i];

#ifdef PIPE_ARCH_LITTLE_ENDIAN
         shift = -shift;
#endif

         if(shift > 0)
            tmp = LLVMBuildLShr(bld->builder, a, lp_build_int_const_uni(type4, shift*type.width), "");
         if(shift < 0)
            tmp = LLVMBuildShl(bld->builder, a, lp_build_int_const_uni(type4, -shift*type.width), "");

         assert(tmp);
         if(tmp)
            a = LLVMBuildOr(bld->builder, a, tmp, "");
      }

      return LLVMBuildBitCast(bld->builder, a, lp_build_vec_type(type), "");
   }
}


LLVMValueRef
lp_build_select_aos(struct lp_build_context *bld,
                    LLVMValueRef a,
                    LLVMValueRef b,
                    boolean cond[4])
{
   const union lp_type type = bld->type;
   const unsigned n = type.length;
   unsigned i, j;

   if(a == b)
      return a;
   if(cond[0] && cond[1] && cond[2] && cond[3])
      return a;
   if(!cond[0] && !cond[1] && !cond[2] && !cond[3])
      return b;
   if(a == bld->undef || b == bld->undef)
      return bld->undef;

   /*
    * There are three major ways of accomplishing this:
    * - with a shuffle,
    * - with a select,
    * - or with a bit mask.
    *
    * Select isn't supported for vector types yet.
    * The flip between these is empirical and might need to be.
    */
   if (n <= 4) {
      /*
       * Shuffle.
       */
      LLVMTypeRef elem_type = LLVMInt32Type();
      LLVMValueRef shuffles[LP_MAX_VECTOR_LENGTH];

      for(j = 0; j < n; j += 4)
         for(i = 0; i < 4; ++i)
            shuffles[j + i] = LLVMConstInt(elem_type, (cond[i] ? 0 : n) + j + i, 0);

      return LLVMBuildShuffleVector(bld->builder, a, b, LLVMConstVector(shuffles, n), "");
   }
#if 0
   else if(0) {
      /* FIXME: Unfortunately select of vectors do not work */
      /* Use a select */
      LLVMTypeRef elem_type = LLVMInt1Type();
      LLVMValueRef cond[LP_MAX_VECTOR_LENGTH];

      for(j = 0; j < n; j += 4)
         for(i = 0; i < 4; ++i)
            cond[j + i] = LLVMConstInt(elem_type, cond[i] ? 1 : 0, 0);

      return LLVMBuildSelect(bld->builder, LLVMConstVector(cond, n), a, b, "");
   }
#endif
   else {
      LLVMValueRef mask = lp_build_const_mask_aos(type, cond);

      /* TODO: On SSE4 we could do this with a single instruction -- PBLENDVB */

      a = LLVMBuildAnd(bld->builder, a, mask, "");

      /* This often gets translated to PANDN, but sometimes the NOT is
       * pre-computed and stored in another constant. The best strategy depends
       * on available registers, so it is not a big deal -- hopefully LLVM does
       * the right decision attending the rest of the program.
       */
      b = LLVMBuildAnd(bld->builder, b, LLVMBuildNot(bld->builder, mask, ""), "");

      return LLVMBuildOr(bld->builder, a, b, "");
   }
}


LLVMValueRef
lp_build_swizzle1_aos(struct lp_build_context *bld,
                      LLVMValueRef a,
                      unsigned char swizzle[4])
{
   const unsigned n = bld->type.length;
   unsigned i, j;

   if(a == bld->undef || a == bld->zero || a == bld->one)
      return a;

   if(swizzle[0] == swizzle[1] && swizzle[1] == swizzle[2] && swizzle[2] == swizzle[3])
      return lp_build_broadcast_aos(bld, a, swizzle[0]);

   {
      /*
       * Shuffle.
       */
      LLVMTypeRef elem_type = LLVMInt32Type();
      LLVMValueRef shuffles[LP_MAX_VECTOR_LENGTH];

      for(j = 0; j < n; j += 4)
         for(i = 0; i < 4; ++i)
            shuffles[j + i] = LLVMConstInt(elem_type, j + swizzle[i], 0);

      return LLVMBuildShuffleVector(bld->builder, a, bld->undef, LLVMConstVector(shuffles, n), "");
   }
}


LLVMValueRef
lp_build_swizzle2_aos(struct lp_build_context *bld,
                      LLVMValueRef a,
                      LLVMValueRef b,
                      unsigned char swizzle[4])
{
   const unsigned n = bld->type.length;
   unsigned i, j;

   if(swizzle[0] < 4 && swizzle[1] < 4 && swizzle[2] < 4 && swizzle[3] < 4)
      return lp_build_swizzle1_aos(bld, a, swizzle);

   if(a == b) {
      swizzle[0] %= 4;
      swizzle[1] %= 4;
      swizzle[2] %= 4;
      swizzle[3] %= 4;
      return lp_build_swizzle1_aos(bld, a, swizzle);
   }

   if(swizzle[0] % 4 == 0 &&
      swizzle[1] % 4 == 1 &&
      swizzle[2] % 4 == 2 &&
      swizzle[3] % 4 == 3) {
      boolean cond[4];
      cond[0] = swizzle[0] / 4;
      cond[1] = swizzle[1] / 4;
      cond[2] = swizzle[2] / 4;
      cond[3] = swizzle[3] / 4;
      return lp_build_select_aos(bld, a, b, cond);
   }

   {
      /*
       * Shuffle.
       */
      LLVMTypeRef elem_type = LLVMInt32Type();
      LLVMValueRef shuffles[LP_MAX_VECTOR_LENGTH];

      for(j = 0; j < n; j += 4)
         for(i = 0; i < 4; ++i)
            shuffles[j + i] = LLVMConstInt(elem_type, j + (swizzle[i] % 4) + (swizzle[i] / 4 * n), 0);

      return LLVMBuildShuffleVector(bld->builder, a, b, LLVMConstVector(shuffles, n), "");
   }
}