1 /*
2 * Copyright (C) 2008 Nicolai Haehnle.
3 *
4 * All Rights Reserved.
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining
7 * a 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, sublicense, 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 above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial
16 * portions of the Software.
17 *
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
19 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
21 * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
22 * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
23 * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
24 * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
25 *
26 */
27
28 /**
29 * @file
30 * Utilities to deal with the somewhat odd restriction on R300 fragment
31 * program swizzles.
32 */
33
34 #include "r300_fragprog_swizzle.h"
35
36 #include <stdio.h>
37
38 #include "util/macros.h"
39
40 #include "r300_reg.h"
41 #include "radeon_compiler.h"
42
43 #define MAKE_SWZ3(x, y, z) (RC_MAKE_SWIZZLE(RC_SWIZZLE_##x, RC_SWIZZLE_##y, RC_SWIZZLE_##z, RC_SWIZZLE_ZERO))
44
45 struct swizzle_data {
46 unsigned int hash; /**< swizzle value this matches */
47 unsigned int base; /**< base value for hw swizzle */
48 unsigned int stride; /**< difference in base between arg0/1/2 */
49 unsigned int srcp_stride; /**< difference in base between arg0/scrp */
50 };
51
52 static const struct swizzle_data native_swizzles[] = {
53 {MAKE_SWZ3(X, Y, Z), R300_ALU_ARGC_SRC0C_XYZ, 4, 15},
54 {MAKE_SWZ3(X, X, X), R300_ALU_ARGC_SRC0C_XXX, 4, 15},
55 {MAKE_SWZ3(Y, Y, Y), R300_ALU_ARGC_SRC0C_YYY, 4, 15},
56 {MAKE_SWZ3(Z, Z, Z), R300_ALU_ARGC_SRC0C_ZZZ, 4, 15},
57 {MAKE_SWZ3(W, W, W), R300_ALU_ARGC_SRC0A, 1, 7},
58 {MAKE_SWZ3(Y, Z, X), R300_ALU_ARGC_SRC0C_YZX, 1, 0},
59 {MAKE_SWZ3(Z, X, Y), R300_ALU_ARGC_SRC0C_ZXY, 1, 0},
60 {MAKE_SWZ3(W, Z, Y), R300_ALU_ARGC_SRC0CA_WZY, 1, 0},
61 {MAKE_SWZ3(ONE, ONE, ONE), R300_ALU_ARGC_ONE, 0, 0},
62 {MAKE_SWZ3(ZERO, ZERO, ZERO), R300_ALU_ARGC_ZERO, 0, 0},
63 {MAKE_SWZ3(HALF, HALF, HALF), R300_ALU_ARGC_HALF, 0, 0}
64 };
65
66 static const int num_native_swizzles = ARRAY_SIZE(native_swizzles);
67 /* Only swizzles with srcp_stride != 0 can be used for presub, so
68 * just the first five from the list. */
69 static const int num_presub_swizzles = 5;
70
71 /**
72 * Find a native RGB swizzle that matches the given swizzle.
73 * Returns 0 if none found.
74 */
lookup_native_swizzle(unsigned int swizzle)75 static const struct swizzle_data* lookup_native_swizzle(unsigned int swizzle)
76 {
77 int i, comp;
78
79 for(i = 0; i < num_native_swizzles; ++i) {
80 const struct swizzle_data* sd = &native_swizzles[i];
81 for(comp = 0; comp < 3; ++comp) {
82 unsigned int swz = GET_SWZ(swizzle, comp);
83 if (swz == RC_SWIZZLE_UNUSED)
84 continue;
85 if (swz != GET_SWZ(sd->hash, comp))
86 break;
87 }
88 if (comp == 3)
89 return sd;
90 }
91
92 return NULL;
93 }
94
95 /**
96 * Determines if the given swizzle is valid for r300/r400. In most situations
97 * it is better to use r300_swizzle_is_native() which can be accessed via
98 * struct radeon_compiler *c; c->SwizzleCaps->IsNative().
99 */
r300_swizzle_is_native_basic(unsigned int swizzle)100 int r300_swizzle_is_native_basic(unsigned int swizzle)
101 {
102 if(lookup_native_swizzle(swizzle))
103 return 1;
104 else
105 return 0;
106 }
107
108 /**
109 * Check whether the given instruction supports the swizzle and negate
110 * combinations in the given source register.
111 */
r300_swizzle_is_native(rc_opcode opcode,struct rc_src_register reg)112 static int r300_swizzle_is_native(rc_opcode opcode, struct rc_src_register reg)
113 {
114 const struct swizzle_data* sd;
115 unsigned int relevant;
116 int j;
117
118 if (opcode == RC_OPCODE_KIL ||
119 opcode == RC_OPCODE_TEX ||
120 opcode == RC_OPCODE_TXB ||
121 opcode == RC_OPCODE_TXP) {
122 if (reg.Abs || reg.Negate)
123 return 0;
124
125 for(j = 0; j < 4; ++j) {
126 unsigned int swz = GET_SWZ(reg.Swizzle, j);
127 if (swz == RC_SWIZZLE_UNUSED)
128 continue;
129 if (swz != j)
130 return 0;
131 }
132
133 return 1;
134 }
135
136 relevant = 0;
137
138 for(j = 0; j < 3; ++j)
139 if (GET_SWZ(reg.Swizzle, j) != RC_SWIZZLE_UNUSED)
140 relevant |= 1 << j;
141
142 if ((reg.Negate & relevant) && ((reg.Negate & relevant) != relevant))
143 return 0;
144
145 sd = lookup_native_swizzle(reg.Swizzle);
146 if (!sd || (reg.File == RC_FILE_PRESUB && sd->srcp_stride == 0))
147 return 0;
148
149 return 1;
150 }
151
152
r300_swizzle_split(struct rc_src_register src,unsigned int mask,struct rc_swizzle_split * split)153 static void r300_swizzle_split(
154 struct rc_src_register src, unsigned int mask,
155 struct rc_swizzle_split * split)
156 {
157 split->NumPhases = 0;
158
159 while(mask) {
160 unsigned int best_matchcount = 0;
161 unsigned int best_matchmask = 0;
162 int i, comp;
163
164 unsigned num_swizzles = src.File == RC_FILE_PRESUB ? num_presub_swizzles : num_native_swizzles;
165
166 for(i = 0; i < num_swizzles; ++i) {
167 const struct swizzle_data *sd = &native_swizzles[i];
168 unsigned int matchcount = 0;
169 unsigned int matchmask = 0;
170 for(comp = 0; comp < 3; ++comp) {
171 unsigned int swz;
172 if (!GET_BIT(mask, comp))
173 continue;
174 swz = GET_SWZ(src.Swizzle, comp);
175 if (swz == RC_SWIZZLE_UNUSED)
176 continue;
177 if (swz == GET_SWZ(sd->hash, comp)) {
178 /* check if the negate bit of current component
179 * is the same for already matched components */
180 if (matchmask && (!!(src.Negate & matchmask) != !!(src.Negate & (1 << comp))))
181 continue;
182
183 matchcount++;
184 matchmask |= 1 << comp;
185 }
186 }
187 if (matchcount > best_matchcount) {
188 best_matchcount = matchcount;
189 best_matchmask = matchmask;
190 if (matchmask == (mask & RC_MASK_XYZ))
191 break;
192 }
193 }
194
195 if (mask & RC_MASK_W)
196 best_matchmask |= RC_MASK_W;
197
198 split->Phase[split->NumPhases++] = best_matchmask;
199 mask &= ~best_matchmask;
200 }
201 }
202
203 const struct rc_swizzle_caps r300_swizzle_caps = {
204 .IsNative = r300_swizzle_is_native,
205 .Split = r300_swizzle_split
206 };
207
208
209 /**
210 * Translate an RGB (XYZ) swizzle into the hardware code for the given
211 * instruction source.
212 */
r300FPTranslateRGBSwizzle(unsigned int src,unsigned int swizzle)213 unsigned int r300FPTranslateRGBSwizzle(unsigned int src, unsigned int swizzle)
214 {
215 const struct swizzle_data* sd = lookup_native_swizzle(swizzle);
216
217 if (!sd || (src == RC_PAIR_PRESUB_SRC && sd->srcp_stride == 0)) {
218 fprintf(stderr, "Not a native swizzle: %08x\n", swizzle);
219 return 0;
220 }
221
222 if (src == RC_PAIR_PRESUB_SRC) {
223 return sd->base + sd->srcp_stride;
224 } else {
225 return sd->base + src*sd->stride;
226 }
227 }
228
229
230 /**
231 * Translate an Alpha (W) swizzle into the hardware code for the given
232 * instruction source.
233 */
r300FPTranslateAlphaSwizzle(unsigned int src,unsigned int swizzle)234 unsigned int r300FPTranslateAlphaSwizzle(unsigned int src, unsigned int swizzle)
235 {
236 unsigned int swz = GET_SWZ(swizzle, 0);
237 if (src == RC_PAIR_PRESUB_SRC) {
238 return R300_ALU_ARGA_SRCP_X + swz;
239 }
240 if (swz < 3)
241 return swz + 3*src;
242
243 switch(swz) {
244 case RC_SWIZZLE_W: return R300_ALU_ARGA_SRC0A + src;
245 case RC_SWIZZLE_ONE: return R300_ALU_ARGA_ONE;
246 case RC_SWIZZLE_ZERO: return R300_ALU_ARGA_ZERO;
247 case RC_SWIZZLE_HALF: return R300_ALU_ARGA_HALF;
248 default: return R300_ALU_ARGA_ONE;
249 }
250 }
251