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
68 /**
69 * Find a native RGB swizzle that matches the given swizzle.
70 * Returns 0 if none found.
71 */
lookup_native_swizzle(unsigned int swizzle)72 static const struct swizzle_data* lookup_native_swizzle(unsigned int swizzle)
73 {
74 int i, comp;
75
76 for(i = 0; i < num_native_swizzles; ++i) {
77 const struct swizzle_data* sd = &native_swizzles[i];
78 for(comp = 0; comp < 3; ++comp) {
79 unsigned int swz = GET_SWZ(swizzle, comp);
80 if (swz == RC_SWIZZLE_UNUSED)
81 continue;
82 if (swz != GET_SWZ(sd->hash, comp))
83 break;
84 }
85 if (comp == 3)
86 return sd;
87 }
88
89 return NULL;
90 }
91
92 /**
93 * Determines if the given swizzle is valid for r300/r400. In most situations
94 * it is better to use r300_swizzle_is_native() which can be accessed via
95 * struct radeon_compiler *c; c->SwizzleCaps->IsNative().
96 */
r300_swizzle_is_native_basic(unsigned int swizzle)97 int r300_swizzle_is_native_basic(unsigned int swizzle)
98 {
99 if(lookup_native_swizzle(swizzle))
100 return 1;
101 else
102 return 0;
103 }
104
105 /**
106 * Check whether the given instruction supports the swizzle and negate
107 * combinations in the given source register.
108 */
r300_swizzle_is_native(rc_opcode opcode,struct rc_src_register reg)109 static int r300_swizzle_is_native(rc_opcode opcode, struct rc_src_register reg)
110 {
111 const struct swizzle_data* sd;
112 unsigned int relevant;
113 int j;
114
115 if (opcode == RC_OPCODE_KIL ||
116 opcode == RC_OPCODE_TEX ||
117 opcode == RC_OPCODE_TXB ||
118 opcode == RC_OPCODE_TXP) {
119 if (reg.Abs || reg.Negate)
120 return 0;
121
122 for(j = 0; j < 4; ++j) {
123 unsigned int swz = GET_SWZ(reg.Swizzle, j);
124 if (swz == RC_SWIZZLE_UNUSED)
125 continue;
126 if (swz != j)
127 return 0;
128 }
129
130 return 1;
131 }
132
133 relevant = 0;
134
135 for(j = 0; j < 3; ++j)
136 if (GET_SWZ(reg.Swizzle, j) != RC_SWIZZLE_UNUSED)
137 relevant |= 1 << j;
138
139 if ((reg.Negate & relevant) && ((reg.Negate & relevant) != relevant))
140 return 0;
141
142 sd = lookup_native_swizzle(reg.Swizzle);
143 if (!sd || (reg.File == RC_FILE_PRESUB && sd->srcp_stride == 0))
144 return 0;
145
146 return 1;
147 }
148
149
r300_swizzle_split(struct rc_src_register src,unsigned int mask,struct rc_swizzle_split * split)150 static void r300_swizzle_split(
151 struct rc_src_register src, unsigned int mask,
152 struct rc_swizzle_split * split)
153 {
154 split->NumPhases = 0;
155
156 while(mask) {
157 unsigned int best_matchcount = 0;
158 unsigned int best_matchmask = 0;
159 int i, comp;
160
161 for(i = 0; i < num_native_swizzles; ++i) {
162 const struct swizzle_data *sd = &native_swizzles[i];
163 unsigned int matchcount = 0;
164 unsigned int matchmask = 0;
165 for(comp = 0; comp < 3; ++comp) {
166 unsigned int swz;
167 if (!GET_BIT(mask, comp))
168 continue;
169 swz = GET_SWZ(src.Swizzle, comp);
170 if (swz == RC_SWIZZLE_UNUSED)
171 continue;
172 if (swz == GET_SWZ(sd->hash, comp)) {
173 /* check if the negate bit of current component
174 * is the same for already matched components */
175 if (matchmask && (!!(src.Negate & matchmask) != !!(src.Negate & (1 << comp))))
176 continue;
177
178 matchcount++;
179 matchmask |= 1 << comp;
180 }
181 }
182 if (matchcount > best_matchcount) {
183 best_matchcount = matchcount;
184 best_matchmask = matchmask;
185 if (matchmask == (mask & RC_MASK_XYZ))
186 break;
187 }
188 }
189
190 if (mask & RC_MASK_W)
191 best_matchmask |= RC_MASK_W;
192
193 split->Phase[split->NumPhases++] = best_matchmask;
194 mask &= ~best_matchmask;
195 }
196 }
197
198 const struct rc_swizzle_caps r300_swizzle_caps = {
199 .IsNative = r300_swizzle_is_native,
200 .Split = r300_swizzle_split
201 };
202
203
204 /**
205 * Translate an RGB (XYZ) swizzle into the hardware code for the given
206 * instruction source.
207 */
r300FPTranslateRGBSwizzle(unsigned int src,unsigned int swizzle)208 unsigned int r300FPTranslateRGBSwizzle(unsigned int src, unsigned int swizzle)
209 {
210 const struct swizzle_data* sd = lookup_native_swizzle(swizzle);
211
212 if (!sd || (src == RC_PAIR_PRESUB_SRC && sd->srcp_stride == 0)) {
213 fprintf(stderr, "Not a native swizzle: %08x\n", swizzle);
214 return 0;
215 }
216
217 if (src == RC_PAIR_PRESUB_SRC) {
218 return sd->base + sd->srcp_stride;
219 } else {
220 return sd->base + src*sd->stride;
221 }
222 }
223
224
225 /**
226 * Translate an Alpha (W) swizzle into the hardware code for the given
227 * instruction source.
228 */
r300FPTranslateAlphaSwizzle(unsigned int src,unsigned int swizzle)229 unsigned int r300FPTranslateAlphaSwizzle(unsigned int src, unsigned int swizzle)
230 {
231 unsigned int swz = GET_SWZ(swizzle, 0);
232 if (src == RC_PAIR_PRESUB_SRC) {
233 return R300_ALU_ARGA_SRCP_X + swz;
234 }
235 if (swz < 3)
236 return swz + 3*src;
237
238 switch(swz) {
239 case RC_SWIZZLE_W: return R300_ALU_ARGA_SRC0A + src;
240 case RC_SWIZZLE_ONE: return R300_ALU_ARGA_ONE;
241 case RC_SWIZZLE_ZERO: return R300_ALU_ARGA_ZERO;
242 case RC_SWIZZLE_HALF: return R300_ALU_ARGA_HALF;
243 default: return R300_ALU_ARGA_ONE;
244 }
245 }
246