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 "../r300_reg.h"
39 #include "radeon_compiler.h"
40
41 #define MAKE_SWZ3(x, y, z) (RC_MAKE_SWIZZLE(RC_SWIZZLE_##x, RC_SWIZZLE_##y, RC_SWIZZLE_##z, RC_SWIZZLE_ZERO))
42
43 struct swizzle_data {
44 unsigned int hash; /**< swizzle value this matches */
45 unsigned int base; /**< base value for hw swizzle */
46 unsigned int stride; /**< difference in base between arg0/1/2 */
47 unsigned int srcp_stride; /**< difference in base between arg0/scrp */
48 };
49
50 static const struct swizzle_data native_swizzles[] = {
51 {MAKE_SWZ3(X, Y, Z), R300_ALU_ARGC_SRC0C_XYZ, 4, 15},
52 {MAKE_SWZ3(X, X, X), R300_ALU_ARGC_SRC0C_XXX, 4, 15},
53 {MAKE_SWZ3(Y, Y, Y), R300_ALU_ARGC_SRC0C_YYY, 4, 15},
54 {MAKE_SWZ3(Z, Z, Z), R300_ALU_ARGC_SRC0C_ZZZ, 4, 15},
55 {MAKE_SWZ3(W, W, W), R300_ALU_ARGC_SRC0A, 1, 7},
56 {MAKE_SWZ3(Y, Z, X), R300_ALU_ARGC_SRC0C_YZX, 1, 0},
57 {MAKE_SWZ3(Z, X, Y), R300_ALU_ARGC_SRC0C_ZXY, 1, 0},
58 {MAKE_SWZ3(W, Z, Y), R300_ALU_ARGC_SRC0CA_WZY, 1, 0},
59 {MAKE_SWZ3(ONE, ONE, ONE), R300_ALU_ARGC_ONE, 0, 0},
60 {MAKE_SWZ3(ZERO, ZERO, ZERO), R300_ALU_ARGC_ZERO, 0, 0},
61 {MAKE_SWZ3(HALF, HALF, HALF), R300_ALU_ARGC_HALF, 0, 0}
62 };
63
64 static const int num_native_swizzles = sizeof(native_swizzles)/sizeof(native_swizzles[0]);
65
66 /**
67 * Find a native RGB swizzle that matches the given swizzle.
68 * Returns 0 if none found.
69 */
lookup_native_swizzle(unsigned int swizzle)70 static const struct swizzle_data* lookup_native_swizzle(unsigned int swizzle)
71 {
72 int i, comp;
73
74 for(i = 0; i < num_native_swizzles; ++i) {
75 const struct swizzle_data* sd = &native_swizzles[i];
76 for(comp = 0; comp < 3; ++comp) {
77 unsigned int swz = GET_SWZ(swizzle, comp);
78 if (swz == RC_SWIZZLE_UNUSED)
79 continue;
80 if (swz != GET_SWZ(sd->hash, comp))
81 break;
82 }
83 if (comp == 3)
84 return sd;
85 }
86
87 return 0;
88 }
89
90 /**
91 * Determines if the given swizzle is valid for r300/r400. In most situations
92 * it is better to use r300_swizzle_is_native() which can be accesed via
93 * struct radeon_compiler *c; c->SwizzleCaps->IsNative().
94 */
r300_swizzle_is_native_basic(unsigned int swizzle)95 int r300_swizzle_is_native_basic(unsigned int swizzle)
96 {
97 if(lookup_native_swizzle(swizzle))
98 return 1;
99 else
100 return 0;
101 }
102
103 /**
104 * Check whether the given instruction supports the swizzle and negate
105 * combinations in the given source register.
106 */
r300_swizzle_is_native(rc_opcode opcode,struct rc_src_register reg)107 static int r300_swizzle_is_native(rc_opcode opcode, struct rc_src_register reg)
108 {
109 const struct swizzle_data* sd;
110 unsigned int relevant;
111 int j;
112
113 if (opcode == RC_OPCODE_KIL ||
114 opcode == RC_OPCODE_TEX ||
115 opcode == RC_OPCODE_TXB ||
116 opcode == RC_OPCODE_TXP) {
117 if (reg.Abs || reg.Negate)
118 return 0;
119
120 for(j = 0; j < 4; ++j) {
121 unsigned int swz = GET_SWZ(reg.Swizzle, j);
122 if (swz == RC_SWIZZLE_UNUSED)
123 continue;
124 if (swz != j)
125 return 0;
126 }
127
128 return 1;
129 }
130
131 relevant = 0;
132
133 for(j = 0; j < 3; ++j)
134 if (GET_SWZ(reg.Swizzle, j) != RC_SWIZZLE_UNUSED)
135 relevant |= 1 << j;
136
137 if ((reg.Negate & relevant) && ((reg.Negate & relevant) != relevant))
138 return 0;
139
140 sd = lookup_native_swizzle(reg.Swizzle);
141 if (!sd || (reg.File == RC_FILE_PRESUB && sd->srcp_stride == 0))
142 return 0;
143
144 return 1;
145 }
146
147
r300_swizzle_split(struct rc_src_register src,unsigned int mask,struct rc_swizzle_split * split)148 static void r300_swizzle_split(
149 struct rc_src_register src, unsigned int mask,
150 struct rc_swizzle_split * split)
151 {
152 split->NumPhases = 0;
153
154 while(mask) {
155 unsigned int best_matchcount = 0;
156 unsigned int best_matchmask = 0;
157 int i, comp;
158
159 for(i = 0; i < num_native_swizzles; ++i) {
160 const struct swizzle_data *sd = &native_swizzles[i];
161 unsigned int matchcount = 0;
162 unsigned int matchmask = 0;
163 for(comp = 0; comp < 3; ++comp) {
164 unsigned int swz;
165 if (!GET_BIT(mask, comp))
166 continue;
167 swz = GET_SWZ(src.Swizzle, comp);
168 if (swz == RC_SWIZZLE_UNUSED)
169 continue;
170 if (swz == GET_SWZ(sd->hash, comp)) {
171 /* check if the negate bit of current component
172 * is the same for already matched components */
173 if (matchmask && (!!(src.Negate & matchmask) != !!(src.Negate & (1 << comp))))
174 continue;
175
176 matchcount++;
177 matchmask |= 1 << comp;
178 }
179 }
180 if (matchcount > best_matchcount) {
181 best_matchcount = matchcount;
182 best_matchmask = matchmask;
183 if (matchmask == (mask & RC_MASK_XYZ))
184 break;
185 }
186 }
187
188 if (mask & RC_MASK_W)
189 best_matchmask |= RC_MASK_W;
190
191 split->Phase[split->NumPhases++] = best_matchmask;
192 mask &= ~best_matchmask;
193 }
194 }
195
196 struct rc_swizzle_caps r300_swizzle_caps = {
197 .IsNative = r300_swizzle_is_native,
198 .Split = r300_swizzle_split
199 };
200
201
202 /**
203 * Translate an RGB (XYZ) swizzle into the hardware code for the given
204 * instruction source.
205 */
r300FPTranslateRGBSwizzle(unsigned int src,unsigned int swizzle)206 unsigned int r300FPTranslateRGBSwizzle(unsigned int src, unsigned int swizzle)
207 {
208 const struct swizzle_data* sd = lookup_native_swizzle(swizzle);
209
210 if (!sd || (src == RC_PAIR_PRESUB_SRC && sd->srcp_stride == 0)) {
211 fprintf(stderr, "Not a native swizzle: %08x\n", swizzle);
212 return 0;
213 }
214
215 if (src == RC_PAIR_PRESUB_SRC) {
216 return sd->base + sd->srcp_stride;
217 } else {
218 return sd->base + src*sd->stride;
219 }
220 }
221
222
223 /**
224 * Translate an Alpha (W) swizzle into the hardware code for the given
225 * instruction source.
226 */
r300FPTranslateAlphaSwizzle(unsigned int src,unsigned int swizzle)227 unsigned int r300FPTranslateAlphaSwizzle(unsigned int src, unsigned int swizzle)
228 {
229 unsigned int swz = GET_SWZ(swizzle, 0);
230 if (src == RC_PAIR_PRESUB_SRC) {
231 return R300_ALU_ARGA_SRCP_X + swz;
232 }
233 if (swz < 3)
234 return swz + 3*src;
235
236 switch(swz) {
237 case RC_SWIZZLE_W: return R300_ALU_ARGA_SRC0A + src;
238 case RC_SWIZZLE_ONE: return R300_ALU_ARGA_ONE;
239 case RC_SWIZZLE_ZERO: return R300_ALU_ARGA_ZERO;
240 case RC_SWIZZLE_HALF: return R300_ALU_ARGA_HALF;
241 default: return R300_ALU_ARGA_ONE;
242 }
243 }
244