i915_fpc_emit.c revision 4a796264dfc1bdba37a7204f3439a8da213109ed 
1/**************************************************************************
2 *
3 * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
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 above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
16 * of the Software.
17 *
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
21 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
25 *
26 **************************************************************************/
27
28#include "i915_reg.h"
29#include "i915_context.h"
30#include "i915_fpc.h"
31#include "util/u_math.h"
32
33
34#define A0_DEST( reg ) (((reg)&UREG_TYPE_NR_MASK)>>UREG_A0_DEST_SHIFT_LEFT)
35#define D0_DEST( reg ) (((reg)&UREG_TYPE_NR_MASK)>>UREG_A0_DEST_SHIFT_LEFT)
36#define T0_DEST( reg ) (((reg)&UREG_TYPE_NR_MASK)>>UREG_A0_DEST_SHIFT_LEFT)
37#define A0_SRC0( reg ) (((reg)&UREG_MASK)>>UREG_A0_SRC0_SHIFT_LEFT)
38#define A1_SRC0( reg ) (((reg)&UREG_MASK)<<UREG_A1_SRC0_SHIFT_RIGHT)
39#define A1_SRC1( reg ) (((reg)&UREG_MASK)>>UREG_A1_SRC1_SHIFT_LEFT)
40#define A2_SRC1( reg ) (((reg)&UREG_MASK)<<UREG_A2_SRC1_SHIFT_RIGHT)
41#define A2_SRC2( reg ) (((reg)&UREG_MASK)>>UREG_A2_SRC2_SHIFT_LEFT)
42
43/* These are special, and don't have swizzle/negate bits.
44 */
45#define T0_SAMPLER( reg )     (GET_UREG_NR(reg)<<T0_SAMPLER_NR_SHIFT)
46#define T1_ADDRESS_REG( reg ) ((GET_UREG_NR(reg)<<T1_ADDRESS_REG_NR_SHIFT) | \
47			       (GET_UREG_TYPE(reg)<<T1_ADDRESS_REG_TYPE_SHIFT))
48
49
50/* Macros for translating UREG's into the various register fields used
51 * by the I915 programmable unit.
52 */
53#define UREG_A0_DEST_SHIFT_LEFT  (UREG_TYPE_SHIFT - A0_DEST_TYPE_SHIFT)
54#define UREG_A0_SRC0_SHIFT_LEFT  (UREG_TYPE_SHIFT - A0_SRC0_TYPE_SHIFT)
55#define UREG_A1_SRC0_SHIFT_RIGHT (A1_SRC0_CHANNEL_W_SHIFT - UREG_CHANNEL_W_SHIFT)
56#define UREG_A1_SRC1_SHIFT_LEFT  (UREG_TYPE_SHIFT - A1_SRC1_TYPE_SHIFT)
57#define UREG_A2_SRC1_SHIFT_RIGHT (A2_SRC1_CHANNEL_W_SHIFT - UREG_CHANNEL_W_SHIFT)
58#define UREG_A2_SRC2_SHIFT_LEFT  (UREG_TYPE_SHIFT - A2_SRC2_TYPE_SHIFT)
59
60#define UREG_MASK         0xffffff00
61#define UREG_TYPE_NR_MASK ((REG_TYPE_MASK << UREG_TYPE_SHIFT) | \
62  			   (REG_NR_MASK << UREG_NR_SHIFT))
63
64
65uint
66i915_get_temp(struct i915_fp_compile *p)
67{
68   int bit = ffs(~p->temp_flag);
69   if (!bit) {
70      i915_program_error(p, "i915_get_temp: out of temporaries\n");
71      return 0;
72   }
73
74   p->temp_flag |= 1 << (bit - 1);
75   return bit - 1;
76}
77
78
79static void
80i915_release_temp(struct i915_fp_compile *p, int reg)
81{
82   p->temp_flag &= ~(1 << reg);
83}
84
85
86/**
87 * Get unpreserved temporary, a temp whose value is not preserved between
88 * PS program phases.
89 */
90uint
91i915_get_utemp(struct i915_fp_compile * p)
92{
93   int bit = ffs(~p->utemp_flag);
94   if (!bit) {
95      i915_program_error(p, "i915_get_utemp: out of temporaries\n");
96      return 0;
97   }
98
99   p->utemp_flag |= 1 << (bit - 1);
100   return UREG(REG_TYPE_U, (bit - 1));
101}
102
103void
104i915_release_utemps(struct i915_fp_compile *p)
105{
106   p->utemp_flag = ~0x7;
107}
108
109
110uint
111i915_emit_decl(struct i915_fp_compile *p,
112               uint type, uint nr, uint d0_flags)
113{
114   uint reg = UREG(type, nr);
115
116   if (type == REG_TYPE_T) {
117      if (p->decl_t & (1 << nr))
118         return reg;
119
120      p->decl_t |= (1 << nr);
121   }
122   else if (type == REG_TYPE_S) {
123      if (p->decl_s & (1 << nr))
124         return reg;
125
126      p->decl_s |= (1 << nr);
127   }
128   else
129      return reg;
130
131   *(p->decl++) = (D0_DCL | D0_DEST(reg) | d0_flags);
132   *(p->decl++) = D1_MBZ;
133   *(p->decl++) = D2_MBZ;
134
135   p->nr_decl_insn++;
136   return reg;
137}
138
139uint
140i915_emit_arith(struct i915_fp_compile * p,
141                uint op,
142                uint dest,
143                uint mask,
144                uint saturate, uint src0, uint src1, uint src2)
145{
146   uint c[3];
147   uint nr_const = 0;
148
149   assert(GET_UREG_TYPE(dest) != REG_TYPE_CONST);
150   dest = UREG(GET_UREG_TYPE(dest), GET_UREG_NR(dest));
151   assert(dest);
152
153   if (GET_UREG_TYPE(src0) == REG_TYPE_CONST)
154      c[nr_const++] = 0;
155   if (GET_UREG_TYPE(src1) == REG_TYPE_CONST)
156      c[nr_const++] = 1;
157   if (GET_UREG_TYPE(src2) == REG_TYPE_CONST)
158      c[nr_const++] = 2;
159
160   /* Recursively call this function to MOV additional const values
161    * into temporary registers.  Use utemp registers for this -
162    * currently shouldn't be possible to run out, but keep an eye on
163    * this.
164    */
165   if (nr_const > 1) {
166      uint s[3], first, i, old_utemp_flag;
167
168      s[0] = src0;
169      s[1] = src1;
170      s[2] = src2;
171      old_utemp_flag = p->utemp_flag;
172
173      first = GET_UREG_NR(s[c[0]]);
174      for (i = 1; i < nr_const; i++) {
175         if (GET_UREG_NR(s[c[i]]) != first) {
176            uint tmp = i915_get_utemp(p);
177
178            i915_emit_arith(p, A0_MOV, tmp, A0_DEST_CHANNEL_ALL, 0,
179                            s[c[i]], 0, 0);
180            s[c[i]] = tmp;
181         }
182      }
183
184      src0 = s[0];
185      src1 = s[1];
186      src2 = s[2];
187      p->utemp_flag = old_utemp_flag;   /* restore */
188   }
189
190   *(p->csr++) = (op | A0_DEST(dest) | mask | saturate | A0_SRC0(src0));
191   *(p->csr++) = (A1_SRC0(src0) | A1_SRC1(src1));
192   *(p->csr++) = (A2_SRC1(src1) | A2_SRC2(src2));
193
194   p->nr_alu_insn++;
195   return dest;
196}
197
198
199/**
200 * Emit a texture load or texkill instruction.
201 * \param dest  the dest i915 register
202 * \param destmask  the dest register writemask
203 * \param sampler  the i915 sampler register
204 * \param coord  the i915 source texcoord operand
205 * \param opcode  the instruction opcode
206 */
207uint i915_emit_texld( struct i915_fp_compile *p,
208			uint dest,
209			uint destmask,
210			uint sampler,
211			uint coord,
212			uint opcode )
213{
214   const uint k = UREG(GET_UREG_TYPE(coord), GET_UREG_NR(coord));
215   int temp = -1;
216
217   if (coord != k) {
218      /* texcoord is swizzled or negated.  Need to allocate a new temporary
219       * register (a utemp / unpreserved temp) won't do.
220       */
221      uint tempReg;
222
223      temp = i915_get_temp(p);           /* get temp reg index */
224      tempReg = UREG(REG_TYPE_R, temp);  /* make i915 register */
225
226      i915_emit_arith( p, A0_MOV,
227                       tempReg, A0_DEST_CHANNEL_ALL, /* dest reg, writemask */
228                       0,                            /* saturate */
229                       coord, 0, 0 );                /* src0, src1, src2 */
230
231      /* new src texcoord is tempReg */
232      coord = tempReg;
233   }
234
235   /* Don't worry about saturate as we only support
236    */
237   if (destmask != A0_DEST_CHANNEL_ALL) {
238      /* if not writing to XYZW... */
239      uint tmp = i915_get_utemp(p);
240      i915_emit_texld( p, tmp, A0_DEST_CHANNEL_ALL, sampler, coord, opcode );
241      i915_emit_arith( p, A0_MOV, dest, destmask, 0, tmp, 0, 0 );
242      /* XXX release utemp here? */
243   }
244   else {
245      assert(GET_UREG_TYPE(dest) != REG_TYPE_CONST);
246      assert(dest == UREG(GET_UREG_TYPE(dest), GET_UREG_NR(dest)));
247
248      /* is the sampler coord a texcoord input reg? */
249      if (GET_UREG_TYPE(coord) != REG_TYPE_T) {
250	 p->nr_tex_indirect++;
251      }
252
253      *(p->csr++) = (opcode |
254		     T0_DEST( dest ) |
255		     T0_SAMPLER( sampler ));
256
257      *(p->csr++) = T1_ADDRESS_REG( coord );
258      *(p->csr++) = T2_MBZ;
259
260      p->nr_tex_insn++;
261   }
262
263   if (temp >= 0)
264      i915_release_temp(p, temp);
265
266   return dest;
267}
268
269
270uint
271i915_emit_const1f(struct i915_fp_compile * p, float c0)
272{
273   struct i915_fragment_shader *ifs = p->shader;
274   unsigned reg, idx;
275
276   if (c0 == 0.0)
277      return swizzle(UREG(REG_TYPE_R, 0), ZERO, ZERO, ZERO, ZERO);
278   if (c0 == 1.0)
279      return swizzle(UREG(REG_TYPE_R, 0), ONE, ONE, ONE, ONE);
280
281   for (reg = 0; reg < I915_MAX_CONSTANT; reg++) {
282      if (ifs->constant_flags[reg] == I915_CONSTFLAG_USER)
283         continue;
284      for (idx = 0; idx < 4; idx++) {
285         if (!(ifs->constant_flags[reg] & (1 << idx)) ||
286             ifs->constants[reg][idx] == c0) {
287            ifs->constants[reg][idx] = c0;
288            ifs->constant_flags[reg] |= 1 << idx;
289            if (reg + 1 > ifs->num_constants)
290               ifs->num_constants = reg + 1;
291            return swizzle(UREG(REG_TYPE_CONST, reg), idx, ZERO, ZERO, ONE);
292         }
293      }
294   }
295
296   i915_program_error(p, "i915_emit_const1f: out of constants\n");
297   return 0;
298}
299
300uint
301i915_emit_const2f(struct i915_fp_compile * p, float c0, float c1)
302{
303   struct i915_fragment_shader *ifs = p->shader;
304   unsigned reg, idx;
305
306   if (c0 == 0.0)
307      return swizzle(i915_emit_const1f(p, c1), ZERO, X, Z, W);
308   if (c0 == 1.0)
309      return swizzle(i915_emit_const1f(p, c1), ONE, X, Z, W);
310
311   if (c1 == 0.0)
312      return swizzle(i915_emit_const1f(p, c0), X, ZERO, Z, W);
313   if (c1 == 1.0)
314      return swizzle(i915_emit_const1f(p, c0), X, ONE, Z, W);
315
316   for (reg = 0; reg < I915_MAX_CONSTANT; reg++) {
317      if (ifs->constant_flags[reg] == 0xf ||
318          ifs->constant_flags[reg] == I915_CONSTFLAG_USER)
319         continue;
320      for (idx = 0; idx < 3; idx++) {
321         if (!(ifs->constant_flags[reg] & (3 << idx))) {
322            ifs->constants[reg][idx + 0] = c0;
323            ifs->constants[reg][idx + 1] = c1;
324            ifs->constant_flags[reg] |= 3 << idx;
325            if (reg + 1 > ifs->num_constants)
326               ifs->num_constants = reg + 1;
327            return swizzle(UREG(REG_TYPE_CONST, reg), idx, idx + 1, ZERO, ONE);
328         }
329      }
330   }
331
332   i915_program_error(p, "i915_emit_const2f: out of constants\n");
333   return 0;
334}
335
336
337
338uint
339i915_emit_const4f(struct i915_fp_compile * p,
340                  float c0, float c1, float c2, float c3)
341{
342   struct i915_fragment_shader *ifs = p->shader;
343   unsigned reg;
344
345   for (reg = 0; reg < I915_MAX_CONSTANT; reg++) {
346      if (ifs->constant_flags[reg] == 0xf &&
347          ifs->constants[reg][0] == c0 &&
348          ifs->constants[reg][1] == c1 &&
349          ifs->constants[reg][2] == c2 &&
350          ifs->constants[reg][3] == c3) {
351         return UREG(REG_TYPE_CONST, reg);
352      }
353      else if (ifs->constant_flags[reg] == 0) {
354
355         ifs->constants[reg][0] = c0;
356         ifs->constants[reg][1] = c1;
357         ifs->constants[reg][2] = c2;
358         ifs->constants[reg][3] = c3;
359         ifs->constant_flags[reg] = 0xf;
360         if (reg + 1 > ifs->num_constants)
361            ifs->num_constants = reg + 1;
362         return UREG(REG_TYPE_CONST, reg);
363      }
364   }
365
366   i915_program_error(p, "i915_emit_const4f: out of constants\n");
367   return 0;
368}
369
370
371uint
372i915_emit_const4fv(struct i915_fp_compile * p, const float * c)
373{
374   return i915_emit_const4f(p, c[0], c[1], c[2], c[3]);
375}
376