1 /*
2  * Copyright © 2014 Intel Corporation
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice (including the next
12  * paragraph) shall be included in all copies or substantial portions of the
13  * Software.
14  *
15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21  * IN THE SOFTWARE.
22  */
23 
24 /**
25  * @file brw_inst.h
26  *
27  * A representation of i965 EU assembly instructions, with helper methods to
28  * get and set various fields.  This is the actual hardware format.
29  */
30 
31 #ifndef BRW_INST_H
32 #define BRW_INST_H
33 
34 #include <assert.h>
35 #include <stdint.h>
36 
37 #include "brw_eu_defines.h"
38 #include "brw_reg_type.h"
39 #include "dev/intel_device_info.h"
40 
41 #ifdef __cplusplus
42 extern "C" {
43 #endif
44 
45 /* brw_context.h has a forward declaration of brw_inst, so name the struct. */
46 typedef struct brw_inst {
47    uint64_t data[2];
48 } brw_inst;
49 
50 static inline uint64_t brw_inst_bits(const brw_inst *inst,
51                                      unsigned high, unsigned low);
52 static inline void brw_inst_set_bits(brw_inst *inst,
53                                      unsigned high, unsigned low,
54                                      uint64_t value);
55 
56 #define FC(name, hi4, lo4, hi12, lo12, assertions)            \
57 static inline void                                            \
58 brw_inst_set_##name(const struct intel_device_info *devinfo,  \
59                     brw_inst *inst, uint64_t v)               \
60 {                                                             \
61    assert(assertions);                                        \
62    if (devinfo->ver >= 12)                                    \
63       brw_inst_set_bits(inst, hi12, lo12, v);                 \
64    else                                                       \
65       brw_inst_set_bits(inst, hi4, lo4, v);                   \
66 }                                                             \
67 static inline uint64_t                                        \
68 brw_inst_##name(const struct intel_device_info *devinfo,      \
69                 const brw_inst *inst)                         \
70 {                                                             \
71    assert(assertions);                                        \
72    if (devinfo->ver >= 12)                                    \
73       return brw_inst_bits(inst, hi12, lo12);                 \
74    else                                                       \
75       return brw_inst_bits(inst, hi4, lo4);                   \
76 }
77 
78 /* A simple macro for fields which stay in the same place on all generations,
79  * except for Gfx12!
80  */
81 #define F(name, hi4, lo4, hi12, lo12) FC(name, hi4, lo4, hi12, lo12, true)
82 
83 #define BOUNDS(hi4, lo4, hi45, lo45, hi5, lo5, hi6, lo6,                     \
84                hi7, lo7, hi8, lo8, hi12, lo12)                               \
85    unsigned high, low;                                                       \
86    if (devinfo->ver >= 12) {                                                 \
87       high = hi12; low = lo12;                                               \
88    } else if (devinfo->ver >= 8) {                                           \
89       high = hi8;  low = lo8;                                                \
90    } else if (devinfo->ver >= 7) {                                           \
91       high = hi7;  low = lo7;                                                \
92    } else if (devinfo->ver >= 6) {                                           \
93       high = hi6;  low = lo6;                                                \
94    } else if (devinfo->ver >= 5) {                                           \
95       high = hi5;  low = lo5;                                                \
96    } else if (devinfo->is_g4x) {                                             \
97       high = hi45; low = lo45;                                               \
98    } else {                                                                  \
99       high = hi4;  low = lo4;                                                \
100    }                                                                         \
101    assert(((int) high) != -1 && ((int) low) != -1);
102 
103 /* A general macro for cases where the field has moved to several different
104  * bit locations across generations.  GCC appears to combine cases where the
105  * bits are identical, removing some of the inefficiency.
106  */
107 #define FF(name, hi4, lo4, hi45, lo45, hi5, lo5, hi6, lo6,                    \
108            hi7, lo7, hi8, lo8, hi12, lo12)                                    \
109 static inline void                                                            \
110 brw_inst_set_##name(const struct intel_device_info *devinfo,                  \
111                     brw_inst *inst, uint64_t value)                           \
112 {                                                                             \
113    BOUNDS(hi4, lo4, hi45, lo45, hi5, lo5, hi6, lo6,                           \
114           hi7, lo7, hi8, lo8, hi12, lo12)                                     \
115    brw_inst_set_bits(inst, high, low, value);                                 \
116 }                                                                             \
117 static inline uint64_t                                                        \
118 brw_inst_##name(const struct intel_device_info *devinfo, const brw_inst *inst)\
119 {                                                                             \
120    BOUNDS(hi4, lo4, hi45, lo45, hi5, lo5, hi6, lo6,                           \
121           hi7, lo7, hi8, lo8, hi12, lo12)                                     \
122    return brw_inst_bits(inst, high, low);                                     \
123 }
124 
125 /* A macro for fields which moved as of Gfx8+. */
126 #define F8(name, gfx4_high, gfx4_low, gfx8_high, gfx8_low, \
127            gfx12_high, gfx12_low)                          \
128 FF(name,                                                   \
129    /* 4:   */ gfx4_high, gfx4_low,                         \
130    /* 4.5: */ gfx4_high, gfx4_low,                         \
131    /* 5:   */ gfx4_high, gfx4_low,                         \
132    /* 6:   */ gfx4_high, gfx4_low,                         \
133    /* 7:   */ gfx4_high, gfx4_low,                         \
134    /* 8:   */ gfx8_high, gfx8_low,                         \
135    /* 12:  */ gfx12_high, gfx12_low);
136 
137 /* Macro for fields that gained extra discontiguous MSBs in Gfx12 (specified
138  * by hi12ex-lo12ex).
139  */
140 #define FFDC(name, hi4, lo4, hi45, lo45, hi5, lo5, hi6, lo6,                  \
141              hi7, lo7, hi8, lo8, hi12ex, lo12ex, hi12, lo12, assertions)      \
142 static inline void                                                            \
143 brw_inst_set_##name(const struct intel_device_info *devinfo,                  \
144                     brw_inst *inst, uint64_t value)                           \
145 {                                                                             \
146    assert(assertions);                                                        \
147    if (devinfo->ver >= 12) {                                                  \
148       const unsigned k = hi12 - lo12 + 1;                                     \
149       if (hi12ex != -1 && lo12ex != -1)                                       \
150          brw_inst_set_bits(inst, hi12ex, lo12ex, value >> k);                 \
151       brw_inst_set_bits(inst, hi12, lo12, value & ((1ull << k) - 1));         \
152    } else {                                                                   \
153       BOUNDS(hi4, lo4, hi45, lo45, hi5, lo5, hi6, lo6,                        \
154              hi7, lo7, hi8, lo8, -1, -1);                                     \
155       brw_inst_set_bits(inst, high, low, value);                              \
156    }                                                                          \
157 }                                                                             \
158 static inline uint64_t                                                        \
159 brw_inst_##name(const struct intel_device_info *devinfo, const brw_inst *inst)\
160 {                                                                             \
161    assert(assertions);                                                        \
162    if (devinfo->ver >= 12) {                                                  \
163       const unsigned k = hi12 - lo12 + 1;                                     \
164       return (hi12ex == -1 || lo12ex == -1 ? 0 :                              \
165               brw_inst_bits(inst, hi12ex, lo12ex) << k) |                     \
166              brw_inst_bits(inst, hi12, lo12);                                 \
167    } else {                                                                   \
168       BOUNDS(hi4, lo4, hi45, lo45, hi5, lo5, hi6, lo6,                        \
169              hi7, lo7, hi8, lo8, -1, -1);                                     \
170       return brw_inst_bits(inst, high, low);                                  \
171    }                                                                          \
172 }
173 
174 #define FD(name, hi4, lo4, hi45, lo45, hi5, lo5, hi6, lo6,        \
175            hi7, lo7, hi8, lo8, hi12ex, lo12ex, hi12, lo12)        \
176    FFDC(name, hi4, lo4, hi45, lo45, hi5, lo5, hi6, lo6,           \
177         hi7, lo7, hi8, lo8, hi12ex, lo12ex, hi12, lo12, true)
178 
179 /* Macro for fields that didn't move across generations until Gfx12, and then
180  * gained extra discontiguous bits.
181  */
182 #define FDC(name, hi4, lo4, hi12ex, lo12ex, hi12, lo12, assertions)     \
183    FFDC(name, hi4, lo4, hi4, lo4, hi4, lo4, hi4, lo4,                   \
184         hi4, lo4, hi4, lo4, hi12ex, lo12ex, hi12, lo12, assertions)
185 
186 
187 /* Macro for the 2-bit register file field, which on Gfx12+ is stored as the
188  * variable length combination of an IsImm (hi12) bit and an additional file
189  * (lo12) bit.
190  */
191 #define FI(name, hi4, lo4, hi8, lo8, hi12, lo12)                              \
192 static inline void                                                            \
193 brw_inst_set_##name(const struct intel_device_info *devinfo,                  \
194                     brw_inst *inst, uint64_t value)                           \
195 {                                                                             \
196    if (devinfo->ver >= 12) {                                                  \
197       brw_inst_set_bits(inst, hi12, hi12, value >> 1);                        \
198       if ((value >> 1) == 0)                                                  \
199          brw_inst_set_bits(inst, lo12, lo12, value & 1);                      \
200    } else {                                                                   \
201       BOUNDS(hi4, lo4, hi4, lo4, hi4, lo4, hi4, lo4,                          \
202              hi4, lo4, hi8, lo8, -1, -1);                                     \
203       brw_inst_set_bits(inst, high, low, value);                              \
204    }                                                                          \
205 }                                                                             \
206 static inline uint64_t                                                        \
207 brw_inst_##name(const struct intel_device_info *devinfo, const brw_inst *inst)\
208 {                                                                             \
209    if (devinfo->ver >= 12) {                                                  \
210       return (brw_inst_bits(inst, hi12, hi12) << 1) |                         \
211              (brw_inst_bits(inst, hi12, hi12) == 0 ?                          \
212               brw_inst_bits(inst, lo12, lo12) : 1);                           \
213    } else {                                                                   \
214       BOUNDS(hi4, lo4, hi4, lo4, hi4, lo4, hi4, lo4,                          \
215              hi4, lo4, hi8, lo8, -1, -1);                                     \
216       return brw_inst_bits(inst, high, low);                                  \
217    }                                                                          \
218 }
219 
220 /* Macro for fields that become a constant in Gfx12+ not actually represented
221  * in the instruction.
222  */
223 #define FK(name, hi4, lo4, const12)                           \
224 static inline void                                            \
225 brw_inst_set_##name(const struct intel_device_info *devinfo,  \
226                     brw_inst *inst, uint64_t v)               \
227 {                                                             \
228    if (devinfo->ver >= 12)                                    \
229       assert(v == (const12));                                 \
230    else                                                       \
231       brw_inst_set_bits(inst, hi4, lo4, v);                   \
232 }                                                             \
233 static inline uint64_t                                        \
234 brw_inst_##name(const struct intel_device_info *devinfo,      \
235                 const brw_inst *inst)                         \
236 {                                                             \
237    if (devinfo->ver >= 12)                                    \
238       return (const12);                                       \
239    else                                                       \
240       return brw_inst_bits(inst, hi4, lo4);                   \
241 }
242 
243 F(src1_vstride,        /* 4+ */ 120, 117, /* 12+ */ 119, 116)
244 F(src1_width,          /* 4+ */ 116, 114, /* 12+ */ 115, 113)
245 F(src1_da16_swiz_w,    /* 4+ */ 115, 114, /* 12+ */ -1, -1)
246 F(src1_da16_swiz_z,    /* 4+ */ 113, 112, /* 12+ */ -1, -1)
247 F(src1_hstride,        /* 4+ */ 113, 112, /* 12+ */ 97, 96)
248 F(src1_address_mode,   /* 4+ */ 111, 111, /* 12+ */ 112, 112)
249 /** Src1.SrcMod @{ */
250 F(src1_negate,         /* 4+ */ 110, 110, /* 12+ */ 121, 121)
251 F(src1_abs,            /* 4+ */ 109, 109, /* 12+ */ 120, 120)
252 /** @} */
253 F8(src1_ia_subreg_nr,  /* 4+ */ 108, 106, /* 8+ */  108, 105, /* 12+ */ 111, 108)
254 F(src1_da_reg_nr,      /* 4+ */ 108, 101, /* 12+ */ 111, 104)
255 F(src1_da16_subreg_nr, /* 4+ */ 100, 100, /* 12+ */ -1, -1)
256 F(src1_da1_subreg_nr,  /* 4+ */ 100,  96, /* 12+ */ 103, 99)
257 F(src1_da16_swiz_y,    /* 4+ */ 99,  98,  /* 12+ */ -1, -1)
258 F(src1_da16_swiz_x,    /* 4+ */ 97,  96,  /* 12+ */ -1, -1)
259 F8(src1_reg_hw_type,   /* 4+ */ 46,  44,  /* 8+ */  94,  91, /* 12+ */ 91, 88)
260 FI(src1_reg_file,      /* 4+ */ 43,  42,  /* 8+ */  90,  89, /* 12+ */ 47, 98)
261 F(src1_is_imm,         /* 4+ */ -1,  -1,  /* 12+ */ 47, 47)
262 F(src0_vstride,        /* 4+ */ 88,  85,  /* 12+ */ 87, 84)
263 F(src0_width,          /* 4+ */ 84,  82,  /* 12+ */ 83, 81)
264 F(src0_da16_swiz_w,    /* 4+ */ 83,  82,  /* 12+ */ -1, -1)
265 F(src0_da16_swiz_z,    /* 4+ */ 81,  80,  /* 12+ */ -1, -1)
266 F(src0_hstride,        /* 4+ */ 81,  80,  /* 12+ */ 65, 64)
267 F(src0_address_mode,   /* 4+ */ 79,  79,  /* 12+ */ 80, 80)
268 /** Src0.SrcMod @{ */
269 F(src0_negate,         /* 4+ */ 78,  78,  /* 12+ */ 45, 45)
270 F(src0_abs,            /* 4+ */ 77,  77,  /* 12+ */ 44, 44)
271 /** @} */
272 F8(src0_ia_subreg_nr,  /* 4+ */ 76,  74,  /* 8+ */  76,  73, /* 12+ */ 79, 76)
273 F(src0_da_reg_nr,      /* 4+ */ 76,  69,  /* 12+ */ 79, 72)
274 F(src0_da16_subreg_nr, /* 4+ */ 68,  68,  /* 12+ */ -1, -1)
275 F(src0_da1_subreg_nr,  /* 4+ */ 68,  64,  /* 12+ */ 71, 67)
276 F(src0_da16_swiz_y,    /* 4+ */ 67,  66,  /* 12+ */ -1, -1)
277 F(src0_da16_swiz_x,    /* 4+ */ 65,  64,  /* 12+ */ -1, -1)
278 F(dst_address_mode,    /* 4+ */ 63,  63,  /* 12+ */ 35, 35)
279 F(dst_hstride,         /* 4+ */ 62,  61,  /* 12+ */ 49, 48)
280 F8(dst_ia_subreg_nr,   /* 4+ */ 60,  58,  /* 8+ */  60,  57, /* 12+ */ 63, 60)
281 F(dst_da_reg_nr,       /* 4+ */ 60,  53,  /* 12+ */ 63, 56)
282 F(dst_da16_subreg_nr,  /* 4+ */ 52,  52,  /* 12+ */ -1, -1)
283 F(dst_da1_subreg_nr,   /* 4+ */ 52,  48,  /* 12+ */ 55, 51)
284 F(da16_writemask,      /* 4+ */ 51,  48,  /* 12+ */ -1, -1) /* Dst.ChanEn */
285 F8(src0_reg_hw_type,   /* 4+ */ 41,  39,  /* 8+ */  46,  43, /* 12+ */ 43, 40)
286 FI(src0_reg_file,      /* 4+ */ 38,  37,  /* 8+ */  42,  41, /* 12+ */ 46, 66)
287 F(src0_is_imm,         /* 4+ */ -1,  -1,  /* 12+ */ 46, 46)
288 F8(dst_reg_hw_type,    /* 4+ */ 36,  34,  /* 8+ */  40,  37, /* 12+ */ 39, 36)
289 F8(dst_reg_file,       /* 4+ */ 33,  32,  /* 8+ */  36,  35, /* 12+ */ 50, 50)
290 F8(mask_control,       /* 4+ */  9,   9,  /* 8+ */  34,  34, /* 12+ */ 31, 31)
291 FF(flag_reg_nr,
292    /* 4-6: doesn't exist */ -1, -1, -1, -1, -1, -1, -1, -1,
293    /* 7: */ 90, 90,
294    /* 8: */ 33, 33,
295    /* 12: */ 23, 23)
296 F8(flag_subreg_nr,     /* 4+ */ 89,  89,  /* 8+ */ 32, 32,   /* 12+ */ 22, 22)
297 F(saturate,            /* 4+ */ 31,  31,  /* 12+ */ 34, 34)
298 F(debug_control,       /* 4+ */ 30,  30,  /* 12+ */ 30, 30)
299 F(cmpt_control,        /* 4+ */ 29,  29,  /* 12+ */ 29, 29)
300 FC(branch_control,     /* 4+ */ 28,  28,  /* 12+ */ 33, 33, devinfo->ver >= 8)
301 FC(acc_wr_control,     /* 4+ */ 28,  28,  /* 12+ */ 33, 33, devinfo->ver >= 6)
302 FC(mask_control_ex,    /* 4+ */ 28,  28,  /* 12+ */ -1, -1, devinfo->is_g4x || devinfo->ver == 5)
303 F(cond_modifier,       /* 4+ */ 27,  24,  /* 12+ */ 95, 92)
304 FC(math_function,      /* 4+ */ 27,  24,  /* 12+ */ 95, 92, devinfo->ver >= 6)
305 F(exec_size,           /* 4+ */ 23,  21,  /* 12+ */ 18, 16)
306 F(pred_inv,            /* 4+ */ 20,  20,  /* 12+ */ 28, 28)
307 F(pred_control,        /* 4+ */ 19,  16,  /* 12+ */ 27, 24)
308 F(thread_control,      /* 4+ */ 15,  14,  /* 12+ */ -1, -1)
309 F(atomic_control,      /* 4+ */ -1,  -1,  /* 12+ */ 32, 32)
310 F(qtr_control,         /* 4+ */ 13,  12,  /* 12+ */ 21, 20)
311 FF(nib_control,
312    /* 4-6: doesn't exist */ -1, -1, -1, -1, -1, -1, -1, -1,
313    /* 7: */ 47, 47,
314    /* 8: */ 11, 11,
315    /* 12: */ 19, 19)
316 F8(no_dd_check,        /* 4+ */  11, 11,  /* 8+ */  10,  10, /* 12+ */ -1, -1)
317 F8(no_dd_clear,        /* 4+ */  10, 10,  /* 8+ */   9,   9, /* 12+ */ -1, -1)
318 F(swsb,                /* 4+ */  -1, -1,  /* 12+ */ 15,  8)
319 FK(access_mode,        /* 4+ */   8,  8,  /* 12+ */ BRW_ALIGN_1)
320 /* Bit 7 is Reserved (for future Opcode expansion) */
321 F(hw_opcode,           /* 4+ */   6,  0,  /* 12+ */ 6,  0)
322 
323 /**
324  * Three-source instructions:
325  *  @{
326  */
327 F(3src_src2_reg_nr,         /* 4+ */ 125, 118, /* 12+ */ 127, 120) /* same in align1 */
328 F(3src_a16_src2_subreg_nr,  /* 4+ */ 117, 115, /* 12+ */ -1, -1) /* Extra discontiguous bit on CHV? */
329 F(3src_a16_src2_swizzle,    /* 4+ */ 114, 107, /* 12+ */ -1, -1)
330 F(3src_a16_src2_rep_ctrl,   /* 4+ */ 106, 106, /* 12+ */ -1, -1)
331 F(3src_src1_reg_nr,         /* 4+ */ 104,  97, /* 12+ */ 111, 104) /* same in align1 */
332 F(3src_a16_src1_subreg_nr,  /* 4+ */ 96,  94,  /* 12+ */ -1, -1) /* Extra discontiguous bit on CHV? */
333 F(3src_a16_src1_swizzle,    /* 4+ */ 93,  86,  /* 12+ */ -1, -1)
334 F(3src_a16_src1_rep_ctrl,   /* 4+ */ 85,  85,  /* 12+ */ -1, -1)
335 F(3src_src0_reg_nr,         /* 4+ */ 83,  76,  /* 12+ */ 79, 72) /* same in align1 */
336 F(3src_a16_src0_subreg_nr,  /* 4+ */ 75,  73,  /* 12+ */ -1, -1) /* Extra discontiguous bit on CHV? */
337 F(3src_a16_src0_swizzle,    /* 4+ */ 72,  65,  /* 12+ */ -1, -1)
338 F(3src_a16_src0_rep_ctrl,   /* 4+ */ 64,  64,  /* 12+ */ -1, -1)
339 F(3src_dst_reg_nr,          /* 4+ */ 63,  56,  /* 12+ */ 63, 56) /* same in align1 */
340 F(3src_a16_dst_subreg_nr,   /* 4+ */ 55,  53,  /* 12+ */ -1, -1)
341 F(3src_a16_dst_writemask,   /* 4+ */ 52,  49,  /* 12+ */ -1, -1)
342 F8(3src_a16_nib_ctrl,       /* 4+ */ 47, 47,   /* 8+ */  11, 11, /* 12+ */ -1, -1) /* only exists on IVB+ */
343 F8(3src_a16_dst_hw_type,    /* 4+ */ 45, 44,   /* 8+ */  48, 46, /* 12+ */ -1, -1) /* only exists on IVB+ */
344 F8(3src_a16_src_hw_type,    /* 4+ */ 43, 42,   /* 8+ */  45, 43, /* 12+ */ -1, -1)
345 F8(3src_src2_negate,        /* 4+ */ 41, 41,   /* 8+ */  42, 42, /* 12+ */ 85, 85)
346 F8(3src_src2_abs,           /* 4+ */ 40, 40,   /* 8+ */  41, 41, /* 12+ */ 84, 84)
347 F8(3src_src1_negate,        /* 4+ */ 39, 39,   /* 8+ */  40, 40, /* 12+ */ 87, 87)
348 F8(3src_src1_abs,           /* 4+ */ 38, 38,   /* 8+ */  39, 39, /* 12+ */ 86, 86)
349 F8(3src_src0_negate,        /* 4+ */ 37, 37,   /* 8+ */  38, 38, /* 12+ */ 45, 45)
350 F8(3src_src0_abs,           /* 4+ */ 36, 36,   /* 8+ */  37, 37, /* 12+ */ 44, 44)
351 F8(3src_a16_src1_type,      /* 4+ */ -1, -1,   /* 8+ */  36, 36, /* 12+ */ -1, -1)
352 F8(3src_a16_src2_type,      /* 4+ */ -1, -1,   /* 8+ */  35, 35, /* 12+ */ -1, -1)
353 F8(3src_a16_flag_reg_nr,    /* 4+ */ 34, 34,   /* 8+ */  33, 33, /* 12+ */ -1, -1)
354 F8(3src_a16_flag_subreg_nr, /* 4+ */ 33, 33,   /* 8+ */  32, 32, /* 12+ */ -1, -1)
355 FF(3src_a16_dst_reg_file,
356    /* 4-5: doesn't exist - no 3-source instructions */ -1, -1, -1, -1, -1, -1,
357    /* 6: */ 32, 32,
358    /* 7-8: doesn't exist - no MRFs */ -1, -1, -1, -1,
359    /* 12: */ -1, -1)
360 F(3src_saturate,            /* 4+ */ 31, 31,   /* 12+ */ 34, 34)
361 F(3src_debug_control,       /* 4+ */ 30, 30,   /* 12+ */ 30, 30)
362 F(3src_cmpt_control,        /* 4+ */ 29, 29,   /* 12+ */ 29, 29)
363 F(3src_acc_wr_control,      /* 4+ */ 28, 28,   /* 12+ */ 33, 33)
364 F(3src_cond_modifier,       /* 4+ */ 27, 24,   /* 12+ */ 95, 92)
365 F(3src_exec_size,           /* 4+ */ 23, 21,   /* 12+ */ 18, 16)
366 F(3src_pred_inv,            /* 4+ */ 20, 20,   /* 12+ */ 28, 28)
367 F(3src_pred_control,        /* 4+ */ 19, 16,   /* 12+ */ 27, 24)
368 F(3src_thread_control,      /* 4+ */ 15, 14,   /* 12+ */ -1, -1)
369 F(3src_atomic_control,      /* 4+ */ -1, -1,   /* 12+ */ 32, 32)
370 F(3src_qtr_control,         /* 4+ */ 13, 12,   /* 12+ */ 21, 20)
371 F8(3src_no_dd_check,        /* 4+ */ 11, 11,   /* 8+ */  10, 10, /* 12+ */ -1, -1)
372 F8(3src_no_dd_clear,        /* 4+ */ 10, 10,   /* 8+ */   9,  9, /* 12+ */ -1, -1)
373 F8(3src_mask_control,       /* 4+ */ 9,  9,    /* 8+ */  34, 34, /* 12+ */ 31, 31)
374 FK(3src_access_mode,        /* 4+ */ 8,  8,    /* 12+ */ BRW_ALIGN_1)
375 F(3src_swsb,                /* 4+ */ -1, -1,   /* 12+ */ 15,  8)
376 /* Bit 7 is Reserved (for future Opcode expansion) */
377 F(3src_hw_opcode,           /* 4+ */ 6,  0,    /* 12+ */ 6, 0)
378 /** @} */
379 
380 #define REG_TYPE(reg)                                                         \
381 static inline void                                                            \
382 brw_inst_set_3src_a16_##reg##_type(const struct intel_device_info *devinfo,   \
383                                    brw_inst *inst, enum brw_reg_type type)    \
384 {                                                                             \
385    unsigned hw_type = brw_reg_type_to_a16_hw_3src_type(devinfo, type);        \
386    brw_inst_set_3src_a16_##reg##_hw_type(devinfo, inst, hw_type);             \
387 }                                                                             \
388                                                                               \
389 static inline enum brw_reg_type                                               \
390 brw_inst_3src_a16_##reg##_type(const struct intel_device_info *devinfo,       \
391                                const brw_inst *inst)                          \
392 {                                                                             \
393    unsigned hw_type = brw_inst_3src_a16_##reg##_hw_type(devinfo, inst);       \
394    return brw_a16_hw_3src_type_to_reg_type(devinfo, hw_type);                 \
395 }
396 
REG_TYPE(dst)397 REG_TYPE(dst)
398 REG_TYPE(src)
399 #undef REG_TYPE
400 
401 /**
402  * Three-source align1 instructions:
403  *  @{
404  */
405 /* Reserved 127:126 */
406 /* src2_reg_nr same in align16 */
407 FC(3src_a1_src2_subreg_nr,  /* 4+ */   117, 113, /* 12+ */ 119, 115, devinfo->ver >= 10)
408 FC(3src_a1_src2_hstride,    /* 4+ */   112, 111, /* 12+ */ 113, 112, devinfo->ver >= 10)
409 /* Reserved 110:109. src2 vstride is an implied parameter */
410 FC(3src_a1_src2_hw_type,    /* 4+ */   108, 106, /* 12+ */ 82, 80, devinfo->ver >= 10)
411 /* Reserved 105 */
412 /* src1_reg_nr same in align16 */
413 FC(3src_a1_src1_subreg_nr,  /* 4+ */   96,  92,  /* 12+ */ 103, 99, devinfo->ver >= 10)
414 FC(3src_a1_src1_hstride,    /* 4+ */   91,  90,  /* 12+ */ 97, 96, devinfo->ver >= 10)
415 FDC(3src_a1_src1_vstride,  /* 4+ */   89,  88,  /* 12+ */ 91, 91, 83, 83, devinfo->ver >= 10)
416 FC(3src_a1_src1_hw_type,    /* 4+ */   87,  85,  /* 12+ */ 90, 88, devinfo->ver >= 10)
417 /* Reserved 84 */
418 /* src0_reg_nr same in align16 */
419 FC(3src_a1_src0_subreg_nr,  /* 4+ */   75,  71,  /* 12+ */ 71, 67, devinfo->ver >= 10)
420 FC(3src_a1_src0_hstride,    /* 4+ */   70,  69,  /* 12+ */ 65, 64, devinfo->ver >= 10)
421 FDC(3src_a1_src0_vstride,  /* 4+ */   68,  67,  /* 12+ */ 43, 43, 35, 35, devinfo->ver >= 10)
422 FC(3src_a1_src0_hw_type,    /* 4+ */   66,  64,  /* 12+ */ 42, 40, devinfo->ver >= 10)
423 /* dst_reg_nr same in align16 */
424 FC(3src_a1_dst_subreg_nr,   /* 4+ */   55,  54,  /* 12+ */ 55, 54, devinfo->ver >= 10)
425 FC(3src_a1_special_acc,     /* 4+ */   55,  52,  /* 12+ */ 54, 51, devinfo->ver >= 10) /* aliases dst_subreg_nr */
426 /* Reserved 51:50 */
427 FC(3src_a1_dst_hstride,     /* 4+ */   49,  49,  /* 12+ */ 48, 48, devinfo->ver >= 10)
428 FC(3src_a1_dst_hw_type,     /* 4+ */   48,  46,  /* 12+ */ 38, 36, devinfo->ver >= 10)
429 FI(3src_a1_src2_reg_file,   /* 4+ */   -1,  -1,  /* 8+ */  45, 45,  /* 12+ */ 47, 114)
430 FC(3src_a1_src1_reg_file,   /* 4+ */   44,  44,  /* 12+ */ 98, 98, devinfo->ver >= 10)
431 FI(3src_a1_src0_reg_file,   /* 4+ */   -1,  -1,  /* 8+ */  43, 43,  /* 12+ */ 46, 66)
432 
433 F(3src_a1_src2_is_imm,      /* 4+ */   -1,  -1,  /* 12+ */ 47, 47)
434 F(3src_a1_src0_is_imm,      /* 4+ */   -1,  -1,  /* 12+ */ 46, 46)
435 
436 /* Source Modifier fields same in align16 */
437 FC(3src_a1_dst_reg_file,    /* 4+ */    36,  36, /* 12+ */ 50, 50, devinfo->ver >= 10)
438 FC(3src_a1_exec_type,       /* 4+ */    35,  35, /* 12+ */ 39, 39, devinfo->ver >= 10)
439 /* Fields below this same in align16 */
440 /** @} */
441 
442 #define REG_TYPE(reg)                                                         \
443 static inline void                                                            \
444 brw_inst_set_3src_a1_##reg##_type(const struct intel_device_info *devinfo,    \
445                                   brw_inst *inst, enum brw_reg_type type)     \
446 {                                                                             \
447    UNUSED enum gfx10_align1_3src_exec_type exec_type =                        \
448       (enum gfx10_align1_3src_exec_type) brw_inst_3src_a1_exec_type(devinfo,  \
449                                                                     inst);    \
450    if (brw_reg_type_is_floating_point(type)) {                                \
451       assert(exec_type == BRW_ALIGN1_3SRC_EXEC_TYPE_FLOAT);                   \
452    } else {                                                                   \
453       assert(exec_type == BRW_ALIGN1_3SRC_EXEC_TYPE_INT);                     \
454    }                                                                          \
455    unsigned hw_type = brw_reg_type_to_a1_hw_3src_type(devinfo, type);         \
456    brw_inst_set_3src_a1_##reg##_hw_type(devinfo, inst, hw_type);              \
457 }                                                                             \
458                                                                               \
459 static inline enum brw_reg_type                                               \
460 brw_inst_3src_a1_##reg##_type(const struct intel_device_info *devinfo,        \
461                               const brw_inst *inst)                           \
462 {                                                                             \
463    enum gfx10_align1_3src_exec_type exec_type =                               \
464       (enum gfx10_align1_3src_exec_type) brw_inst_3src_a1_exec_type(devinfo,  \
465                                                                     inst);    \
466    unsigned hw_type = brw_inst_3src_a1_##reg##_hw_type(devinfo, inst);        \
467    return brw_a1_hw_3src_type_to_reg_type(devinfo, hw_type, exec_type);       \
468 }
469 
470 REG_TYPE(dst)
471 REG_TYPE(src0)
472 REG_TYPE(src1)
473 REG_TYPE(src2)
474 #undef REG_TYPE
475 
476 /**
477  * Three-source align1 instruction immediates:
478  *  @{
479  */
480 static inline uint16_t
481 brw_inst_3src_a1_src0_imm(ASSERTED const struct intel_device_info *devinfo,
482                           const brw_inst *insn)
483 {
484    assert(devinfo->ver >= 10);
485    if (devinfo->ver >= 12)
486       return brw_inst_bits(insn, 79, 64);
487    else
488       return brw_inst_bits(insn, 82, 67);
489 }
490 
491 static inline uint16_t
brw_inst_3src_a1_src2_imm(ASSERTED const struct intel_device_info * devinfo,const brw_inst * insn)492 brw_inst_3src_a1_src2_imm(ASSERTED const struct intel_device_info *devinfo,
493                           const brw_inst *insn)
494 {
495    assert(devinfo->ver >= 10);
496    if (devinfo->ver >= 12)
497       return brw_inst_bits(insn, 127, 112);
498    else
499       return brw_inst_bits(insn, 124, 109);
500 }
501 
502 static inline void
brw_inst_set_3src_a1_src0_imm(ASSERTED const struct intel_device_info * devinfo,brw_inst * insn,uint16_t value)503 brw_inst_set_3src_a1_src0_imm(ASSERTED const struct intel_device_info *devinfo,
504                               brw_inst *insn, uint16_t value)
505 {
506    assert(devinfo->ver >= 10);
507    if (devinfo->ver >= 12)
508       brw_inst_set_bits(insn, 79, 64, value);
509    else
510       brw_inst_set_bits(insn, 82, 67, value);
511 }
512 
513 static inline void
brw_inst_set_3src_a1_src2_imm(ASSERTED const struct intel_device_info * devinfo,brw_inst * insn,uint16_t value)514 brw_inst_set_3src_a1_src2_imm(ASSERTED const struct intel_device_info *devinfo,
515                               brw_inst *insn, uint16_t value)
516 {
517    assert(devinfo->ver >= 10);
518    if (devinfo->ver >= 12)
519       brw_inst_set_bits(insn, 127, 112, value);
520    else
521       brw_inst_set_bits(insn, 124, 109, value);
522 }
523 /** @} */
524 
525 /**
526  * Flow control instruction bits:
527  *  @{
528  */
529 static inline void
brw_inst_set_uip(const struct intel_device_info * devinfo,brw_inst * inst,int32_t value)530 brw_inst_set_uip(const struct intel_device_info *devinfo,
531                  brw_inst *inst, int32_t value)
532 {
533    assert(devinfo->ver >= 6);
534 
535    if (devinfo->ver >= 12)
536       brw_inst_set_src1_is_imm(devinfo, inst, 1);
537 
538    if (devinfo->ver >= 8) {
539       brw_inst_set_bits(inst, 95, 64, (uint32_t)value);
540    } else {
541       assert(value <= (1 << 16) - 1);
542       assert(value > -(1 << 16));
543       brw_inst_set_bits(inst, 127, 112, (uint16_t)value);
544    }
545 }
546 
547 static inline int32_t
brw_inst_uip(const struct intel_device_info * devinfo,const brw_inst * inst)548 brw_inst_uip(const struct intel_device_info *devinfo, const brw_inst *inst)
549 {
550    assert(devinfo->ver >= 6);
551 
552    if (devinfo->ver >= 8) {
553       return brw_inst_bits(inst, 95, 64);
554    } else {
555       return (int16_t)brw_inst_bits(inst, 127, 112);
556    }
557 }
558 
559 static inline void
brw_inst_set_jip(const struct intel_device_info * devinfo,brw_inst * inst,int32_t value)560 brw_inst_set_jip(const struct intel_device_info *devinfo,
561                  brw_inst *inst, int32_t value)
562 {
563    assert(devinfo->ver >= 6);
564 
565    if (devinfo->ver >= 12)
566       brw_inst_set_src0_is_imm(devinfo, inst, 1);
567 
568    if (devinfo->ver >= 8) {
569       brw_inst_set_bits(inst, 127, 96, (uint32_t)value);
570    } else {
571       assert(value <= (1 << 15) - 1);
572       assert(value >= -(1 << 15));
573       brw_inst_set_bits(inst, 111, 96, (uint16_t)value);
574    }
575 }
576 
577 static inline int32_t
brw_inst_jip(const struct intel_device_info * devinfo,const brw_inst * inst)578 brw_inst_jip(const struct intel_device_info *devinfo, const brw_inst *inst)
579 {
580    assert(devinfo->ver >= 6);
581 
582    if (devinfo->ver >= 8) {
583       return brw_inst_bits(inst, 127, 96);
584    } else {
585       return (int16_t)brw_inst_bits(inst, 111, 96);
586    }
587 }
588 
589 /** Like FC, but using int16_t to handle negative jump targets. */
590 #define FJ(name, high, low, assertions)                                       \
591 static inline void                                                            \
592 brw_inst_set_##name(const struct intel_device_info *devinfo, brw_inst *inst, int16_t v) \
593 {                                                                             \
594    assert(assertions);                                                        \
595    (void) devinfo;                                                            \
596    brw_inst_set_bits(inst, high, low, (uint16_t) v);                          \
597 }                                                                             \
598 static inline int16_t                                                         \
599 brw_inst_##name(const struct intel_device_info *devinfo, const brw_inst *inst)\
600 {                                                                             \
601    assert(assertions);                                                        \
602    (void) devinfo;                                                            \
603    return brw_inst_bits(inst, high, low);                                     \
604 }
605 
606 FJ(gfx6_jump_count,  63,  48, devinfo->ver == 6)
607 FJ(gfx4_jump_count, 111,  96, devinfo->ver < 6)
608 FC(gfx4_pop_count,  /* 4+ */ 115, 112, /* 12+ */ -1, -1, devinfo->ver < 6)
609 /** @} */
610 
611 /**
612  * SEND instructions:
613  *  @{
614  */
615 FC(send_ex_desc_ia_subreg_nr, /* 4+ */ 82, 80, /* 12+ */  42,  40, devinfo->ver >= 9)
616 FC(send_src0_address_mode,    /* 4+ */ 79, 79, /* 12+ */  -1,  -1, devinfo->ver >= 9)
617 FC(send_sel_reg32_desc,       /* 4+ */ 77, 77, /* 12+ */  48,  48, devinfo->ver >= 9)
618 FC(send_sel_reg32_ex_desc,    /* 4+ */ 61, 61, /* 12+ */  49,  49, devinfo->ver >= 9)
619 F8(send_src0_reg_file,        /* 4+ */ 38, 37, /* 8+ */   42,  41, /* 12+ */ 66, 66)
620 FC(send_src1_reg_nr,          /* 4+ */ 51, 44, /* 12+ */ 111, 104, devinfo->ver >= 9)
621 FC(send_src1_reg_file,        /* 4+ */ 36, 36, /* 12+ */  98,  98, devinfo->ver >= 9)
622 FC(send_dst_reg_file,         /* 4+ */ 35, 35, /* 12+ */  50,  50, devinfo->ver >= 9)
623 /** @} */
624 
625 /* Message descriptor bits */
626 #define MD(x) ((x) + 96)
627 #define MD12(x) ((x) >= 30 ? (x) - 30 + 122 :        \
628                  (x) >= 25 ? (x) - 25 + 67 :         \
629                  (x) >= 20 ? (x) - 20 + 51 :         \
630                  (x) >= 11 ? (x) - 11 + 113 :        \
631                  (x) - 0 + 81)
632 
633 /**
634  * Set the SEND(C) message descriptor immediate.
635  *
636  * This doesn't include the SFID nor the EOT field that were considered to be
637  * part of the message descriptor by ancient versions of the BSpec, because
638  * they are present in the instruction even if the message descriptor is
639  * provided indirectly in the address register, so we want to specify them
640  * separately.
641  */
642 static inline void
brw_inst_set_send_desc(const struct intel_device_info * devinfo,brw_inst * inst,uint32_t value)643 brw_inst_set_send_desc(const struct intel_device_info *devinfo,
644                        brw_inst *inst, uint32_t value)
645 {
646    if (devinfo->ver >= 12) {
647       brw_inst_set_bits(inst, 123, 122, GET_BITS(value, 31, 30));
648       brw_inst_set_bits(inst, 71, 67, GET_BITS(value, 29, 25));
649       brw_inst_set_bits(inst, 55, 51, GET_BITS(value, 24, 20));
650       brw_inst_set_bits(inst, 121, 113, GET_BITS(value, 19, 11));
651       brw_inst_set_bits(inst, 91, 81, GET_BITS(value, 10, 0));
652    } else if (devinfo->ver >= 9) {
653       brw_inst_set_bits(inst, 126, 96, value);
654       assert(value >> 31 == 0);
655    } else if (devinfo->ver >= 5) {
656       brw_inst_set_bits(inst, 124, 96, value);
657       assert(value >> 29 == 0);
658    } else {
659       brw_inst_set_bits(inst, 119, 96, value);
660       assert(value >> 24 == 0);
661    }
662 }
663 
664 /**
665  * Get the SEND(C) message descriptor immediate.
666  *
667  * \sa brw_inst_set_send_desc().
668  */
669 static inline uint32_t
brw_inst_send_desc(const struct intel_device_info * devinfo,const brw_inst * inst)670 brw_inst_send_desc(const struct intel_device_info *devinfo,
671                    const brw_inst *inst)
672 {
673    if (devinfo->ver >= 12) {
674       return (brw_inst_bits(inst, 123, 122) << 30 |
675               brw_inst_bits(inst, 71, 67) << 25 |
676               brw_inst_bits(inst, 55, 51) << 20 |
677               brw_inst_bits(inst, 121, 113) << 11 |
678               brw_inst_bits(inst, 91, 81));
679    } else if (devinfo->ver >= 9) {
680       return brw_inst_bits(inst, 126, 96);
681    } else if (devinfo->ver >= 5) {
682       return brw_inst_bits(inst, 124, 96);
683    } else {
684       return brw_inst_bits(inst, 119, 96);
685    }
686 }
687 
688 /**
689  * Set the SEND(C) message extended descriptor immediate.
690  *
691  * This doesn't include the SFID nor the EOT field that were considered to be
692  * part of the extended message descriptor by some versions of the BSpec,
693  * because they are present in the instruction even if the extended message
694  * descriptor is provided indirectly in a register, so we want to specify them
695  * separately.
696  */
697 static inline void
brw_inst_set_send_ex_desc(const struct intel_device_info * devinfo,brw_inst * inst,uint32_t value)698 brw_inst_set_send_ex_desc(const struct intel_device_info *devinfo,
699                           brw_inst *inst, uint32_t value)
700 {
701    if (devinfo->ver >= 12) {
702       brw_inst_set_bits(inst, 127, 124, GET_BITS(value, 31, 28));
703       brw_inst_set_bits(inst, 97, 96, GET_BITS(value, 27, 26));
704       brw_inst_set_bits(inst, 65, 64, GET_BITS(value, 25, 24));
705       brw_inst_set_bits(inst, 47, 35, GET_BITS(value, 23, 11));
706       brw_inst_set_bits(inst, 103, 99, GET_BITS(value, 10, 6));
707       assert(GET_BITS(value, 5, 0) == 0);
708    } else {
709       assert(devinfo->ver >= 9);
710       brw_inst_set_bits(inst, 94, 91, GET_BITS(value, 31, 28));
711       brw_inst_set_bits(inst, 88, 85, GET_BITS(value, 27, 24));
712       brw_inst_set_bits(inst, 83, 80, GET_BITS(value, 23, 20));
713       brw_inst_set_bits(inst, 67, 64, GET_BITS(value, 19, 16));
714       assert(GET_BITS(value, 15, 0) == 0);
715    }
716 }
717 
718 /**
719  * Set the SENDS(C) message extended descriptor immediate.
720  *
721  * This doesn't include the SFID nor the EOT field that were considered to be
722  * part of the extended message descriptor by some versions of the BSpec,
723  * because they are present in the instruction even if the extended message
724  * descriptor is provided indirectly in a register, so we want to specify them
725  * separately.
726  */
727 static inline void
brw_inst_set_sends_ex_desc(const struct intel_device_info * devinfo,brw_inst * inst,uint32_t value)728 brw_inst_set_sends_ex_desc(const struct intel_device_info *devinfo,
729                            brw_inst *inst, uint32_t value)
730 {
731    if (devinfo->ver >= 12) {
732       brw_inst_set_send_ex_desc(devinfo, inst, value);
733    } else {
734       brw_inst_set_bits(inst, 95, 80, GET_BITS(value, 31, 16));
735       assert(GET_BITS(value, 15, 10) == 0);
736       brw_inst_set_bits(inst, 67, 64, GET_BITS(value, 9, 6));
737       assert(GET_BITS(value, 5, 0) == 0);
738    }
739 }
740 
741 /**
742  * Get the SEND(C) message extended descriptor immediate.
743  *
744  * \sa brw_inst_set_send_ex_desc().
745  */
746 static inline uint32_t
brw_inst_send_ex_desc(const struct intel_device_info * devinfo,const brw_inst * inst)747 brw_inst_send_ex_desc(const struct intel_device_info *devinfo,
748                       const brw_inst *inst)
749 {
750    if (devinfo->ver >= 12) {
751       return (brw_inst_bits(inst, 127, 124) << 28 |
752               brw_inst_bits(inst, 97, 96) << 26 |
753               brw_inst_bits(inst, 65, 64) << 24 |
754               brw_inst_bits(inst, 47, 35) << 11 |
755               brw_inst_bits(inst, 103, 99) << 6);
756    } else {
757       assert(devinfo->ver >= 9);
758       return (brw_inst_bits(inst, 94, 91) << 28 |
759               brw_inst_bits(inst, 88, 85) << 24 |
760               brw_inst_bits(inst, 83, 80) << 20 |
761               brw_inst_bits(inst, 67, 64) << 16);
762    }
763 }
764 
765 /**
766  * Get the SENDS(C) message extended descriptor immediate.
767  *
768  * \sa brw_inst_set_send_ex_desc().
769  */
770 static inline uint32_t
brw_inst_sends_ex_desc(const struct intel_device_info * devinfo,const brw_inst * inst)771 brw_inst_sends_ex_desc(const struct intel_device_info *devinfo,
772                        const brw_inst *inst)
773 {
774    if (devinfo->ver >= 12) {
775       return brw_inst_send_ex_desc(devinfo, inst);
776    } else {
777       return (brw_inst_bits(inst, 95, 80) << 16 |
778               brw_inst_bits(inst, 67, 64) << 6);
779    }
780 }
781 
782 /**
783  * Fields for SEND messages:
784  *  @{
785  */
786 F(eot,                 /* 4+ */ 127, 127, /* 12+ */ 34, 34)
787 FF(mlen,
788    /* 4:   */ 119, 116,
789    /* 4.5: */ 119, 116,
790    /* 5:   */ 124, 121,
791    /* 6:   */ 124, 121,
792    /* 7:   */ 124, 121,
793    /* 8:   */ 124, 121,
794    /* 12:  */ MD12(28), MD12(25));
795 FF(rlen,
796    /* 4:   */ 115, 112,
797    /* 4.5: */ 115, 112,
798    /* 5:   */ 120, 116,
799    /* 6:   */ 120, 116,
800    /* 7:   */ 120, 116,
801    /* 8:   */ 120, 116,
802    /* 12:  */ MD12(24), MD12(20));
803 FF(header_present,
804    /* 4: doesn't exist */ -1, -1, -1, -1,
805    /* 5:   */ 115, 115,
806    /* 6:   */ 115, 115,
807    /* 7:   */ 115, 115,
808    /* 8:   */ 115, 115,
809    /* 12:  */ MD12(19), MD12(19))
810 F(gateway_notify, /* 4+ */ MD(16), MD(15), /* 12+ */ -1, -1)
811 FD(function_control,
812    /* 4:   */ 111,  96,
813    /* 4.5: */ 111,  96,
814    /* 5:   */ 114,  96,
815    /* 6:   */ 114,  96,
816    /* 7:   */ 114,  96,
817    /* 8:   */ 114,  96,
818    /* 12:  */ MD12(18), MD12(11), MD12(10), MD12(0))
819 FF(gateway_subfuncid,
820    /* 4:   */ MD(1), MD(0),
821    /* 4.5: */ MD(1), MD(0),
822    /* 5:   */ MD(1), MD(0), /* 2:0, but bit 2 is reserved MBZ */
823    /* 6:   */ MD(2), MD(0),
824    /* 7:   */ MD(2), MD(0),
825    /* 8:   */ MD(2), MD(0),
826    /* 12:  */ MD12(2), MD12(0))
827 FF(sfid,
828    /* 4:   */ 123, 120, /* called msg_target */
829    /* 4.5  */ 123, 120,
830    /* 5:   */  95,  92,
831    /* 6:   */  27,  24,
832    /* 7:   */  27,  24,
833    /* 8:   */  27,  24,
834    /* 12:  */  95,  92)
835 FF(null_rt,
836    /* 4-7: */ -1, -1, -1, -1, -1, -1, -1, -1, -1, -1,
837    /* 8:   */ 80, 80,
838    /* 12:  */ 44, 44) /* actually only Gfx11+ */
839 FC(base_mrf,   /* 4+ */ 27,  24, /* 12+ */ -1, -1, devinfo->ver < 6);
840 FF(send_rta_index,
841    /* 4:   */  -1,  -1,
842    /* 4.5  */  -1,  -1,
843    /* 5:   */  -1,  -1,
844    /* 6:   */  -1,  -1,
845    /* 7:   */  -1,  -1,
846    /* 8:   */  -1,  -1,
847    /* 12:  */  38,  36)
848 /** @} */
849 
850 /**
851  * URB message function control bits:
852  *  @{
853  */
854 FF(urb_per_slot_offset,
855    /* 4-6: */ -1, -1, -1, -1, -1, -1, -1, -1,
856    /* 7:   */ MD(16), MD(16),
857    /* 8:   */ MD(17), MD(17),
858    /* 12:  */ MD12(17), MD12(17))
859 FC(urb_channel_mask_present, /* 4+ */ MD(15), MD(15), /* 12+ */ MD12(15), MD12(15), devinfo->ver >= 8)
860 FC(urb_complete, /* 4+ */ MD(15), MD(15), /* 12+ */ -1, -1, devinfo->ver < 8)
861 FC(urb_used,     /* 4+ */ MD(14), MD(14), /* 12+ */ -1, -1, devinfo->ver < 7)
862 FC(urb_allocate, /* 4+ */ MD(13), MD(13), /* 12+ */ -1, -1, devinfo->ver < 7)
863 FF(urb_swizzle_control,
864    /* 4:   */ MD(11), MD(10),
865    /* 4.5: */ MD(11), MD(10),
866    /* 5:   */ MD(11), MD(10),
867    /* 6:   */ MD(11), MD(10),
868    /* 7:   */ MD(14), MD(14),
869    /* 8:   */ MD(15), MD(15),
870    /* 12:  */ -1, -1)
871 FD(urb_global_offset,
872    /* 4:   */ MD( 9), MD(4),
873    /* 4.5: */ MD( 9), MD(4),
874    /* 5:   */ MD( 9), MD(4),
875    /* 6:   */ MD( 9), MD(4),
876    /* 7:   */ MD(13), MD(3),
877    /* 8:   */ MD(14), MD(4),
878    /* 12:  */ MD12(14), MD12(11), MD12(10), MD12(4))
879 FF(urb_opcode,
880    /* 4:   */ MD( 3), MD(0),
881    /* 4.5: */ MD( 3), MD(0),
882    /* 5:   */ MD( 3), MD(0),
883    /* 6:   */ MD( 3), MD(0),
884    /* 7:   */ MD( 2), MD(0),
885    /* 8:   */ MD( 3), MD(0),
886    /* 12:  */ MD12(3), MD12(0))
887 /** @} */
888 
889 /**
890  * Gfx4-5 math messages:
891  *  @{
892  */
893 FC(math_msg_data_type,  /* 4+ */ MD(7), MD(7), /* 12+ */ -1, -1, devinfo->ver < 6)
894 FC(math_msg_saturate,   /* 4+ */ MD(6), MD(6), /* 12+ */ -1, -1, devinfo->ver < 6)
895 FC(math_msg_precision,  /* 4+ */ MD(5), MD(5), /* 12+ */ -1, -1, devinfo->ver < 6)
896 FC(math_msg_signed_int, /* 4+ */ MD(4), MD(4), /* 12+ */ -1, -1, devinfo->ver < 6)
897 FC(math_msg_function,   /* 4+ */ MD(3), MD(0), /* 12+ */ -1, -1, devinfo->ver < 6)
898 /** @} */
899 
900 /**
901  * Sampler message function control bits:
902  *  @{
903  */
904 FF(sampler_simd_mode,
905    /* 4: doesn't exist */ -1, -1, -1, -1,
906    /* 5:   */ MD(17), MD(16),
907    /* 6:   */ MD(17), MD(16),
908    /* 7:   */ MD(18), MD(17),
909    /* 8:   */ MD(18), MD(17),
910    /* 12:  */ MD12(18), MD12(17))
911 FF(sampler_msg_type,
912    /* 4:   */ MD(15), MD(14),
913    /* 4.5: */ MD(15), MD(12),
914    /* 5:   */ MD(15), MD(12),
915    /* 6:   */ MD(15), MD(12),
916    /* 7:   */ MD(16), MD(12),
917    /* 8:   */ MD(16), MD(12),
918    /* 12:  */ MD12(16), MD12(12))
919 FC(sampler_return_format, /* 4+ */ MD(13), MD(12), /* 12+ */ -1, -1, devinfo->ver == 4 && !devinfo->is_g4x)
920 FD(sampler,
921    /* 4:   */ MD(11), MD(8),
922    /* 4.5: */ MD(11), MD(8),
923    /* 5:   */ MD(11), MD(8),
924    /* 6:   */ MD(11), MD(8),
925    /* 7:   */ MD(11), MD(8),
926    /* 8:   */ MD(11), MD(8),
927    /* 12:   */ MD12(11), MD12(11), MD12(10), MD12(8))
928 F(binding_table_index,    /* 4+ */ MD(7), MD(0),  /* 12+ */ MD12(7), MD12(0)) /* also used by other messages */
929 /** @} */
930 
931 /**
932  * Data port message function control bits:
933  *  @{
934  */
935 FC(dp_category,           /* 4+ */ MD(18), MD(18), /* 12+ */ MD12(18), MD12(18), devinfo->ver >= 7)
936 
937 /* Gfx4-5 store fields in different bits for read/write messages. */
938 FF(dp_read_msg_type,
939    /* 4:   */ MD(13), MD(12),
940    /* 4.5: */ MD(13), MD(11),
941    /* 5:   */ MD(13), MD(11),
942    /* 6:   */ MD(16), MD(13),
943    /* 7:   */ MD(17), MD(14),
944    /* 8:   */ MD(17), MD(14),
945    /* 12:  */ MD12(17), MD12(14))
946 FF(dp_write_msg_type,
947    /* 4:   */ MD(14), MD(12),
948    /* 4.5: */ MD(14), MD(12),
949    /* 5:   */ MD(14), MD(12),
950    /* 6:   */ MD(16), MD(13),
951    /* 7:   */ MD(17), MD(14),
952    /* 8:   */ MD(17), MD(14),
953    /* 12:  */ MD12(17), MD12(14))
954 FD(dp_read_msg_control,
955    /* 4:   */ MD(11), MD( 8),
956    /* 4.5: */ MD(10), MD( 8),
957    /* 5:   */ MD(10), MD( 8),
958    /* 6:   */ MD(12), MD( 8),
959    /* 7:   */ MD(13), MD( 8),
960    /* 8:   */ MD(13), MD( 8),
961    /* 12:  */ MD12(13), MD12(11), MD12(10), MD12(8))
962 FD(dp_write_msg_control,
963    /* 4:   */ MD(11), MD( 8),
964    /* 4.5: */ MD(11), MD( 8),
965    /* 5:   */ MD(11), MD( 8),
966    /* 6:   */ MD(12), MD( 8),
967    /* 7:   */ MD(13), MD( 8),
968    /* 8:   */ MD(13), MD( 8),
969    /* 12:  */ MD12(13), MD12(11), MD12(10), MD12(8))
970 FC(dp_read_target_cache, /* 4+ */ MD(15), MD(14), /* 12+ */ -1, -1, devinfo->ver < 6);
971 
972 FF(dp_write_commit,
973    /* 4:   */ MD(15),  MD(15),
974    /* 4.5: */ MD(15),  MD(15),
975    /* 5:   */ MD(15),  MD(15),
976    /* 6:   */ MD(17),  MD(17),
977    /* 7+: does not exist */ -1, -1, -1, -1,
978    /* 12:  */ -1, -1)
979 
980 /* Gfx6+ use the same bit locations for everything. */
981 FF(dp_msg_type,
982    /* 4-5: use dp_read_msg_type or dp_write_msg_type instead */
983    -1, -1, -1, -1, -1, -1,
984    /* 6:   */ MD(16), MD(13),
985    /* 7:   */ MD(17), MD(14),
986    /* 8:   */ MD(18), MD(14),
987    /* 12:  */ MD12(18), MD12(14))
988 FD(dp_msg_control,
989    /* 4:   */ MD(11), MD( 8),
990    /* 4.5-5: use dp_read_msg_control or dp_write_msg_control */ -1, -1, -1, -1,
991    /* 6:   */ MD(12), MD( 8),
992    /* 7:   */ MD(13), MD( 8),
993    /* 8:   */ MD(13), MD( 8),
994    /* 12:  */ MD12(13), MD12(11), MD12(10), MD12(8))
995 /** @} */
996 
997 /**
998  * Scratch message bits (Gfx7+):
999  *  @{
1000  */
1001 FC(scratch_read_write, /* 4+ */ MD(17), MD(17), /* 12+ */ MD12(17), MD12(17), devinfo->ver >= 7) /* 0 = read,  1 = write */
1002 FC(scratch_type,       /* 4+ */ MD(16), MD(16), /* 12+ */ -1, -1, devinfo->ver >= 7) /* 0 = OWord, 1 = DWord */
1003 FC(scratch_invalidate_after_read, /* 4+ */ MD(15), MD(15), /* 12+ */ MD12(15), MD12(15), devinfo->ver >= 7)
1004 FC(scratch_block_size, /* 4+ */ MD(13), MD(12), /* 12+ */ MD12(13), MD12(12), devinfo->ver >= 7)
1005 FD(scratch_addr_offset,
1006    /* 4:   */ -1, -1,
1007    /* 4.5: */ -1, -1,
1008    /* 5:   */ -1, -1,
1009    /* 6:   */ -1, -1,
1010    /* 7:   */ MD(11), MD(0),
1011    /* 8:   */ MD(11), MD(0),
1012    /* 12:  */ MD12(11), MD12(11), MD12(10), MD12(0))
1013 /** @} */
1014 
1015 /**
1016  * Render Target message function control bits:
1017  *  @{
1018  */
1019 FF(rt_last,
1020    /* 4:   */ MD(11), MD(11),
1021    /* 4.5: */ MD(11), MD(11),
1022    /* 5:   */ MD(11), MD(11),
1023    /* 6:   */ MD(12), MD(12),
1024    /* 7:   */ MD(12), MD(12),
1025    /* 8:   */ MD(12), MD(12),
1026    /* 12:  */ MD12(12), MD12(12))
1027 FC(rt_slot_group,      /* 4+ */ MD(11),  MD(11), /* 12+ */ MD12(11), MD12(11), devinfo->ver >= 6)
1028 F(rt_message_type,     /* 4+ */ MD(10),  MD( 8), /* 12+ */ MD12(10), MD12(8))
1029 /** @} */
1030 
1031 /**
1032  * Thread Spawn message function control bits:
1033  *  @{
1034  */
1035 FC(ts_resource_select,  /* 4+ */ MD( 4),  MD( 4), /* 12+ */ -1, -1, devinfo->ver < 11)
1036 FC(ts_request_type,     /* 4+ */ MD( 1),  MD( 1), /* 12+ */ -1, -1, devinfo->ver < 11)
1037 F(ts_opcode,           /* 4+ */ MD( 0),  MD( 0), /* 12+ */ MD12(0), MD12(0))
1038 /** @} */
1039 
1040 /**
1041  * Pixel Interpolator message function control bits:
1042  *  @{
1043  */
1044 F(pi_simd_mode,        /* 4+ */ MD(16),  MD(16), /* 12+ */ MD12(16), MD12(16))
1045 F(pi_nopersp,          /* 4+ */ MD(14),  MD(14), /* 12+ */ MD12(14), MD12(14))
1046 F(pi_message_type,     /* 4+ */ MD(13),  MD(12), /* 12+ */ MD12(13), MD12(12))
1047 F(pi_slot_group,       /* 4+ */ MD(11),  MD(11), /* 12+ */ MD12(11), MD12(11))
1048 F(pi_message_data,     /* 4+ */ MD(7),   MD(0),  /* 12+ */  MD12(7), MD12(0))
1049 /** @} */
1050 
1051 /**
1052  * Immediates:
1053  *  @{
1054  */
1055 static inline int
brw_inst_imm_d(const struct intel_device_info * devinfo,const brw_inst * insn)1056 brw_inst_imm_d(const struct intel_device_info *devinfo, const brw_inst *insn)
1057 {
1058    (void) devinfo;
1059    return brw_inst_bits(insn, 127, 96);
1060 }
1061 
1062 static inline unsigned
brw_inst_imm_ud(const struct intel_device_info * devinfo,const brw_inst * insn)1063 brw_inst_imm_ud(const struct intel_device_info *devinfo, const brw_inst *insn)
1064 {
1065    (void) devinfo;
1066    return brw_inst_bits(insn, 127, 96);
1067 }
1068 
1069 static inline uint64_t
brw_inst_imm_uq(ASSERTED const struct intel_device_info * devinfo,const brw_inst * insn)1070 brw_inst_imm_uq(ASSERTED const struct intel_device_info *devinfo,
1071                 const brw_inst *insn)
1072 {
1073    assert(devinfo->ver >= 8);
1074    return brw_inst_bits(insn, 127, 64);
1075 }
1076 
1077 static inline float
brw_inst_imm_f(const struct intel_device_info * devinfo,const brw_inst * insn)1078 brw_inst_imm_f(const struct intel_device_info *devinfo, const brw_inst *insn)
1079 {
1080    union {
1081       float f;
1082       uint32_t u;
1083    } ft;
1084    (void) devinfo;
1085    ft.u = brw_inst_bits(insn, 127, 96);
1086    return ft.f;
1087 }
1088 
1089 static inline double
brw_inst_imm_df(const struct intel_device_info * devinfo,const brw_inst * insn)1090 brw_inst_imm_df(const struct intel_device_info *devinfo, const brw_inst *insn)
1091 {
1092    union {
1093       double d;
1094       uint64_t u;
1095    } dt;
1096    (void) devinfo;
1097    dt.u = brw_inst_bits(insn, 127, 64);
1098    return dt.d;
1099 }
1100 
1101 static inline void
brw_inst_set_imm_d(const struct intel_device_info * devinfo,brw_inst * insn,int value)1102 brw_inst_set_imm_d(const struct intel_device_info *devinfo,
1103                    brw_inst *insn, int value)
1104 {
1105    (void) devinfo;
1106    return brw_inst_set_bits(insn, 127, 96, value);
1107 }
1108 
1109 static inline void
brw_inst_set_imm_ud(const struct intel_device_info * devinfo,brw_inst * insn,unsigned value)1110 brw_inst_set_imm_ud(const struct intel_device_info *devinfo,
1111                     brw_inst *insn, unsigned value)
1112 {
1113    (void) devinfo;
1114    return brw_inst_set_bits(insn, 127, 96, value);
1115 }
1116 
1117 static inline void
brw_inst_set_imm_f(const struct intel_device_info * devinfo,brw_inst * insn,float value)1118 brw_inst_set_imm_f(const struct intel_device_info *devinfo,
1119                    brw_inst *insn, float value)
1120 {
1121    union {
1122       float f;
1123       uint32_t u;
1124    } ft;
1125    (void) devinfo;
1126    ft.f = value;
1127    brw_inst_set_bits(insn, 127, 96, ft.u);
1128 }
1129 
1130 static inline void
brw_inst_set_imm_df(const struct intel_device_info * devinfo,brw_inst * insn,double value)1131 brw_inst_set_imm_df(const struct intel_device_info *devinfo,
1132                     brw_inst *insn, double value)
1133 {
1134    union {
1135       double d;
1136       uint64_t u;
1137    } dt;
1138    (void) devinfo;
1139    dt.d = value;
1140 
1141    if (devinfo->ver >= 12) {
1142       brw_inst_set_bits(insn, 95, 64, dt.u >> 32);
1143       brw_inst_set_bits(insn, 127, 96, dt.u & 0xFFFFFFFF);
1144    } else {
1145       brw_inst_set_bits(insn, 127, 64, dt.u);
1146    }
1147 }
1148 
1149 static inline void
brw_inst_set_imm_uq(const struct intel_device_info * devinfo,brw_inst * insn,uint64_t value)1150 brw_inst_set_imm_uq(const struct intel_device_info *devinfo,
1151                     brw_inst *insn, uint64_t value)
1152 {
1153    (void) devinfo;
1154    if (devinfo->ver >= 12) {
1155       brw_inst_set_bits(insn, 95, 64, value >> 32);
1156       brw_inst_set_bits(insn, 127, 96, value & 0xFFFFFFFF);
1157    } else {
1158       brw_inst_set_bits(insn, 127, 64, value);
1159    }
1160 }
1161 
1162 /** @} */
1163 
1164 #define REG_TYPE(reg)                                                         \
1165 static inline void                                                            \
1166 brw_inst_set_##reg##_file_type(const struct intel_device_info *devinfo,       \
1167                                brw_inst *inst, enum brw_reg_file file,        \
1168                                enum brw_reg_type type)                        \
1169 {                                                                             \
1170    assert(file <= BRW_IMMEDIATE_VALUE);                                       \
1171    unsigned hw_type = brw_reg_type_to_hw_type(devinfo, file, type);           \
1172    brw_inst_set_##reg##_reg_file(devinfo, inst, file);                        \
1173    brw_inst_set_##reg##_reg_hw_type(devinfo, inst, hw_type);                  \
1174 }                                                                             \
1175                                                                               \
1176 static inline enum brw_reg_type                                               \
1177 brw_inst_##reg##_type(const struct intel_device_info *devinfo,                \
1178                       const brw_inst *inst)                                   \
1179 {                                                                             \
1180    unsigned file = __builtin_strcmp("dst", #reg) == 0 ?                       \
1181                    (unsigned) BRW_GENERAL_REGISTER_FILE :                     \
1182                    brw_inst_##reg##_reg_file(devinfo, inst);                  \
1183    unsigned hw_type = brw_inst_##reg##_reg_hw_type(devinfo, inst);            \
1184    return brw_hw_type_to_reg_type(devinfo, (enum brw_reg_file)file, hw_type); \
1185 }
1186 
1187 REG_TYPE(dst)
REG_TYPE(src0)1188 REG_TYPE(src0)
1189 REG_TYPE(src1)
1190 #undef REG_TYPE
1191 
1192 
1193 /* The AddrImm fields are split into two discontiguous sections on Gfx8+ */
1194 #define BRW_IA1_ADDR_IMM(reg, g4_high, g4_low, g8_nine, g8_high, g8_low, \
1195                          g12_high, g12_low)                              \
1196 static inline void                                                       \
1197 brw_inst_set_##reg##_ia1_addr_imm(const struct                           \
1198                                   intel_device_info *devinfo,            \
1199                                   brw_inst *inst,                        \
1200                                   unsigned value)                        \
1201 {                                                                        \
1202    assert((value & ~0x3ff) == 0);                                        \
1203    if (devinfo->ver >= 12) {                                             \
1204       brw_inst_set_bits(inst, g12_high, g12_low, value);                 \
1205    } else if (devinfo->ver >= 8) {                                       \
1206       brw_inst_set_bits(inst, g8_high, g8_low, value & 0x1ff);           \
1207       brw_inst_set_bits(inst, g8_nine, g8_nine, value >> 9);             \
1208    } else {                                                              \
1209       brw_inst_set_bits(inst, g4_high, g4_low, value);                   \
1210    }                                                                     \
1211 }                                                                        \
1212 static inline unsigned                                                   \
1213 brw_inst_##reg##_ia1_addr_imm(const struct intel_device_info *devinfo,   \
1214                               const brw_inst *inst)                      \
1215 {                                                                        \
1216    if (devinfo->ver >= 12) {                                             \
1217       return brw_inst_bits(inst, g12_high, g12_low);                     \
1218    } else if (devinfo->ver >= 8) {                                       \
1219       return brw_inst_bits(inst, g8_high, g8_low) |                      \
1220              (brw_inst_bits(inst, g8_nine, g8_nine) << 9);               \
1221    } else {                                                              \
1222       return brw_inst_bits(inst, g4_high, g4_low);                       \
1223    }                                                                     \
1224 }
1225 
1226 /* AddrImm[9:0] for Align1 Indirect Addressing        */
1227 /*                     -Gen 4-  ----Gfx8----  -Gfx12- */
1228 BRW_IA1_ADDR_IMM(src1, 105, 96, 121, 104, 96, 107, 98)
1229 BRW_IA1_ADDR_IMM(src0,  73, 64,  95,  72, 64,  75, 66)
1230 BRW_IA1_ADDR_IMM(dst,   57, 48,  47,  56, 48,  59, 50)
1231 
1232 #define BRW_IA16_ADDR_IMM(reg, g4_high, g4_low, g8_nine, g8_high, g8_low) \
1233 static inline void                                                        \
1234 brw_inst_set_##reg##_ia16_addr_imm(const struct                           \
1235                                    intel_device_info *devinfo,            \
1236                                    brw_inst *inst, unsigned value)        \
1237 {                                                                         \
1238    assert(devinfo->ver < 12);                                             \
1239    assert((value & ~0x3ff) == 0);                                         \
1240    if (devinfo->ver >= 8) {                                               \
1241       assert(GET_BITS(value, 3, 0) == 0);                                 \
1242       brw_inst_set_bits(inst, g8_high, g8_low, GET_BITS(value, 8, 4));    \
1243       brw_inst_set_bits(inst, g8_nine, g8_nine, GET_BITS(value, 9, 9));   \
1244    } else {                                                               \
1245       brw_inst_set_bits(inst, g4_high, g4_low, value);                    \
1246    }                                                                      \
1247 }                                                                         \
1248 static inline unsigned                                                    \
1249 brw_inst_##reg##_ia16_addr_imm(const struct intel_device_info *devinfo,   \
1250                                const brw_inst *inst)                      \
1251 {                                                                         \
1252    assert(devinfo->ver < 12);                                             \
1253    if (devinfo->ver >= 8) {                                               \
1254       return (brw_inst_bits(inst, g8_high, g8_low) << 4) |                \
1255              (brw_inst_bits(inst, g8_nine, g8_nine) << 9);                \
1256    } else {                                                               \
1257       return brw_inst_bits(inst, g4_high, g4_low);                        \
1258    }                                                                      \
1259 }
1260 
1261 /* AddrImm[9:0] for Align16 Indirect Addressing:
1262  * Compared to Align1, these are missing the low 4 bits.
1263  *                     -Gen 4-  ----Gfx8----
1264  */
1265 BRW_IA16_ADDR_IMM(src1,       105, 96, 121, 104, 100)
1266 BRW_IA16_ADDR_IMM(src0,        73, 64,  95,  72,  68)
1267 BRW_IA16_ADDR_IMM(dst,         57, 52,  47,  56,  52)
1268 BRW_IA16_ADDR_IMM(send_src0,   -1, -1,  78,  72,  68)
1269 BRW_IA16_ADDR_IMM(send_dst,    -1, -1,  62,  56,  52)
1270 
1271 /**
1272  * Fetch a set of contiguous bits from the instruction.
1273  *
1274  * Bits indices range from 0..127; fields may not cross 64-bit boundaries.
1275  */
1276 static inline uint64_t
1277 brw_inst_bits(const brw_inst *inst, unsigned high, unsigned low)
1278 {
1279    assume(high < 128);
1280    assume(high >= low);
1281    /* We assume the field doesn't cross 64-bit boundaries. */
1282    const unsigned word = high / 64;
1283    assert(word == low / 64);
1284 
1285    high %= 64;
1286    low %= 64;
1287 
1288    const uint64_t mask = (~0ull >> (64 - (high - low + 1)));
1289 
1290    return (inst->data[word] >> low) & mask;
1291 }
1292 
1293 /**
1294  * Set bits in the instruction, with proper shifting and masking.
1295  *
1296  * Bits indices range from 0..127; fields may not cross 64-bit boundaries.
1297  */
1298 static inline void
brw_inst_set_bits(brw_inst * inst,unsigned high,unsigned low,uint64_t value)1299 brw_inst_set_bits(brw_inst *inst, unsigned high, unsigned low, uint64_t value)
1300 {
1301    assume(high < 128);
1302    assume(high >= low);
1303    const unsigned word = high / 64;
1304    assert(word == low / 64);
1305 
1306    high %= 64;
1307    low %= 64;
1308 
1309    const uint64_t mask = (~0ull >> (64 - (high - low + 1))) << low;
1310 
1311    /* Make sure the supplied value actually fits in the given bitfield. */
1312    assert((value & (mask >> low)) == value);
1313 
1314    inst->data[word] = (inst->data[word] & ~mask) | (value << low);
1315 }
1316 
1317 #undef BRW_IA16_ADDR_IMM
1318 #undef BRW_IA1_ADDR_IMM
1319 #undef MD
1320 #undef F8
1321 #undef FF
1322 #undef BOUNDS
1323 #undef F
1324 #undef FC
1325 
1326 typedef struct {
1327    uint64_t data;
1328 } brw_compact_inst;
1329 
1330 /**
1331  * Fetch a set of contiguous bits from the compacted instruction.
1332  *
1333  * Bits indices range from 0..63.
1334  */
1335 static inline unsigned
brw_compact_inst_bits(const brw_compact_inst * inst,unsigned high,unsigned low)1336 brw_compact_inst_bits(const brw_compact_inst *inst, unsigned high, unsigned low)
1337 {
1338    const uint64_t mask = (1ull << (high - low + 1)) - 1;
1339 
1340    return (inst->data >> low) & mask;
1341 }
1342 
1343 /**
1344  * Set bits in the compacted instruction.
1345  *
1346  * Bits indices range from 0..63.
1347  */
1348 static inline void
brw_compact_inst_set_bits(brw_compact_inst * inst,unsigned high,unsigned low,uint64_t value)1349 brw_compact_inst_set_bits(brw_compact_inst *inst, unsigned high, unsigned low,
1350                           uint64_t value)
1351 {
1352    const uint64_t mask = ((1ull << (high - low + 1)) - 1) << low;
1353 
1354    /* Make sure the supplied value actually fits in the given bitfield. */
1355    assert((value & (mask >> low)) == value);
1356 
1357    inst->data = (inst->data & ~mask) | (value << low);
1358 }
1359 
1360 #define FC(name, high, low, gfx12_high, gfx12_low, assertions)     \
1361 static inline void                                                 \
1362 brw_compact_inst_set_##name(const struct                           \
1363                             intel_device_info *devinfo,            \
1364                             brw_compact_inst *inst, unsigned v)    \
1365 {                                                                  \
1366    assert(assertions);                                             \
1367    if (devinfo->ver >= 12)                                         \
1368       brw_compact_inst_set_bits(inst, gfx12_high, gfx12_low, v);   \
1369    else                                                            \
1370       brw_compact_inst_set_bits(inst, high, low, v);               \
1371 }                                                                  \
1372 static inline unsigned                                             \
1373 brw_compact_inst_##name(const struct intel_device_info *devinfo,   \
1374                         const brw_compact_inst *inst)              \
1375 {                                                                  \
1376    assert(assertions);                                             \
1377    if (devinfo->ver >= 12)                                         \
1378       return brw_compact_inst_bits(inst, gfx12_high, gfx12_low);   \
1379    else                                                            \
1380       return brw_compact_inst_bits(inst, high, low);               \
1381 }
1382 
1383 /* A simple macro for fields which stay in the same place on all generations
1384  * except for Gfx12.
1385  */
1386 #define F(name, high, low, gfx12_high, gfx12_low)       \
1387    FC(name, high, low, gfx12_high, gfx12_low, true)
1388 
1389 F(src1_reg_nr,      /* 4+ */ 63, 56, /* 12+ */ 63, 56)
1390 F(src0_reg_nr,      /* 4+ */ 55, 48, /* 12+ */ 47, 40)
1391 F(dst_reg_nr,       /* 4+ */ 47, 40, /* 12+ */ 23, 16)
1392 F(src1_index,       /* 4+ */ 39, 35, /* 12+ */ 55, 52)
1393 F(src0_index,       /* 4+ */ 34, 30, /* 12+ */ 51, 48)
1394 F(cmpt_control,     /* 4+ */ 29, 29, /* 12+ */ 29, 29) /* Same location as brw_inst */
1395 FC(flag_subreg_nr,  /* 4+ */ 28, 28, /* 12+ */ -1, -1, devinfo->ver <= 6)
1396 F(cond_modifier,    /* 4+ */ 27, 24, /* 12+ */ -1, -1) /* Same location as brw_inst */
1397 FC(acc_wr_control,  /* 4+ */ 23, 23, /* 12+ */ -1, -1, devinfo->ver >= 6)
1398 FC(mask_control_ex, /* 4+ */ 23, 23, /* 12+ */ -1, -1, devinfo->is_g4x || devinfo->ver == 5)
1399 F(subreg_index,     /* 4+ */ 22, 18, /* 12+ */ 39, 35)
1400 F(datatype_index,   /* 4+ */ 17, 13, /* 12+ */ 34, 30)
1401 F(control_index,    /* 4+ */ 12,  8, /* 12+ */ 28, 24)
1402 FC(swsb,            /* 4+ */ -1, -1, /* 12+ */ 15,  8, devinfo->ver >= 12)
1403 F(debug_control,    /* 4+ */  7,  7, /* 12+ */  7,  7)
1404 F(hw_opcode,        /* 4+ */  6,  0, /* 12+ */  6,  0) /* Same location as brw_inst */
1405 
1406 static inline unsigned
brw_compact_inst_imm(const struct intel_device_info * devinfo,const brw_compact_inst * inst)1407 brw_compact_inst_imm(const struct intel_device_info *devinfo,
1408                      const brw_compact_inst *inst)
1409 {
1410    if (devinfo->ver >= 12) {
1411       return brw_compact_inst_bits(inst, 63, 52);
1412    } else {
1413       return (brw_compact_inst_bits(inst, 39, 35) << 8) |
1414              (brw_compact_inst_bits(inst, 63, 56));
1415    }
1416 }
1417 
1418 /**
1419  * (Gfx8+) Compacted three-source instructions:
1420  *  @{
1421  */
1422 FC(3src_src2_reg_nr,    /* 4+ */ 63, 57, /* 12+ */ 55, 48, devinfo->ver >= 8)
1423 FC(3src_src1_reg_nr,    /* 4+ */ 56, 50, /* 12+ */ 63, 56, devinfo->ver >= 8)
1424 FC(3src_src0_reg_nr,    /* 4+ */ 49, 43, /* 12+ */ 47, 40, devinfo->ver >= 8)
1425 FC(3src_src2_subreg_nr, /* 4+ */ 42, 40, /* 12+ */ -1, -1, devinfo->ver >= 8)
1426 FC(3src_src1_subreg_nr, /* 4+ */ 39, 37, /* 12+ */ -1, -1, devinfo->ver >= 8)
1427 FC(3src_src0_subreg_nr, /* 4+ */ 36, 34, /* 12+ */ -1, -1, devinfo->ver >= 8)
1428 FC(3src_src2_rep_ctrl,  /* 4+ */ 33, 33, /* 12+ */ -1, -1, devinfo->ver >= 8)
1429 FC(3src_src1_rep_ctrl,  /* 4+ */ 32, 32, /* 12+ */ -1, -1, devinfo->ver >= 8)
1430 FC(3src_saturate,       /* 4+ */ 31, 31, /* 12+ */ -1, -1, devinfo->ver >= 8)
1431 FC(3src_debug_control,  /* 4+ */ 30, 30, /* 12+ */  7,  7, devinfo->ver >= 8)
1432 FC(3src_cmpt_control,   /* 4+ */ 29, 29, /* 12+ */ 29, 29, devinfo->ver >= 8)
1433 FC(3src_src0_rep_ctrl,  /* 4+ */ 28, 28, /* 12+ */ -1, -1, devinfo->ver >= 8)
1434 /* Reserved */
1435 FC(3src_dst_reg_nr,     /* 4+ */ 18, 12, /* 12+ */ 23, 16, devinfo->ver >= 8)
1436 FC(3src_source_index,   /* 4+ */ 11, 10, /* 12+ */ 34, 30, devinfo->ver >= 8)
1437 FC(3src_subreg_index,   /* 4+ */ -1, -1, /* 12+ */ 39, 35, devinfo->ver >= 12)
1438 FC(3src_control_index,  /* 4+ */  9,  8, /* 12+ */ 28, 24, devinfo->ver >= 8)
1439 FC(3src_swsb,           /* 4+ */ -1, -1, /* 12+ */ 15,  8, devinfo->ver >= 8)
1440 /* Bit 7 is Reserved (for future Opcode expansion) */
1441 FC(3src_hw_opcode,      /* 4+ */  6,  0, /* 12+ */  6,  0, devinfo->ver >= 8)
1442 /** @} */
1443 
1444 #undef F
1445 
1446 #ifdef __cplusplus
1447 }
1448 #endif
1449 
1450 #endif
1451