xref: /openbmc/qemu/target/hexagon/mmvec/macros.h (revision a6caeee8)
1 /*
2  *  Copyright(c) 2019-2022 Qualcomm Innovation Center, Inc. All Rights Reserved.
3  *
4  *  This program is free software; you can redistribute it and/or modify
5  *  it under the terms of the GNU General Public License as published by
6  *  the Free Software Foundation; either version 2 of the License, or
7  *  (at your option) any later version.
8  *
9  *  This program is distributed in the hope that it will be useful,
10  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
11  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  *  GNU General Public License for more details.
13  *
14  *  You should have received a copy of the GNU General Public License
15  *  along with this program; if not, see <http://www.gnu.org/licenses/>.
16  */
17 
18 #ifndef HEXAGON_MMVEC_MACROS_H
19 #define HEXAGON_MMVEC_MACROS_H
20 
21 #include "qemu/osdep.h"
22 #include "qemu/host-utils.h"
23 #include "arch.h"
24 #include "mmvec/system_ext_mmvec.h"
25 
26 #ifndef QEMU_GENERATE
27 #define VdV      (*(MMVector *)(VdV_void))
28 #define VsV      (*(MMVector *)(VsV_void))
29 #define VuV      (*(MMVector *)(VuV_void))
30 #define VvV      (*(MMVector *)(VvV_void))
31 #define VwV      (*(MMVector *)(VwV_void))
32 #define VxV      (*(MMVector *)(VxV_void))
33 #define VyV      (*(MMVector *)(VyV_void))
34 
35 #define VddV     (*(MMVectorPair *)(VddV_void))
36 #define VuuV     (*(MMVectorPair *)(VuuV_void))
37 #define VvvV     (*(MMVectorPair *)(VvvV_void))
38 #define VxxV     (*(MMVectorPair *)(VxxV_void))
39 
40 #define QeV      (*(MMQReg *)(QeV_void))
41 #define QdV      (*(MMQReg *)(QdV_void))
42 #define QsV      (*(MMQReg *)(QsV_void))
43 #define QtV      (*(MMQReg *)(QtV_void))
44 #define QuV      (*(MMQReg *)(QuV_void))
45 #define QvV      (*(MMQReg *)(QvV_void))
46 #define QxV      (*(MMQReg *)(QxV_void))
47 #endif
48 
49 #define LOG_VTCM_BYTE(VA, MASK, VAL, IDX) \
50     do { \
51         env->vtcm_log.data.ub[IDX] = (VAL); \
52         if (MASK) { \
53             set_bit((IDX), env->vtcm_log.mask); \
54         } else { \
55             clear_bit((IDX), env->vtcm_log.mask); \
56         } \
57         env->vtcm_log.va[IDX] = (VA); \
58     } while (0)
59 
60 #define fNOTQ(VAL) \
61     ({ \
62         MMQReg _ret;  \
63         int _i_;  \
64         for (_i_ = 0; _i_ < fVECSIZE() / 64; _i_++) { \
65             _ret.ud[_i_] = ~VAL.ud[_i_]; \
66         } \
67         _ret;\
68      })
69 #define fGETQBITS(REG, WIDTH, MASK, BITNO) \
70     ((MASK) & (REG.w[(BITNO) >> 5] >> ((BITNO) & 0x1f)))
71 #define fGETQBIT(REG, BITNO) fGETQBITS(REG, 1, 1, BITNO)
72 #define fGENMASKW(QREG, IDX) \
73     (((fGETQBIT(QREG, (IDX * 4 + 0)) ? 0xFF : 0x0) << 0)  | \
74      ((fGETQBIT(QREG, (IDX * 4 + 1)) ? 0xFF : 0x0) << 8)  | \
75      ((fGETQBIT(QREG, (IDX * 4 + 2)) ? 0xFF : 0x0) << 16) | \
76      ((fGETQBIT(QREG, (IDX * 4 + 3)) ? 0xFF : 0x0) << 24))
77 #define fGETNIBBLE(IDX, SRC) (fSXTN(4, 8, (SRC >> (4 * IDX)) & 0xF))
78 #define fGETCRUMB(IDX, SRC) (fSXTN(2, 8, (SRC >> (2 * IDX)) & 0x3))
79 #define fGETCRUMB_SYMMETRIC(IDX, SRC) \
80     ((fGETCRUMB(IDX, SRC) >= 0 ? (2 - fGETCRUMB(IDX, SRC)) \
81                                : fGETCRUMB(IDX, SRC)))
82 #define fGENMASKH(QREG, IDX) \
83     (((fGETQBIT(QREG, (IDX * 2 + 0)) ? 0xFF : 0x0) << 0) | \
84      ((fGETQBIT(QREG, (IDX * 2 + 1)) ? 0xFF : 0x0) << 8))
85 #define fGETMASKW(VREG, QREG, IDX) (VREG.w[IDX] & fGENMASKW((QREG), IDX))
86 #define fGETMASKH(VREG, QREG, IDX) (VREG.h[IDX] & fGENMASKH((QREG), IDX))
87 #define fCONDMASK8(QREG, IDX, YESVAL, NOVAL) \
88     (fGETQBIT(QREG, IDX) ? (YESVAL) : (NOVAL))
89 #define fCONDMASK16(QREG, IDX, YESVAL, NOVAL) \
90     ((fGENMASKH(QREG, IDX) & (YESVAL)) | \
91      (fGENMASKH(fNOTQ(QREG), IDX) & (NOVAL)))
92 #define fCONDMASK32(QREG, IDX, YESVAL, NOVAL) \
93     ((fGENMASKW(QREG, IDX) & (YESVAL)) | \
94      (fGENMASKW(fNOTQ(QREG), IDX) & (NOVAL)))
95 #define fSETQBITS(REG, WIDTH, MASK, BITNO, VAL) \
96     do { \
97         uint32_t __TMP = (VAL); \
98         REG.w[(BITNO) >> 5] &= ~((MASK) << ((BITNO) & 0x1f)); \
99         REG.w[(BITNO) >> 5] |= (((__TMP) & (MASK)) << ((BITNO) & 0x1f)); \
100     } while (0)
101 #define fSETQBIT(REG, BITNO, VAL) fSETQBITS(REG, 1, 1, BITNO, VAL)
102 #define fVBYTES() (fVECSIZE())
103 #define fVALIGN(ADDR, LOG2_ALIGNMENT) (ADDR = ADDR & ~(LOG2_ALIGNMENT - 1))
104 #define fVLASTBYTE(ADDR, LOG2_ALIGNMENT) (ADDR = ADDR | (LOG2_ALIGNMENT - 1))
105 #define fVELEM(WIDTH) ((fVECSIZE() * 8) / WIDTH)
106 #define fVECLOGSIZE() (7)
107 #define fVECSIZE() (1 << fVECLOGSIZE())
108 #define fSWAPB(A, B) do { uint8_t tmp = A; A = B; B = tmp; } while (0)
109 #define fV_AL_CHECK(EA, MASK) \
110     if ((EA) & (MASK)) { \
111         warn("aligning misaligned vector. EA=%08x", (EA)); \
112     }
113 #define fSCATTER_INIT(REGION_START, LENGTH, ELEMENT_SIZE) \
114     mem_vector_scatter_init(env)
115 #define fGATHER_INIT(REGION_START, LENGTH, ELEMENT_SIZE) \
116     mem_vector_gather_init(env)
117 #define fSCATTER_FINISH(OP)
118 #define fGATHER_FINISH()
119 #define fLOG_SCATTER_OP(SIZE) \
120     do { \
121         env->vtcm_log.op = true; \
122         env->vtcm_log.op_size = SIZE; \
123     } while (0)
124 #define fVLOG_VTCM_WORD_INCREMENT(EA, OFFSET, INC, IDX, ALIGNMENT, LEN) \
125     do { \
126         int log_byte = 0; \
127         target_ulong va = EA; \
128         target_ulong va_high = EA + LEN; \
129         for (int i0 = 0; i0 < 4; i0++) { \
130             log_byte = (va + i0) <= va_high; \
131             LOG_VTCM_BYTE(va + i0, log_byte, INC. ub[4 * IDX + i0], \
132                           4 * IDX + i0); \
133         } \
134     } while (0)
135 #define fVLOG_VTCM_HALFWORD_INCREMENT(EA, OFFSET, INC, IDX, ALIGNMENT, LEN) \
136     do { \
137         int log_byte = 0; \
138         target_ulong va = EA; \
139         target_ulong va_high = EA + LEN; \
140         for (int i0 = 0; i0 < 2; i0++) { \
141             log_byte = (va + i0) <= va_high; \
142             LOG_VTCM_BYTE(va + i0, log_byte, INC.ub[2 * IDX + i0], \
143                           2 * IDX + i0); \
144         } \
145     } while (0)
146 
147 #define fVLOG_VTCM_HALFWORD_INCREMENT_DV(EA, OFFSET, INC, IDX, IDX2, IDX_H, \
148                                          ALIGNMENT, LEN) \
149     do { \
150         int log_byte = 0; \
151         target_ulong va = EA; \
152         target_ulong va_high = EA + LEN; \
153         for (int i0 = 0; i0 < 2; i0++) { \
154             log_byte = (va + i0) <= va_high; \
155             LOG_VTCM_BYTE(va + i0, log_byte, INC.ub[2 * IDX + i0], \
156                           2 * IDX + i0); \
157         } \
158     } while (0)
159 
160 /* NOTE - Will this always be tmp_VRegs[0]; */
161 #define GATHER_FUNCTION(EA, OFFSET, IDX, LEN, ELEMENT_SIZE, BANK_IDX, QVAL) \
162     do { \
163         int i0; \
164         target_ulong va = EA; \
165         target_ulong va_high = EA + LEN; \
166         uintptr_t ra = GETPC(); \
167         int log_byte = 0; \
168         for (i0 = 0; i0 < ELEMENT_SIZE; i0++) { \
169             log_byte = ((va + i0) <= va_high) && QVAL; \
170             uint8_t B; \
171             B = cpu_ldub_data_ra(env, EA + i0, ra); \
172             env->tmp_VRegs[0].ub[ELEMENT_SIZE * IDX + i0] = B; \
173             LOG_VTCM_BYTE(va + i0, log_byte, B, ELEMENT_SIZE * IDX + i0); \
174         } \
175     } while (0)
176 #define fVLOG_VTCM_GATHER_WORD(EA, OFFSET, IDX, LEN) \
177     do { \
178         GATHER_FUNCTION(EA, OFFSET, IDX, LEN, 4, IDX, 1); \
179     } while (0)
180 #define fVLOG_VTCM_GATHER_HALFWORD(EA, OFFSET, IDX, LEN) \
181     do { \
182         GATHER_FUNCTION(EA, OFFSET, IDX, LEN, 2, IDX, 1); \
183     } while (0)
184 #define fVLOG_VTCM_GATHER_HALFWORD_DV(EA, OFFSET, IDX, IDX2, IDX_H, LEN) \
185     do { \
186         GATHER_FUNCTION(EA, OFFSET, IDX, LEN, 2, (2 * IDX2 + IDX_H), 1); \
187     } while (0)
188 #define fVLOG_VTCM_GATHER_WORDQ(EA, OFFSET, IDX, Q, LEN) \
189     do { \
190         GATHER_FUNCTION(EA, OFFSET, IDX, LEN, 4, IDX, \
191                         fGETQBIT(QsV, 4 * IDX + i0)); \
192     } while (0)
193 #define fVLOG_VTCM_GATHER_HALFWORDQ(EA, OFFSET, IDX, Q, LEN) \
194     do { \
195         GATHER_FUNCTION(EA, OFFSET, IDX, LEN, 2, IDX, \
196                         fGETQBIT(QsV, 2 * IDX + i0)); \
197     } while (0)
198 #define fVLOG_VTCM_GATHER_HALFWORDQ_DV(EA, OFFSET, IDX, IDX2, IDX_H, Q, LEN) \
199     do { \
200         GATHER_FUNCTION(EA, OFFSET, IDX, LEN, 2, (2 * IDX2 + IDX_H), \
201                         fGETQBIT(QsV, 2 * IDX + i0)); \
202     } while (0)
203 #define SCATTER_OP_WRITE_TO_MEM(TYPE) \
204     do { \
205         uintptr_t ra = GETPC(); \
206         for (int i = 0; i < sizeof(MMVector); i += sizeof(TYPE)) { \
207             if (test_bit(i, env->vtcm_log.mask)) { \
208                 TYPE dst = 0; \
209                 TYPE inc = 0; \
210                 for (int j = 0; j < sizeof(TYPE); j++) { \
211                     uint8_t val; \
212                     val = cpu_ldub_data_ra(env, env->vtcm_log.va[i + j], ra); \
213                     dst |= val << (8 * j); \
214                     inc |= env->vtcm_log.data.ub[j + i] << (8 * j); \
215                     clear_bit(j + i, env->vtcm_log.mask); \
216                     env->vtcm_log.data.ub[j + i] = 0; \
217                 } \
218                 dst += inc; \
219                 for (int j = 0; j < sizeof(TYPE); j++) { \
220                     cpu_stb_data_ra(env, env->vtcm_log.va[i + j], \
221                                     (dst >> (8 * j)) & 0xFF, ra); \
222                 } \
223             } \
224         } \
225     } while (0)
226 #define SCATTER_OP_PROBE_MEM(TYPE, MMU_IDX, RETADDR) \
227     do { \
228         for (int i = 0; i < sizeof(MMVector); i += sizeof(TYPE)) { \
229             if (test_bit(i, env->vtcm_log.mask)) { \
230                 for (int j = 0; j < sizeof(TYPE); j++) { \
231                     probe_read(env, env->vtcm_log.va[i + j], 1, \
232                                MMU_IDX, RETADDR); \
233                     probe_write(env, env->vtcm_log.va[i + j], 1, \
234                                 MMU_IDX, RETADDR); \
235                 } \
236             } \
237         } \
238     } while (0)
239 #define SCATTER_FUNCTION(EA, OFFSET, IDX, LEN, ELEM_SIZE, BANK_IDX, QVAL, IN) \
240     do { \
241         int i0; \
242         target_ulong va = EA; \
243         target_ulong va_high = EA + LEN; \
244         int log_byte = 0; \
245         for (i0 = 0; i0 < ELEM_SIZE; i0++) { \
246             log_byte = ((va + i0) <= va_high) && QVAL; \
247             LOG_VTCM_BYTE(va + i0, log_byte, IN.ub[ELEM_SIZE * IDX + i0], \
248                           ELEM_SIZE * IDX + i0); \
249         } \
250     } while (0)
251 #define fVLOG_VTCM_HALFWORD(EA, OFFSET, IN, IDX, LEN) \
252     do { \
253         SCATTER_FUNCTION(EA, OFFSET, IDX, LEN, 2, IDX, 1, IN); \
254     } while (0)
255 #define fVLOG_VTCM_WORD(EA, OFFSET, IN, IDX, LEN) \
256     do { \
257         SCATTER_FUNCTION(EA, OFFSET, IDX, LEN, 4, IDX, 1, IN); \
258     } while (0)
259 #define fVLOG_VTCM_HALFWORDQ(EA, OFFSET, IN, IDX, Q, LEN) \
260     do { \
261         SCATTER_FUNCTION(EA, OFFSET, IDX, LEN, 2, IDX, \
262                          fGETQBIT(QsV, 2 * IDX + i0), IN); \
263     } while (0)
264 #define fVLOG_VTCM_WORDQ(EA, OFFSET, IN, IDX, Q, LEN) \
265     do { \
266         SCATTER_FUNCTION(EA, OFFSET, IDX, LEN, 4, IDX, \
267                          fGETQBIT(QsV, 4 * IDX + i0), IN); \
268     } while (0)
269 #define fVLOG_VTCM_HALFWORD_DV(EA, OFFSET, IN, IDX, IDX2, IDX_H, LEN) \
270     do { \
271         SCATTER_FUNCTION(EA, OFFSET, IDX, LEN, 2, \
272                          (2 * IDX2 + IDX_H), 1, IN); \
273     } while (0)
274 #define fVLOG_VTCM_HALFWORDQ_DV(EA, OFFSET, IN, IDX, Q, IDX2, IDX_H, LEN) \
275     do { \
276         SCATTER_FUNCTION(EA, OFFSET, IDX, LEN, 2, (2 * IDX2 + IDX_H), \
277                          fGETQBIT(QsV, 2 * IDX + i0), IN); \
278     } while (0)
279 #define fSTORERELEASE(EA, TYPE) \
280     do { \
281         fV_AL_CHECK(EA, fVECSIZE() - 1); \
282     } while (0)
283 #ifdef QEMU_GENERATE
284 #define fLOADMMV(EA, DST) gen_vreg_load(ctx, DST##_off, EA, true)
285 #endif
286 #ifdef QEMU_GENERATE
287 #define fLOADMMVU(EA, DST) gen_vreg_load(ctx, DST##_off, EA, false)
288 #endif
289 #ifdef QEMU_GENERATE
290 #define fSTOREMMV(EA, SRC) \
291     gen_vreg_store(ctx, insn, pkt, EA, SRC##_off, insn->slot, true)
292 #endif
293 #ifdef QEMU_GENERATE
294 #define fSTOREMMVQ(EA, SRC, MASK) \
295     gen_vreg_masked_store(ctx, EA, SRC##_off, MASK##_off, insn->slot, false)
296 #endif
297 #ifdef QEMU_GENERATE
298 #define fSTOREMMVNQ(EA, SRC, MASK) \
299     gen_vreg_masked_store(ctx, EA, SRC##_off, MASK##_off, insn->slot, true)
300 #endif
301 #ifdef QEMU_GENERATE
302 #define fSTOREMMVU(EA, SRC) \
303     gen_vreg_store(ctx, insn, pkt, EA, SRC##_off, insn->slot, false)
304 #endif
305 #define fVFOREACH(WIDTH, VAR) for (VAR = 0; VAR < fVELEM(WIDTH); VAR++)
306 #define fVARRAY_ELEMENT_ACCESS(ARRAY, TYPE, INDEX) \
307     ARRAY.v[(INDEX) / (fVECSIZE() / (sizeof(ARRAY.TYPE[0])))].TYPE[(INDEX) % \
308     (fVECSIZE() / (sizeof(ARRAY.TYPE[0])))]
309 
310 #define fVSATDW(U, V) fVSATW(((((long long)U) << 32) | fZXTN(32, 64, V)))
311 #define fVASL_SATHI(U, V) fVSATW(((U) << 1) | ((V) >> 31))
312 #define fVUADDSAT(WIDTH, U, V) \
313     fVSATUN(WIDTH, fZXTN(WIDTH, 2 * WIDTH, U) + fZXTN(WIDTH, 2 * WIDTH, V))
314 #define fVSADDSAT(WIDTH, U, V) \
315     fVSATN(WIDTH, fSXTN(WIDTH, 2 * WIDTH, U) + fSXTN(WIDTH, 2 * WIDTH, V))
316 #define fVUSUBSAT(WIDTH, U, V) \
317     fVSATUN(WIDTH, fZXTN(WIDTH, 2 * WIDTH, U) - fZXTN(WIDTH, 2 * WIDTH, V))
318 #define fVSSUBSAT(WIDTH, U, V) \
319     fVSATN(WIDTH, fSXTN(WIDTH, 2 * WIDTH, U) - fSXTN(WIDTH, 2 * WIDTH, V))
320 #define fVAVGU(WIDTH, U, V) \
321     ((fZXTN(WIDTH, 2 * WIDTH, U) + fZXTN(WIDTH, 2 * WIDTH, V)) >> 1)
322 #define fVAVGURND(WIDTH, U, V) \
323     ((fZXTN(WIDTH, 2 * WIDTH, U) + fZXTN(WIDTH, 2 * WIDTH, V) + 1) >> 1)
324 #define fVNAVGU(WIDTH, U, V) \
325     ((fZXTN(WIDTH, 2 * WIDTH, U) - fZXTN(WIDTH, 2 * WIDTH, V)) >> 1)
326 #define fVNAVGURNDSAT(WIDTH, U, V) \
327     fVSATUN(WIDTH, ((fZXTN(WIDTH, 2 * WIDTH, U) - \
328                      fZXTN(WIDTH, 2 * WIDTH, V) + 1) >> 1))
329 #define fVAVGS(WIDTH, U, V) \
330     ((fSXTN(WIDTH, 2 * WIDTH, U) + fSXTN(WIDTH, 2 * WIDTH, V)) >> 1)
331 #define fVAVGSRND(WIDTH, U, V) \
332     ((fSXTN(WIDTH, 2 * WIDTH, U) + fSXTN(WIDTH, 2 * WIDTH, V) + 1) >> 1)
333 #define fVNAVGS(WIDTH, U, V) \
334     ((fSXTN(WIDTH, 2 * WIDTH, U) - fSXTN(WIDTH, 2 * WIDTH, V)) >> 1)
335 #define fVNAVGSRND(WIDTH, U, V) \
336     ((fSXTN(WIDTH, 2 * WIDTH, U) - fSXTN(WIDTH, 2 * WIDTH, V) + 1) >> 1)
337 #define fVNAVGSRNDSAT(WIDTH, U, V) \
338     fVSATN(WIDTH, ((fSXTN(WIDTH, 2 * WIDTH, U) - \
339                     fSXTN(WIDTH, 2 * WIDTH, V) + 1) >> 1))
340 #define fVNOROUND(VAL, SHAMT) VAL
341 #define fVNOSAT(VAL) VAL
342 #define fVROUND(VAL, SHAMT) \
343     ((VAL) + (((SHAMT) > 0) ? (1LL << ((SHAMT) - 1)) : 0))
344 #define fCARRY_FROM_ADD32(A, B, C) \
345     (((fZXTN(32, 64, A) + fZXTN(32, 64, B) + C) >> 32) & 1)
346 #define fUARCH_NOTE_PUMP_4X()
347 #define fUARCH_NOTE_PUMP_2X()
348 
349 #define IV1DEAD()
350 #endif
351