5 minor opcode allocation
8 | ------ |--| --------- |
14 | 101 |0 | ternarycr |
20 | dest | src1 | subop | op |
21 | ---- | ---- | ----- | -------- |
22 | RT | RA | .. | bmatflip |
26 | dest | src1 | src2 | subop | op |
27 | ---- | ---- | ---- | ----- | -------- |
28 | RT | RA | RB | or | bmatflip |
29 | RT | RA | RB | xor | bmatflip |
30 | RT | RA | RB | bdep | dep/ext |
31 | RT | RA | RB | bext | dep/ext |
32 | RT | RA | RB | | grev |
33 | RT | RA | RB | | clmul* |
34 | RT | RA | RB | | gorc |
35 | RT | RA | RB | shuf | shuffle |
36 | RT | RA | RB | unshuf| shuffle |
37 | RT | RA | RB | width | xperm |
38 | RT | RA | RB | type | minmax |
49 | 0.5|6.10|11.15|16.20|21..25 | 26....30 |31| name |
50 | -- | -- | --- | --- | ----- | -------- |--| ------ |
51 | NN | RT | RA | RB | RC | mode 001 |Rc| ternary |
52 | NN | RT | RA | RB | im0-4 | im5-7 00 |Rc| ternaryi |
53 | NN | RS | RA | RB | deg | 00 011 |Rc| gfmul |
54 | NN | RS | RA | RB | deg | 01 011 |Rc| gfadd |
55 | NN | RT | RA | RB | deg | 10 011 |Rc| gfinv |
56 | NN | RS | RA | RB | deg | 11 011 |Rc| gf rsvd |
58 | 0.5|6.10|11.15| 16.23 |24.27 | 28.30 |31| name |
59 | -- | -- | --- | ----- | ---- | ----- |--| ------ |
60 | NN | RT | RA | imm | mask | 101 |1 | ternaryv |
62 | 0.5|6.8 | 9.11|12.14|15|16.23|24.27 | 28.30|31| name |
63 | -- | -- | --- | --- |- |-----|----- | -----|--| -------|
64 | NN | BA | BB | BC |0 |imm | mask | 101 |0 | ternarycr |
68 | 0.5|6.10|11.15|16.20| 21.22 | 23 | 24....30 |31| name |
69 | -- | -- | --- | --- | ----- | -- | -------- |--| ---- |
70 | NN | RA | RB | | | 0 | 0000 110 |Rc| rsvd |
71 | NN | RA | RB | RC | itype | 1 | 0000 110 |Rc| xperm |
72 | NN | RA | RB | RC | itype | 0 | 0100 110 |Rc| minmax |
73 | NN | RA | RB | | | 1 | 0100 110 |Rc| rsvd |
74 | NN | RA | RB | sh | itype | SH | 1000 110 |Rc| bmopsi |
75 | NN | RA | RB | | | | 1100 110 |Rc| rsvd |
76 | NN | RA | RB | | | 0 | 0001 110 |Rc| rsvd |
77 | NN | RA | RB | | | 0 | 0101 110 |Rc| rsvd |
78 | NN | RA | RB | RC | 00 | 0 | 0010 110 |Rc| gorc |
79 | NN | RA | RB | sh | 00 | SH | 1010 110 |Rc| gorci |
80 | NN | RA | RB | RC | 00 | 0 | 0110 110 |Rc| gorcw |
81 | NN | RA | RB | sh | 00 | 0 | 1110 110 |Rc| gorcwi |
82 | NN | RA | RB | RC | 00 | 1 | 1110 110 |Rc| bmator |
83 | NN | RA | RB | RC | 01 | 0 | 0010 110 |Rc| grev |
84 | NN | RA | RB | RC | 01 | 1 | 0010 110 |Rc| clmul |
85 | NN | RA | RB | sh | 01 | SH | 1010 110 |Rc| grevi |
86 | NN | RA | RB | RC | 01 | 0 | 0110 110 |Rc| grevw |
87 | NN | RA | RB | sh | 01 | 0 | 1110 110 |Rc| grevwi |
88 | NN | RA | RB | RC | 01 | 1 | 1110 110 |Rc| bmatxor |
89 | NN | RA | RB | RC | 10 | 0 | 0010 110 |Rc| shfl |
90 | NN | RA | RB | sh | 10 | SH | 1010 110 |Rc| shfli |
91 | NN | RA | RB | RC | 10 | 0 | 0110 110 |Rc| shflw |
92 | NN | RA | RB | RC | 10 | 0 | 1110 110 |Rc| bdep |
93 | NN | RA | RB | RC | 10 | 1 | 1110 110 |Rc| bext |
94 | NN | RA | RB | RC | 11 | 0 | 1110 110 |Rc| clmulr |
95 | NN | RA | RB | RC | 11 | 1 | 1110 110 |Rc| clmulh |
96 | NN | RA | RB | | | | NN11 110 |Rc| rsvd |
100 similar to matrix permute in RV bitmanip, which has XOR and OR variants
104 b = VSR[VRB+32].dword[i].byte[k].bit[j]
105 VSR[VRT+32].dword[i].byte[j].bit[k] = b
109 vpdepd VRT,VRA,VRB, identical to RV bitmamip bdep
112 if VSR[VRB+32].dword[i].bit[63-m]=1 then do
113 result = VSR[VRA+32].dword[i].bit[63-k]
114 VSR[VRT+32].dword[i].bit[63-m] = result
120 uint_xlen_t bdep(uint_xlen_t RA, uint_xlen_t RB)
123 for (int i = 0, j = 0; i < XLEN; i++)
126 r |= uint_xlen_t(1) << i;
136 other way round: identical to RV bext
139 uint_xlen_t bext(uint_xlen_t RA, uint_xlen_t RB)
142 for (int i = 0, j = 0; i < XLEN; i++)
145 r |= uint_xlen_t(1) << j;
154 signed and unsigned min/max for integer. this is sort-of partly synthesiseable in [[sv/svp64]] with pred-result as long as the dest reg is one of the sources, but not both signed and unsigned. when the dest is also one of the srces and the mv fails due to the CR bittest failing this will only overwrite the dest where the src is greater (or less).
156 signed/unsigned min/max gives more flexibility.
159 uint_xlen_t min(uint_xlen_t rs1, uint_xlen_t rs2)
160 { return (int_xlen_t)rs1 < (int_xlen_t)rs2 ? rs1 : rs2;
162 uint_xlen_t max(uint_xlen_t rs1, uint_xlen_t rs2)
163 { return (int_xlen_t)rs1 > (int_xlen_t)rs2 ? rs1 : rs2;
165 uint_xlen_t minu(uint_xlen_t rs1, uint_xlen_t rs2)
166 { return rs1 < rs2 ? rs1 : rs2;
168 uint_xlen_t maxu(uint_xlen_t rs1, uint_xlen_t rs2)
169 { return rs1 > rs2 ? rs1 : rs2;
176 Similar to FPGA LUTs: for every bit perform a lookup into a table using an 8bit immediate, or in another register
178 | 0.5|6.10|11.15|16.20| 21..25| 26..30 |31|
179 | -- | -- | --- | --- | ----- | -------- |--|
180 | NN | RT | RA | RB | im0-4 | im5-7 00 |Rc|
183 idx = RT[i] << 2 | RA[i] << 1 | RB[i]
184 RT[i] = (imm & (1<<idx)) != 0
186 bits 21..22 may be used to specify a mode, such as treating the whole integer zero/nonzero and putting 1/0 in the result, rather than bitwise test.
188 a 4 operand variant which becomes more along the lines of an FPGA:
190 | 0.5|6.10|11.15|16.20|21.25| 26...30 |31|
191 | -- | -- | --- | --- | --- | -------- |--|
192 | NN | RT | RA | RB | RC | mode 001 |Rc|
195 idx = RT[i] << 2 | RA[i] << 1 | RB[i]
196 RT[i] = (RC & (1<<idx)) != 0
198 mode (2 bit) may be used to do inversion of ordering, similar to carryless mul,
201 also, another possible variant involving swizzle and vec4:
203 | 0.5|6.10|11.15| 16.23 |24.27 | 28.30 |31|
204 | -- | -- | --- | ----- | ---- | ----- |--|
205 | NN | RT | RA | imm | mask | 101 |1 |
208 idx = RA.x[i] << 2 | RA.y[i] << 1 | RA.z[i]
209 res = (imm & (1<<idx)) != 0
211 if mask[j]: RT[i+j*8] = res
213 another mode selection would be CRs not Ints.
215 | 0.5|6.8 | 9.11|12.14|15|16.23|24.27 | 28.30|31|
216 | -- | -- | --- | --- |- |-----|----- | -----|--|
217 | NN | BA | BB | BC |0 |imm | mask | 101 |0 |
220 if not mask[i] continue
221 idx = crregs[BA][i] << 2 |
224 crregs[BA][i] = (imm & (1<<idx)) != 0
228 based on RV bitmanip singlebit set, instruction format similar to shift
229 [[isa/fixedshift]]. bmext is actually covered already (shift-with-mask rldicl but only immediate version).
230 however bitmask-invert is not, and set/clr are not covered, although they can use the same Shift ALU.
232 bmext (RB) version is not the same as rldicl because bmext is a right shift by RC, where rldicl is a left rotate. for the immediate version this does not matter.
234 | 0.5|6.10|11.15|16.20|21.25| 26..30 |31|
235 | -- | -- | --- | --- | --- | ------- |--|
236 | NN | RT | RA | RB | RC | mode 010 |Rc|
239 uint_xlen_t bmset(RA, RB, sh)
241 int shamt = RB & (XLEN - 1);
243 return RA | (mask << shamt);
246 uint_xlen_t bmclr(RA, RB, sh)
248 int shamt = RB & (XLEN - 1);
250 return RA & ~(mask << shamt);
253 uint_xlen_t bminv(RA, RB, sh)
255 int shamt = RB & (XLEN - 1);
257 return RA ^ (mask << shamt);
260 uint_xlen_t bmext(RA, RB, sh)
262 int shamt = RB & (XLEN - 1);
264 return mask & (RA >> shamt);
268 bitmask extract with reverse
271 rev[0:msb] = ra[msb:0];
280 uint64_t grev64(uint64_t RA, uint64_t RB)
284 if (shamt & 1) x = ((x & 0x5555555555555555LL) << 1) |
285 ((x & 0xAAAAAAAAAAAAAAAALL) >> 1);
286 if (shamt & 2) x = ((x & 0x3333333333333333LL) << 2) |
287 ((x & 0xCCCCCCCCCCCCCCCCLL) >> 2);
288 if (shamt & 4) x = ((x & 0x0F0F0F0F0F0F0F0FLL) << 4) |
289 ((x & 0xF0F0F0F0F0F0F0F0LL) >> 4);
290 if (shamt & 8) x = ((x & 0x00FF00FF00FF00FFLL) << 8) |
291 ((x & 0xFF00FF00FF00FF00LL) >> 8);
292 if (shamt & 16) x = ((x & 0x0000FFFF0000FFFFLL) << 16) |
293 ((x & 0xFFFF0000FFFF0000LL) >> 16);
294 if (shamt & 32) x = ((x & 0x00000000FFFFFFFFLL) << 32) |
295 ((x & 0xFFFFFFFF00000000LL) >> 32);
301 # shuffle / unshuffle
306 uint32_t shfl32(uint32_t RA, uint32_t RB)
310 if (shamt & 8) x = shuffle32_stage(x, 0x00ff0000, 0x0000ff00, 8);
311 if (shamt & 4) x = shuffle32_stage(x, 0x0f000f00, 0x00f000f0, 4);
312 if (shamt & 2) x = shuffle32_stage(x, 0x30303030, 0x0c0c0c0c, 2);
313 if (shamt & 1) x = shuffle32_stage(x, 0x44444444, 0x22222222, 1);
316 uint32_t unshfl32(uint32_t RA, uint32_t RB)
320 if (shamt & 1) x = shuffle32_stage(x, 0x44444444, 0x22222222, 1);
321 if (shamt & 2) x = shuffle32_stage(x, 0x30303030, 0x0c0c0c0c, 2);
322 if (shamt & 4) x = shuffle32_stage(x, 0x0f000f00, 0x00f000f0, 4);
323 if (shamt & 8) x = shuffle32_stage(x, 0x00ff0000, 0x0000ff00, 8);
327 uint64_t shuffle64_stage(uint64_t src, uint64_t maskL, uint64_t maskR, int N)
329 uint64_t x = src & ~(maskL | maskR);
330 x |= ((src << N) & maskL) | ((src >> N) & maskR);
333 uint64_t shfl64(uint64_t RA, uint64_t RB)
337 if (shamt & 16) x = shuffle64_stage(x, 0x0000ffff00000000LL,
338 0x00000000ffff0000LL, 16);
339 if (shamt & 8) x = shuffle64_stage(x, 0x00ff000000ff0000LL,
340 0x0000ff000000ff00LL, 8);
341 if (shamt & 4) x = shuffle64_stage(x, 0x0f000f000f000f00LL,
342 0x00f000f000f000f0LL, 4);
343 if (shamt & 2) x = shuffle64_stage(x, 0x3030303030303030LL,
344 0x0c0c0c0c0c0c0c0cLL, 2);
345 if (shamt & 1) x = shuffle64_stage(x, 0x4444444444444444LL,
346 0x2222222222222222LL, 1);
349 uint64_t unshfl64(uint64_t RA, uint64_t RB)
353 if (shamt & 1) x = shuffle64_stage(x, 0x4444444444444444LL,
354 0x2222222222222222LL, 1);
355 if (shamt & 2) x = shuffle64_stage(x, 0x3030303030303030LL,
356 0x0c0c0c0c0c0c0c0cLL, 2);
357 if (shamt & 4) x = shuffle64_stage(x, 0x0f000f000f000f00LL,
358 0x00f000f000f000f0LL, 4);
359 if (shamt & 8) x = shuffle64_stage(x, 0x00ff000000ff0000LL,
360 0x0000ff000000ff00LL, 8);
361 if (shamt & 16) x = shuffle64_stage(x, 0x0000ffff00000000LL,
362 0x00000000ffff0000LL, 16);
372 uint_xlen_t xperm(uint_xlen_t RA, uint_xlen_t RB, int sz_log2)
375 uint_xlen_t sz = 1LL << sz_log2;
376 uint_xlen_t mask = (1LL << sz) - 1;
377 for (int i = 0; i < XLEN; i += sz) {
378 uint_xlen_t pos = ((RB >> i) & mask) << sz_log2;
380 r |= ((RA >> pos) & mask) << i;
384 uint_xlen_t xperm_n (uint_xlen_t RA, uint_xlen_t RB)
385 { return xperm(RA, RB, 2); }
386 uint_xlen_t xperm_b (uint_xlen_t RA, uint_xlen_t RB)
387 { return xperm(RA, RB, 3); }
388 uint_xlen_t xperm_h (uint_xlen_t RA, uint_xlen_t RB)
389 { return xperm(RA, RB, 4); }
390 uint_xlen_t xperm_w (uint_xlen_t RA, uint_xlen_t RB)
391 { return xperm(RA, RB, 5); }
399 uint32_t gorc32(uint32_t RA, uint32_t RB)
403 if (shamt & 1) x |= ((x & 0x55555555) << 1) | ((x & 0xAAAAAAAA) >> 1);
404 if (shamt & 2) x |= ((x & 0x33333333) << 2) | ((x & 0xCCCCCCCC) >> 2);
405 if (shamt & 4) x |= ((x & 0x0F0F0F0F) << 4) | ((x & 0xF0F0F0F0) >> 4);
406 if (shamt & 8) x |= ((x & 0x00FF00FF) << 8) | ((x & 0xFF00FF00) >> 8);
407 if (shamt & 16) x |= ((x & 0x0000FFFF) << 16) | ((x & 0xFFFF0000) >> 16);
410 uint64_t gorc64(uint64_t RA, uint64_t RB)
414 if (shamt & 1) x |= ((x & 0x5555555555555555LL) << 1) |
415 ((x & 0xAAAAAAAAAAAAAAAALL) >> 1);
416 if (shamt & 2) x |= ((x & 0x3333333333333333LL) << 2) |
417 ((x & 0xCCCCCCCCCCCCCCCCLL) >> 2);
418 if (shamt & 4) x |= ((x & 0x0F0F0F0F0F0F0F0FLL) << 4) |
419 ((x & 0xF0F0F0F0F0F0F0F0LL) >> 4);
420 if (shamt & 8) x |= ((x & 0x00FF00FF00FF00FFLL) << 8) |
421 ((x & 0xFF00FF00FF00FF00LL) >> 8);
422 if (shamt & 16) x |= ((x & 0x0000FFFF0000FFFFLL) << 16) |
423 ((x & 0xFFFF0000FFFF0000LL) >> 16);
424 if (shamt & 32) x |= ((x & 0x00000000FFFFFFFFLL) << 32) |
425 ((x & 0xFFFFFFFF00000000LL) >> 32);
433 based on RV bitmanip, covered by ternary bitops
436 uint_xlen_t cmix(uint_xlen_t RA, uint_xlen_t RB, uint_xlen_t RC) {
437 return (RA & RB) | (RC & ~RB);
444 see https://en.wikipedia.org/wiki/CLMUL_instruction_set
447 uint_xlen_t clmul(uint_xlen_t RA, uint_xlen_t RB)
450 for (int i = 0; i < XLEN; i++)
455 uint_xlen_t clmulh(uint_xlen_t RA, uint_xlen_t RB)
458 for (int i = 1; i < XLEN; i++)
463 uint_xlen_t clmulr(uint_xlen_t RA, uint_xlen_t RB)
466 for (int i = 0; i < XLEN; i++)
468 x ^= RA >> (XLEN-i-1);
476 this requires 3 parameters and a "degree"
478 RT = GFMUL(RA, RB, gfdegree, modulo=RC)
480 realistically with the degree also needing to be an immediate it should be brought down to an overwrite version:
482 RS = GFMUL(RS, RA, gfdegree, modulo=RB)
484 | 0.5|6.10|11.15|16.20|21.25| 26..30 |31|
485 | -- | -- | --- | --- | --- | ------- |--|
486 | NN | RS | RA | RB | deg | 00 011 |Rc|
488 where the SimpleV variant may override RS-as-src differently from RS-as-dest
493 from functools import reduce
495 # constants used in the multGF2 function
496 mask1 = mask2 = polyred = None
498 def setGF2(degree, irPoly):
499 """Define parameters of binary finite field GF(2^m)/g(x)
500 - degree: extension degree of binary field
501 - irPoly: coefficients of irreducible polynomial g(x)
504 """Convert an integer into a polynomial"""
505 return [(sInt >> i) & 1
506 for i in reversed(range(sInt.bit_length()))]
508 global mask1, mask2, polyred
509 mask1 = mask2 = 1 << degree
511 polyred = reduce(lambda x, y: (x << 1) + y, i2P(irPoly)[1:])
514 """Multiply two polynomials in GF(2^m)/g(x)"""
525 if __name__ == "__main__":
527 # Define binary field GF(2^3)/x^3 + x + 1
530 # Evaluate the product (x^2 + x + 1)(x^2 + 1)
531 print("{:02x}".format(multGF2(0b111, 0b101)))
533 # Define binary field GF(2^8)/x^8 + x^4 + x^3 + x + 1
534 # (used in the Advanced Encryption Standard-AES)
535 setGF2(8, 0b100011011)
537 # Evaluate the product (x^7)(x^7 + x + 1)
538 print("{:02x}".format(multGF2(0b10000000, 0b10000011)))
542 RS = GFADD(RS, RA|0, gfdegree, modulo=RB)
544 | 0.5|6.10|11.15|16.20|21.25| 26..30 |31|
545 | -- | -- | --- | --- | --- | ------- |--|
546 | NN | RS | RA | RB | deg | 01 011 |Rc|
559 def gf_invert(a, mod=0x1B) :
574 a %= 256 # Emulating 8-bit overflow
575 g1 %= 256 # Emulating 8-bit overflow
577 j = gf_degree(a) - gf_degree(v)
585 uint64_t bmatflip(uint64_t RA)
593 uint64_t bmatxor(uint64_t RA, uint64_t RB)
596 uint64_t RBt = bmatflip(RB);
597 uint8_t u[8]; // rows of RA
598 uint8_t v[8]; // cols of RB
599 for (int i = 0; i < 8; i++) {
604 for (int i = 0; i < 64; i++) {
605 if (pcnt(u[i / 8] & v[i % 8]) & 1)
610 uint64_t bmator(uint64_t RA, uint64_t RB)
613 uint64_t RBt = bmatflip(RB);
614 uint8_t u[8]; // rows of RA
615 uint8_t v[8]; // cols of RB
616 for (int i = 0; i < 8; i++) {
621 for (int i = 0; i < 64; i++) {
622 if ((u[i / 8] & v[i % 8]) != 0)