Mercurial > vec
diff utils/gengcc.c @ 40:55cadb1fac4b
*: add mod operation, add GCC vector backend
need to test it with old gcc though. :)
author | Paper <paper@tflc.us> |
---|---|
date | Sun, 27 Apr 2025 02:49:53 -0400 |
parents | |
children | c6e0df09b86f |
line wrap: on
line diff
--- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/utils/gengcc.c Sun Apr 27 02:49:53 2025 -0400 @@ -0,0 +1,390 @@ +/** + * vec - a tiny SIMD vector library in C99 + * + * Copyright (c) 2024-2025 Paper + * + * Permission is hereby granted, free of charge, to any person obtaining a copy + * of this software and associated documentation files (the "Software"), to deal + * in the Software without restriction, including without limitation the rights + * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell + * copies of the Software, and to permit persons to whom the Software is + * furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in all + * copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE + * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. +**/ + +/* Use this file to generate include/vec/impl/generic.h !! + * + * `gcc -o gengeneric gengeneric.c` */ + +#include <stdio.h> +#include <stdlib.h> +#include <string.h> +#include <ctype.h> + +#define ARRAY_SIZE(x) (sizeof(x)/sizeof((x)[0])) + +/* ------------------------------------------------------------------------ */ + +enum op { + /* return vector, take in a integer */ + OP_SPLAT = 0, + + /* return vector, take in an array */ + OP_LOAD_ALIGNED, + OP_LOAD, + + /* void, take in vector and array */ + OP_STORE_ALIGNED, + OP_STORE, + + /* return vector, takes in two vectors */ + OP_ADD, + OP_SUB, + OP_MUL, + OP_AND, + OP_OR, + OP_XOR, + OP_CMPLT, + OP_CMPEQ, + OP_CMPGT, + OP_CMPLE, + OP_CMPGE, + OP_MIN, + OP_MAX, + OP_AVG, + + /* return vector, takes in a vector and an explicitly unsigned vector */ + OP_LSHIFT, + OP_RSHIFT, + OP_LRSHIFT, + + /* return vector, takes in a vector */ + OP_NOT, + + OP_FINAL_, + + /* operations that have some sort of "caveat" should go here, until + * they are fixed or removed */ + + OP_DIV, /* this one causes a floating point exception on my machine. + * possibly we could change the behavior of divide-by-zero + * with some gcc pragma ? --paper */ + OP_MOD, /* ditto with the above */ +}; + +/* convert op -> string */ +static struct { + const char *u; + const char *l; +} op_names[] = { + [OP_SPLAT] = {"SPLAT", "splat"}, + [OP_LOAD_ALIGNED] = {"LOAD_ALIGNED", "load_aligned"}, + [OP_LOAD] = {"LOAD", "load"}, + [OP_STORE_ALIGNED] = {"STORE_ALIGNED", "store_aligned"}, + [OP_STORE] = {"STORE", "store"}, + [OP_ADD] = {"ADD", "add"}, + [OP_SUB] = {"SUB", "sub"}, + [OP_MUL] = {"MUL", "mul"}, + /*[OP_DIV] = {"DIV", "div"},*/ + [OP_AVG] = {"AVG", "avg"}, + [OP_AND] = {"AND", "and"}, + [OP_OR] = {"OR", "or"}, + [OP_XOR] = {"XOR", "xor"}, + [OP_NOT] = {"NOT", "not"}, + [OP_CMPLT] = {"CMPLT", "cmplt"}, + [OP_CMPEQ] = {"CMPEQ", "cmpeq"}, + [OP_CMPGT] = {"CMPGT", "cmpgt"}, + [OP_CMPLE] = {"CMPLE", "cmple"}, + [OP_CMPGE] = {"CMPGE", "cmpge"}, + [OP_MIN] = {"MIN", "min"}, + [OP_MAX] = {"MAX", "max"}, + [OP_RSHIFT] = {"RSHIFT", "rshift"}, + [OP_LRSHIFT] = {"LRSHIFT", "lrshift"}, + [OP_LSHIFT] = {"LSHIFT", "lshift"}, +}; + +#define UPSIGN(x) ((x) ? "" : "U") +#define LOSIGN(x) ((x) ? "" : "u") + +static void print_gcc_op(enum op op, int is_signed, int bits, int size) +{ + int i; + + printf("#ifndef V%sINT%dx%d_%s_DEFINED\n", UPSIGN(is_signed), bits, size, op_names[op].u); + + printf("VEC_FUNC_IMPL "); + + /* first; the return value */ + switch (op) { + case OP_SPLAT: + case OP_LOAD_ALIGNED: + case OP_LOAD: + case OP_ADD: + case OP_SUB: + case OP_MUL: + case OP_DIV: + case OP_AND: + case OP_OR: + case OP_XOR: + case OP_CMPLT: + case OP_CMPEQ: + case OP_CMPGT: + case OP_CMPLE: + case OP_CMPGE: + case OP_MIN: + case OP_MAX: + case OP_AVG: + case OP_RSHIFT: + case OP_LRSHIFT: + case OP_LSHIFT: + case OP_NOT: + printf("v%sint%dx%d", LOSIGN(is_signed), bits, size); + break; + case OP_STORE_ALIGNED: + case OP_STORE: + printf("void"); + break; + } + + /* whitespace and function name */ + printf(" v%sint%dx%d_%s(", LOSIGN(is_signed), bits, size, op_names[op].l); + + /* parameters */ + switch (op) { + case OP_SPLAT: + printf("vec_%sint%d x", LOSIGN(is_signed), bits); + break; + case OP_LOAD_ALIGNED: + case OP_LOAD: + printf("const vec_%sint%d x[%d]", LOSIGN(is_signed), bits, size); + break; + case OP_STORE_ALIGNED: + case OP_STORE: + printf("v%sint%dx%d vec, vec_%sint%d arr[%d]", LOSIGN(is_signed), bits, size, LOSIGN(is_signed), bits, size); + break; + case OP_ADD: + case OP_SUB: + case OP_MUL: + case OP_DIV: + case OP_AND: + case OP_OR: + case OP_XOR: + case OP_CMPLT: + case OP_CMPEQ: + case OP_CMPGT: + case OP_CMPLE: + case OP_CMPGE: + case OP_MIN: + case OP_MAX: + case OP_AVG: + printf("v%sint%dx%d vec1, v%sint%dx%d vec2", LOSIGN(is_signed), bits, size, LOSIGN(is_signed), bits, size); + break; + case OP_RSHIFT: + case OP_LRSHIFT: + case OP_LSHIFT: + printf("v%sint%dx%d vec1, vuint%dx%d vec2", LOSIGN(is_signed), bits, size, bits, size); + break; + case OP_NOT: + printf("v%sint%dx%d vec", LOSIGN(is_signed), bits, size); + break; + } + + puts(")\n{"); + + switch (op) { + case OP_SPLAT: + printf("\tv%sint%dx%d vec;\n", LOSIGN(is_signed), bits, size); + printf("\tvec.gcc = (__typeof__(vec.gcc)){"); + for (i = 0; i < size; i++) + printf("x,"); + printf("};\n"); + printf("\treturn vec;\n"); + break; + case OP_LOAD_ALIGNED: + printf("\tv%sint%dx%d vec;\n", LOSIGN(is_signed), bits, size); + puts("\tvec.gcc = *(__typeof__(vec.gcc) *)x;"); + printf("\treturn vec;\n"); + break; + case OP_LOAD: + printf("\tv%sint%dx%d vec;\n", LOSIGN(is_signed), bits, size); + puts("\tmemcpy(&vec, x, sizeof(vec));"); + printf("\treturn vec;\n"); + break; + case OP_STORE_ALIGNED: + puts("\t*(__typeof__(vec.gcc) *)arr = vec.gcc;"); + break; + case OP_STORE: + puts("\tmemcpy(arr, &vec, sizeof(vec));"); + break; + case OP_ADD: + case OP_SUB: + case OP_MUL: + case OP_DIV: + case OP_AND: + case OP_OR: + case OP_XOR: + case OP_CMPLT: + case OP_CMPEQ: + case OP_CMPGT: + case OP_CMPLE: + case OP_CMPGE: { + const char *op_builtins[OP_CMPGE - OP_ADD + 1] = {"+", "-", "*", /*"/", */"&", "|", "^", "<", "==", ">", "<=", ">="}; + + printf("\tvec1.gcc = (vec1.gcc %s vec2.gcc);\n", op_builtins[op - OP_ADD]); + printf("\treturn vec1;\n"); + break; + } + + case OP_LSHIFT: + case OP_RSHIFT: { + const char *op_builtins[OP_RSHIFT - OP_LSHIFT + 1] = {"<<", ">>"}; + + printf("\tvec1.gcc = (vec1.gcc %s vec2.gcc);\n", op_builtins[op - OP_LSHIFT]); + printf("\treturn vec1;\n"); + break; + } + + case OP_LRSHIFT: { + /* sigh */ + printf("\tvec1.gcc = (__typeof__(vec1.gcc))((vec_uint%d __attribute__((__vector_size__(%d))))vec1.gcc >> vec2.gcc);\n", bits, bits * size / 8); + printf("\treturn vec1;\n"); + break; + } + case OP_MIN: + case OP_MAX: { + const char *op_builtins[OP_MAX - OP_MIN + 1] = {"<", ">"}; + + printf("\tv%sint%dx%d mask;\n", LOSIGN(is_signed), bits, size); + printf("\tmask.gcc = (vec1.gcc %s vec2.gcc);\n", op_builtins[op - OP_MIN]); + printf("\tvec1.gcc = (vec1.gcc & mask.gcc) | (vec2.gcc & ~mask.gcc);\n"); + printf("\treturn vec1;\n"); + break; + } + case OP_AVG: + if (is_signed) { + printf("\tvint%dx%d ones = vint%dx%d_splat(1);\n", bits, size, bits, size); + printf("\t__typeof__(vec1.gcc) x_d_rem = (vec1.gcc % 2);\n"); + printf("\t__typeof__(vec1.gcc) y_d_rem = (vec2.gcc % 2);\n"); + printf("\t__typeof__(vec1.gcc) rem_d_quot = ((x_d_rem + y_d_rem) / 2);\n"); + printf("\t__typeof__(vec1.gcc) rem_d_rem = ((x_d_rem + y_d_rem) % 2);\n"); + puts(""); + printf("\tvec1.gcc = ((vec1.gcc / 2) + (vec2.gcc / 2)) + (rem_d_quot) + ((rem_d_rem == 1) & ones.gcc);\n"); + } else { + printf("\tvec1.gcc = (vec1.gcc >> 1) + (vec2.gcc >> 1) + ((vec1.gcc | vec2.gcc) & 1);\n"); + } + + printf("\treturn vec1;\n"); + break; + case OP_NOT: + printf("\tvec.gcc = ~vec.gcc;\n"); + printf("\treturn vec;\n"); + break; + default: + printf("#error implement this operation"); + break; + } + + /* end function definition */ + puts("}"); + + printf("# define V%sINT%dx%d_%s_DEFINED\n", UPSIGN(is_signed), bits, size, op_names[op].u); + puts("#endif"); +} + +static inline void print_ops(int is_signed, int bits, int size) +{ + int i; + + printf("\n\n/* v%sint%dx%d */\n\n", (is_signed ? "u" : ""), bits, size); + + for (i = 0; i < OP_FINAL_; i++) + print_gcc_op(i, is_signed, bits, size); +} + +static const char *header = + "/**\n" + " * vec - a tiny SIMD vector library in C99\n" + " * \n" + " * Copyright (c) 2024-2025 Paper\n" + " * \n" + " * Permission is hereby granted, free of charge, to any person obtaining a copy\n" + " * of this software and associated documentation files (the \"Software\"), to deal\n" + " * in the Software without restriction, including without limitation the rights\n" + " * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell\n" + " * copies of the Software, and to permit persons to whom the Software is\n" + " * furnished to do so, subject to the following conditions:\n" + " * \n" + " * The above copyright notice and this permission notice shall be included in all\n" + " * copies or substantial portions of the Software.\n" + " * \n" + " * THE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR\n" + " * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,\n" + " * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE\n" + " * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER\n" + " * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,\n" + " * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE\n" + " * SOFTWARE.\n" + "**/\n" + "\n" + "/* This file is automatically generated! Do not edit it directly!\n" + " * Edit the code that generates it in utils/gengcc.c --paper */\n" + "\n" + "#ifndef VEC_IMPL_GCC_H_\n" + "#define VEC_IMPL_GCC_H_\n" + "\n"; + +static const char *footer = + "#endif /* VEC_IMPL_GCC_H_ */\n"; + +int main(void) +{ + static struct { + int bits, size; + } defs[] = { + /* -- 8-bit */ + {8, 2}, + {8, 4}, + {8, 8}, + {8, 16}, + {8, 32}, + {8, 64}, + + /* -- 16-bit */ + {16, 2}, + {16, 4}, + {16, 8}, + {16, 16}, + {16, 32}, + + /* -- 32-bit */ + {32, 2}, + {32, 4}, + {32, 8}, + {32, 16}, + + /* -- 64-bit */ + {64, 2}, + {64, 4}, + {64, 8}, + }; + int i; + + puts(header); + + for (i = 0; i < ARRAY_SIZE(defs); i++) { + print_ops(1, defs[i].bits, defs[i].size); + print_ops(0, defs[i].bits, defs[i].size); + } + + puts(footer); +}