Mercurial > vec
view 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 source
/** * 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); }