Last active
October 19, 2025 02:46
-
-
Save estshorter/19717de2d18780d6d754cd3efd3c8021 to your computer and use it in GitHub Desktop.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| // clang -fomit-frame-pointer -target arm-linux-gnueabihf -march=armv8-a -mcpu=cortex-a53 -mfloat-abi=hard -marm | |
| // --sysroot=/usr/lib/arm-linux-gnueabihf test.c | |
| // のようにomit-frame-pointerを指定していても、オプション指定なし(O0）だと以下のように怒られる。 | |
| // test.c:12:9: warning: inline asm clobber list contains reserved registers: R11 [-Winline-asm] | |
| // 12 | "sub sp, sp, #0x200\n\t" | |
| // | ^ | |
| // test.c:12:9: note: Reserved registers on the clobber list may not be preserved across the asm statement, and clobbering them may lead to undefined behaviour. | |
| // 1 warning generated. | |
| // O1以上を指定すると出ない。 | |
| #include <stdio.h> | |
| #include <stdint.h> | |
| typedef int32_t sp_digit; | |
| static void sp_2048_mul_64(sp_digit* r_p, const sp_digit* a_p, const sp_digit* b_p) | |
| { | |
| register sp_digit* r asm ("r0") = r_p; | |
| register const sp_digit* a asm ("r1") = a_p; | |
| register const sp_digit* b asm ("r2") = b_p; | |
| __asm__ __volatile__ ( | |
| "sub sp, sp, #0x200\n\t" | |
| "mov r5, #0\n\t" | |
| "mov r6, #0\n\t" | |
| "mov r7, #0\n\t" | |
| "mov r8, #0\n\t" | |
| "\n" | |
| "L_sp_2048_mul_64_outer_%=: \n\t" | |
| "subs r3, r5, #0xfc\n\t" | |
| "it cc\n\t" | |
| "movcc r3, #0\n\t" | |
| "sub r4, r5, r3\n\t" | |
| "\n" | |
| "L_sp_2048_mul_64_inner_%=: \n\t" | |
| "ldr lr, [%[a], r3]\n\t" | |
| "ldr r11, [%[b], r4]\n\t" | |
| "umull r9, r10, lr, r11\n\t" | |
| "adds r6, r6, r9\n\t" | |
| "adcs r7, r7, r10\n\t" | |
| "adc r8, r8, #0\n\t" | |
| "add r3, r3, #4\n\t" | |
| "sub r4, r4, #4\n\t" | |
| "cmp r3, #0x100\n\t" | |
| "beq L_sp_2048_mul_64_inner_done_%=\n\t" | |
| "cmp r3, r5\n\t" | |
| "ble L_sp_2048_mul_64_inner_%=\n\t" | |
| "\n" | |
| "L_sp_2048_mul_64_inner_done_%=: \n\t" | |
| "str r6, [sp, r5]\n\t" | |
| "mov r6, r7\n\t" | |
| "mov r7, r8\n\t" | |
| "mov r8, #0\n\t" | |
| "add r5, r5, #4\n\t" | |
| "cmp r5, #0x1f8\n\t" | |
| "ble L_sp_2048_mul_64_outer_%=\n\t" | |
| "str r6, [sp, r5]\n\t" | |
| "\n" | |
| "L_sp_2048_mul_64_store_%=: \n\t" | |
| "ldm sp!, {r6, r7, r8, r9}\n\t" | |
| "stm %[r]!, {r6, r7, r8, r9}\n\t" | |
| "subs r5, r5, #16\n\t" | |
| "bgt L_sp_2048_mul_64_store_%=\n\t" | |
| : [r] "+r" (r), [a] "+r" (a), [b] "+r" (b) | |
| : | |
| : "memory", "r3", "r4", "r5", "r6", "r7", "r8", "r9", "r10", "lr", "r11" | |
| ); | |
| } | |
| int main(void) { | |
| sp_digit r_p[32]; | |
| sp_digit a_p[32]; | |
| sp_digit b_p[32]; | |
| for (int i=0;i<32;i++) { | |
| r_p[i] = 0; | |
| a_p[i] = 2; | |
| b_p[i] = 2; | |
| } | |
| sp_2048_mul_64(r_p, a_p, b_p); | |
| for (int i=0;i<32;i++) { | |
| printf("i: %02d, r: %d, a: %d, b: %d", i, r_p[i], a_p[i], b_p[i]); | |
| } | |
| } |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment