Add an intermediate optimization level where only add/sub are "fast".
Also made uECC_SQUARE_FUNC separately controllable for now.
diff --git a/asm_arm_fast.inc b/asm_arm.inc
similarity index 66%
rename from asm_arm_fast.inc
rename to asm_arm.inc
index 150afa0..d5bbe57 100644
--- a/asm_arm_fast.inc
+++ b/asm_arm.inc
@@ -1,7 +1,9 @@
/* Copyright 2015, Kenneth MacKay. Licensed under the BSD 2-clause license. */
-#ifndef _UECC_ASM_ARM_FAST_H_
-#define _UECC_ASM_ARM_FAST_H_
+#ifndef _UECC_ASM_ARM_H_
+#define _UECC_ASM_ARM_H_
+
+#include "asm_arm_mult_square.inc"
#if (uECC_SUPPORTS_secp256r1 || uECC_SUPPORTS_secp256k1)
#define uECC_MIN_WORDS 8
@@ -33,6 +35,8 @@
#define RESUME_SYNTAX ".syntax divided \n\t"
#endif
+#if (uECC_OPTIMIZATION_LEVEL >= 2)
+
static uECC_word_t vli_add(uECC_word_t *result,
const uECC_word_t *left,
const uECC_word_t *right,
@@ -186,6 +190,10 @@
}
#define asm_sub 1
+#endif /* (uECC_OPTIMIZATION_LEVEL >= 2) */
+
+#if (uECC_OPTIMIZATION_LEVEL >= 3)
+
#define FAST_MULT_ASM_5_TO_6 \
"cmp r3, #5 \n\t" \
"beq 1f \n\t" \
@@ -842,4 +850,410 @@
#endif /* uECC_PLATFORM != uECC_arm_thumb */
-#endif /* _UECC_ASM_ARM_FAST_H_ */
+#endif /* (uECC_OPTIMIZATION_LEVEL >= 3) */
+
+/* ---- "Small" implementations ---- */
+
+#if !asm_add
+static uECC_word_t vli_add(uECC_word_t *result,
+ const uECC_word_t *left,
+ const uECC_word_t *right,
+ wordcount_t num_words) {
+ uint32_t carry = 0;
+ uint32_t left_word;
+ uint32_t right_word;
+
+ __asm__ volatile (
+ ".syntax unified \n\t"
+ "1: \n\t"
+ "ldmia %[lptr]!, {%[left]} \n\t" /* Load left word. */
+ "ldmia %[rptr]!, {%[right]} \n\t" /* Load right word. */
+ "lsrs %[carry], #1 \n\t" /* Set up carry flag (carry = 0 after this). */
+ "adcs %[left], %[right] \n\t" /* Add with carry. */
+ "adcs %[carry], %[carry] \n\t" /* Store carry bit. */
+ "stmia %[dptr]!, {%[left]} \n\t" /* Store result word. */
+ "subs %[ctr], #1 \n\t" /* Decrement counter. */
+ "bne 1b \n\t" /* Loop until counter == 0. */
+ RESUME_SYNTAX
+ : [dptr] REG_RW (result), [lptr] REG_RW (left), [rptr] REG_RW (right),
+ [ctr] REG_RW (num_words), [carry] REG_RW (carry),
+ [left] REG_WRITE (left_word), [right] REG_WRITE (right_word)
+ :
+ : "cc", "memory"
+ );
+ return carry;
+}
+#define asm_add 1
+#endif
+
+#if !asm_sub
+static uECC_word_t vli_sub(uECC_word_t *result,
+ const uECC_word_t *left,
+ const uECC_word_t *right,
+ wordcount_t num_words) {
+ uint32_t carry = 1; /* carry = 1 initially (means don't borrow) */
+ uint32_t left_word;
+ uint32_t right_word;
+
+ __asm__ volatile (
+ ".syntax unified \n\t"
+ "1: \n\t"
+ "ldmia %[lptr]!, {%[left]} \n\t" /* Load left word. */
+ "ldmia %[rptr]!, {%[right]} \n\t" /* Load right word. */
+ "lsrs %[carry], #1 \n\t" /* Set up carry flag (carry = 0 after this). */
+ "sbcs %[left], %[right] \n\t" /* Subtract with borrow. */
+ "adcs %[carry], %[carry] \n\t" /* Store carry bit. */
+ "stmia %[dptr]!, {%[left]} \n\t" /* Store result word. */
+ "subs %[ctr], #1 \n\t" /* Decrement counter. */
+ "bne 1b \n\t" /* Loop until counter == 0. */
+ RESUME_SYNTAX
+ : [dptr] REG_RW (result), [lptr] REG_RW (left), [rptr] REG_RW (right),
+ [ctr] REG_RW (num_words), [carry] REG_RW (carry),
+ [left] REG_WRITE (left_word), [right] REG_WRITE (right_word)
+ :
+ : "cc", "memory"
+ );
+ return !carry;
+}
+#define asm_sub 1
+#endif
+
+#if !asm_mult
+static void vli_mult(uECC_word_t *result,
+ const uECC_word_t *left,
+ const uECC_word_t *right,
+ wordcount_t num_words) {
+#if (uECC_PLATFORM != uECC_arm_thumb)
+ uint32_t c0 = 0;
+ uint32_t c1 = 0;
+ uint32_t c2 = 0;
+ uint32_t k = 0;
+ uint32_t i;
+ uint32_t t0, t1;
+
+ __asm__ volatile (
+ ".syntax unified \n\t"
+
+ "1: \n\t" /* outer loop (k < num_words) */
+ "movs %[i], #0 \n\t" /* i = 0 */
+ "b 3f \n\t"
+
+ "2: \n\t" /* outer loop (k >= num_words) */
+ "movs %[i], %[k] \n\t" /* i = k */
+ "subs %[i], %[last_word] \n\t" /* i = k - (num_words - 1) (times 4) */
+
+ "3: \n\t" /* inner loop */
+ "subs %[t0], %[k], %[i] \n\t" /* t0 = k-i */
+
+ "ldr %[t1], [%[right], %[t0]] \n\t" /* t1 = right[k - i] */
+ "ldr %[t0], [%[left], %[i]] \n\t" /* t0 = left[i] */
+
+ "umull %[t0], %[t1], %[t0], %[t1] \n\t" /* (t0, t1) = left[i] * right[k - i] */
+
+ "adds %[c0], %[t0] \n\t" /* add low word to c0 */
+ "adcs %[c1], %[t1] \n\t" /* add high word to c1, including carry */
+ "adcs %[c2], #0 \n\t" /* add carry to c2 */
+
+ "adds %[i], #4 \n\t" /* i += 4 */
+ "cmp %[i], %[last_word] \n\t" /* i > (num_words - 1) (times 4)? */
+ "bgt 4f \n\t" /* if so, exit the loop */
+ "cmp %[i], %[k] \n\t" /* i <= k? */
+ "ble 3b \n\t" /* if so, continue looping */
+
+ "4: \n\t" /* end inner loop */
+
+ "str %[c0], [%[result], %[k]] \n\t" /* result[k] = c0 */
+ "mov %[c0], %[c1] \n\t" /* c0 = c1 */
+ "mov %[c1], %[c2] \n\t" /* c1 = c2 */
+ "movs %[c2], #0 \n\t" /* c2 = 0 */
+ "adds %[k], #4 \n\t" /* k += 4 */
+ "cmp %[k], %[last_word] \n\t" /* k <= (num_words - 1) (times 4) ? */
+ "ble 1b \n\t" /* if so, loop back, start with i = 0 */
+ "cmp %[k], %[last_word], lsl #1 \n\t" /* k <= (num_words * 2 - 2) (times 4) ? */
+ "ble 2b \n\t" /* if so, loop back, start with i = (k + 1) - num_words */
+ /* end outer loop */
+
+ "str %[c0], [%[result], %[k]] \n\t" /* result[num_words * 2 - 1] = c0 */
+ RESUME_SYNTAX
+ : [c0] "+r" (c0), [c1] "+r" (c1), [c2] "+r" (c2),
+ [k] "+r" (k), [i] "=&r" (i), [t0] "=&r" (t0), [t1] "=&r" (t1)
+ : [result] "r" (result), [left] "r" (left), [right] "r" (right),
+ [last_word] "r" ((num_words - 1) * 4)
+ : "cc", "memory"
+ );
+
+#else /* Thumb-1 */
+
+ register uint32_t *r0 __asm__("r0") = result;
+ register const uint32_t *r1 __asm__("r1") = left;
+ register const uint32_t *r2 __asm__("r2") = right;
+
+ __asm__ volatile (
+ ".syntax unified \n\t"
+ "movs r3, #0 \n\t" /* c0 = 0 */
+ "movs r4, #0 \n\t" /* c1 = 0 */
+ "movs r5, #0 \n\t" /* c2 = 0 */
+ "movs r6, #0 \n\t" /* k = 0 */
+
+ "push {r0} \n\t" /* keep result on the stack */
+
+ "1: \n\t" /* outer loop (k < num_words) */
+ "movs r7, #0 \n\t" /* r7 = i = 0 */
+ "b 3f \n\t"
+
+ "2: \n\t" /* outer loop (k >= num_words) */
+ "movs r7, r6 \n\t" /* r7 = k */
+ "subs r7, %[last_word] \n\t" /* r7 = i = k - (num_words - 1) (times 4) */
+
+ "3: \n\t" /* inner loop */
+ "push {r3, r4, r5, r6} \n\t" /* push things, r3 (c0) is at the top of stack. */
+ "subs r0, r6, r7 \n\t" /* r0 = k - i */
+
+ "ldr r4, [r2, r0] \n\t" /* r4 = right[k - i] */
+ "ldr r0, [r1, r7] \n\t" /* r0 = left[i] */
+
+ "lsrs r3, r0, #16 \n\t" /* r3 = a1 */
+ "uxth r0, r0 \n\t" /* r0 = a0 */
+
+ "lsrs r5, r4, #16 \n\t" /* r5 = b1 */
+ "uxth r4, r4 \n\t" /* r4 = b0 */
+
+ "movs r6, r3 \n\t" /* r6 = a1 */
+ "muls r6, r5, r6 \n\t" /* r6 = a1 * b1 */
+ "muls r3, r4, r3 \n\t" /* r3 = b0 * a1 */
+ "muls r5, r0, r5 \n\t" /* r5 = a0 * b1 */
+ "muls r0, r4, r0 \n\t" /* r0 = a0 * b0 */
+
+ "movs r4, #0 \n\t" /* r4 = 0 */
+ "adds r3, r5 \n\t" /* r3 = b0 * a1 + a0 * b1 */
+ "adcs r4, r4 \n\t" /* r4 = carry */
+ "lsls r4, #16 \n\t" /* r4 = carry << 16 */
+ "adds r6, r4 \n\t" /* r6 = a1 * b1 + carry */
+
+ "lsls r4, r3, #16 \n\t" /* r4 = (b0 * a1 + a0 * b1) << 16 */
+ "lsrs r3, #16 \n\t" /* r3 = (b0 * a1 + a0 * b1) >> 16 */
+ "adds r0, r4 \n\t" /* r0 = low word = a0 * b0 + ((b0 * a1 + a0 * b1) << 16) */
+ "adcs r6, r3 \n\t" /* r6 = high word = a1 * b1 + carry + ((b0 * a1 + a0 * b1) >> 16) */
+
+ "pop {r3, r4, r5} \n\t" /* r3 = c0, r4 = c1, r5 = c2 */
+ "adds r3, r0 \n\t" /* add low word to c0 */
+ "adcs r4, r6 \n\t" /* add high word to c1, including carry */
+ "movs r0, #0 \n\t" /* r0 = 0 (does not affect carry bit) */
+ "adcs r5, r0 \n\t" /* add carry to c2 */
+
+ "pop {r6} \n\t" /* r6 = k */
+
+ "adds r7, #4 \n\t" /* i += 4 */
+ "cmp r7, %[last_word] \n\t" /* i > (num_words - 1) (times 4)? */
+ "bgt 4f \n\t" /* if so, exit the loop */
+ "cmp r7, r6 \n\t" /* i <= k? */
+ "ble 3b \n\t" /* if so, continue looping */
+
+ "4: \n\t" /* end inner loop */
+
+ "ldr r0, [sp, #0] \n\t" /* r0 = result */
+
+ "str r3, [r0, r6] \n\t" /* result[k] = c0 */
+ "mov r3, r4 \n\t" /* c0 = c1 */
+ "mov r4, r5 \n\t" /* c1 = c2 */
+ "movs r5, #0 \n\t" /* c2 = 0 */
+ "adds r6, #4 \n\t" /* k += 4 */
+ "cmp r6, %[last_word] \n\t" /* k <= (num_words - 1) (times 4) ? */
+ "ble 1b \n\t" /* if so, loop back, start with i = 0 */
+ "cmp r6, %[lw2] \n\t" /* k <= (num_words * 2 - 2) (times 4) ? */
+ "ble 2b \n\t" /* if so, loop back, start with i = (k + 1) - num_words */
+ /* end outer loop */
+
+ "str r3, [r0, r6] \n\t" /* result[num_words * 2 - 1] = c0 */
+ "pop {r0} \n\t" /* pop result off the stack */
+
+ ".syntax divided \n\t"
+ :
+ : [r0] "l" (r0), [r1] "l" (r1), [r2] "l" (r2),
+ [last_word] "r" ((num_words - 1) * 4), [lw2] "r" ((num_words - 1) * 4 * 2)
+ : "r3", "r4", "r5", "r6", "r7", "cc", "memory"
+ );
+#endif
+}
+#define asm_mult 1
+#endif
+
+#if uECC_SQUARE_FUNC
+#if !asm_square
+static void vli_square(uECC_word_t *result, const uECC_word_t *left, wordcount_t num_words) {
+#if (uECC_PLATFORM != uECC_arm_thumb)
+ uint32_t c0 = 0;
+ uint32_t c1 = 0;
+ uint32_t c2 = 0;
+ uint32_t k = 0;
+ uint32_t i, tt;
+ uint32_t t0, t1;
+
+ __asm__ volatile (
+ ".syntax unified \n\t"
+
+ "1: \n\t" /* outer loop (k < num_words) */
+ "movs %[i], #0 \n\t" /* i = 0 */
+ "b 3f \n\t"
+
+ "2: \n\t" /* outer loop (k >= num_words) */
+ "movs %[i], %[k] \n\t" /* i = k */
+ "subs %[i], %[last_word] \n\t" /* i = k - (num_words - 1) (times 4) */
+
+ "3: \n\t" /* inner loop */
+ "subs %[tt], %[k], %[i] \n\t" /* tt = k-i */
+
+ "ldr %[t1], [%[left], %[tt]] \n\t" /* t1 = left[k - i] */
+ "ldr %[t0], [%[left], %[i]] \n\t" /* t0 = left[i] */
+
+ "umull %[t0], %[t1], %[t0], %[t1] \n\t" /* (t0, t1) = left[i] * right[k - i] */
+
+ "cmp %[i], %[tt] \n\t" /* (i < k - i) ? */
+ "bge 4f \n\t" /* if i >= k - i, skip */
+ "lsls %[t1], #1 \n\t" /* high word << 1 */
+ "adc %[c2], #0 \n\t" /* add carry bit to c2 */
+ "lsls %[t0], #1 \n\t" /* low word << 1 */
+ "adc %[t1], #0 \n\t" /* add carry bit to high word */
+
+ "4: \n\t"
+
+ "adds %[c0], %[t0] \n\t" /* add low word to c0 */
+ "adcs %[c1], %[t1] \n\t" /* add high word to c1, including carry */
+ "adc %[c2], #0 \n\t" /* add carry to c2 */
+
+ "adds %[i], #4 \n\t" /* i += 4 */
+ "cmp %[i], %[k] \n\t" /* i >= k? */
+ "bge 5f \n\t" /* if so, exit the loop */
+ "subs %[tt], %[k], %[i] \n\t" /* tt = k - i */
+ "cmp %[i], %[tt] \n\t" /* i <= k - i? */
+ "ble 3b \n\t" /* if so, continue looping */
+
+ "5: \n\t" /* end inner loop */
+
+ "str %[c0], [%[result], %[k]] \n\t" /* result[k] = c0 */
+ "mov %[c0], %[c1] \n\t" /* c0 = c1 */
+ "mov %[c1], %[c2] \n\t" /* c1 = c2 */
+ "movs %[c2], #0 \n\t" /* c2 = 0 */
+ "adds %[k], #4 \n\t" /* k += 4 */
+ "cmp %[k], %[last_word] \n\t" /* k <= (num_words - 1) (times 4) ? */
+ "ble 1b \n\t" /* if so, loop back, start with i = 0 */
+ "cmp %[k], %[last_word], lsl #1 \n\t" /* k <= (num_words * 2 - 2) (times 4) ? */
+ "ble 2b \n\t" /* if so, loop back, start with i = (k + 1) - num_words */
+ /* end outer loop */
+
+ "str %[c0], [%[result], %[k]] \n\t" /* result[num_words * 2 - 1] = c0 */
+ RESUME_SYNTAX
+ : [c0] "+r" (c0), [c1] "+r" (c1), [c2] "+r" (c2),
+ [k] "+r" (k), [i] "=&r" (i), [tt] "=&r" (tt), [t0] "=&r" (t0), [t1] "=&r" (t1)
+ : [result] "r" (result), [left] "r" (left), [last_word] "r" ((num_words - 1) * 4)
+ : "cc", "memory"
+ );
+
+#else
+
+ register uint32_t *r0 __asm__("r0") = result;
+ register const uint32_t *r1 __asm__("r1") = left;
+
+ __asm__ volatile (
+ ".syntax unified \n\t"
+ "movs r2, #0 \n\t" /* c0 = 0 */
+ "movs r3, #0 \n\t" /* c1 = 0 */
+ "movs r4, #0 \n\t" /* c2 = 0 */
+ "movs r5, #0 \n\t" /* k = 0 */
+
+ "push {r0} \n\t" /* keep result on the stack */
+
+ "1: \n\t" /* outer loop (k < num_words) */
+ "movs r6, #0 \n\t" /* r6 = i = 0 */
+ "b 3f \n\t"
+
+ "2: \n\t" /* outer loop (k >= num_words) */
+ "movs r6, r5 \n\t" /* r6 = k */
+ "subs r6, %[last_word] \n\t" /* r6 = i = k - (num_words - 1) (times 4) */
+
+ "3: \n\t" /* inner loop */
+ "push {r2, r3, r4, r5} \n\t" /* push things, r2 (c0) is at the top of stack. */
+ "subs r7, r5, r6 \n\t" /* r7 = k - i */
+
+ "ldr r3, [r1, r7] \n\t" /* r3 = left[k - i] */
+ "ldr r0, [r1, r6] \n\t" /* r0 = left[i] */
+
+ "lsrs r2, r0, #16 \n\t" /* r2 = a1 */
+ "uxth r0, r0 \n\t" /* r0 = a0 */
+
+ "lsrs r4, r3, #16 \n\t" /* r4 = b1 */
+ "uxth r3, r3 \n\t" /* r3 = b0 */
+
+ "movs r5, r2 \n\t" /* r5 = a1 */
+ "muls r5, r4, r5 \n\t" /* r5 = a1 * b1 */
+ "muls r2, r3, r2 \n\t" /* r2 = b0 * a1 */
+ "muls r4, r0, r4 \n\t" /* r4 = a0 * b1 */
+ "muls r0, r3, r0 \n\t" /* r0 = a0 * b0 */
+
+ "movs r3, #0 \n\t" /* r3 = 0 */
+ "adds r2, r4 \n\t" /* r2 = b0 * a1 + a0 * b1 */
+ "adcs r3, r3 \n\t" /* r3 = carry */
+ "lsls r3, #16 \n\t" /* r3 = carry << 16 */
+ "adds r5, r3 \n\t" /* r5 = a1 * b1 + carry */
+
+ "lsls r3, r2, #16 \n\t" /* r3 = (b0 * a1 + a0 * b1) << 16 */
+ "lsrs r2, #16 \n\t" /* r2 = (b0 * a1 + a0 * b1) >> 16 */
+ "adds r0, r3 \n\t" /* r0 = low word = a0 * b0 + ((b0 * a1 + a0 * b1) << 16) */
+ "adcs r5, r2 \n\t" /* r5 = high word = a1 * b1 + carry + ((b0 * a1 + a0 * b1) >> 16) */
+
+ "movs r3, #0 \n\t" /* r3 = 0 */
+ "cmp r6, r7 \n\t" /* (i < k - i) ? */
+ "mov r7, r3 \n\t" /* r7 = 0 (does not affect condition)*/
+ "bge 4f \n\t" /* if i >= k - i, skip */
+ "lsls r5, #1 \n\t" /* high word << 1 */
+ "adcs r7, r3 \n\t" /* r7 = carry bit for c2 */
+ "lsls r0, #1 \n\t" /* low word << 1 */
+ "adcs r5, r3 \n\t" /* add carry from shift to high word */
+
+ "4: \n\t"
+ "pop {r2, r3, r4} \n\t" /* r2 = c0, r3 = c1, r4 = c2 */
+ "adds r2, r0 \n\t" /* add low word to c0 */
+ "adcs r3, r5 \n\t" /* add high word to c1, including carry */
+ "movs r0, #0 \n\t" /* r0 = 0 (does not affect carry bit) */
+ "adcs r4, r0 \n\t" /* add carry to c2 */
+ "adds r4, r7 \n\t" /* add carry from doubling (if any) */
+
+ "pop {r5} \n\t" /* r5 = k */
+
+ "adds r6, #4 \n\t" /* i += 4 */
+ "cmp r6, r5 \n\t" /* i >= k? */
+ "bge 5f \n\t" /* if so, exit the loop */
+ "subs r7, r5, r6 \n\t" /* r7 = k - i */
+ "cmp r6, r7 \n\t" /* i <= k - i? */
+ "ble 3b \n\t" /* if so, continue looping */
+
+ "5: \n\t" /* end inner loop */
+
+ "ldr r0, [sp, #0] \n\t" /* r0 = result */
+
+ "str r2, [r0, r5] \n\t" /* result[k] = c0 */
+ "mov r2, r3 \n\t" /* c0 = c1 */
+ "mov r3, r4 \n\t" /* c1 = c2 */
+ "movs r4, #0 \n\t" /* c2 = 0 */
+ "adds r5, #4 \n\t" /* k += 4 */
+ "cmp r5, %[last_word] \n\t" /* k <= (num_words - 1) (times 4) ? */
+ "ble 1b \n\t" /* if so, loop back, start with i = 0 */
+ "cmp r5, %[lw2] \n\t" /* k <= (num_words * 2 - 2) (times 4) ? */
+ "ble 2b \n\t" /* if so, loop back, start with i = (k + 1) - num_words */
+ /* end outer loop */
+
+ "str r2, [r0, r5] \n\t" /* result[num_words * 2 - 1] = c0 */
+ "pop {r0} \n\t" /* pop result off the stack */
+
+ ".syntax divided \n\t"
+ : [r0] "+l" (r0), [r1] "+l" (r1)
+ : [last_word] "r" ((num_words - 1) * 4), [lw2] "r" ((num_words - 1) * 4 * 2)
+ : "r2", "r3", "r4", "r5", "r6", "r7", "cc", "memory"
+ );
+#endif
+}
+#define asm_square 1
+#endif
+#endif /* uECC_SQUARE_FUNC */
+
+#endif /* _UECC_ASM_ARM_H_ */
diff --git a/asm_arm_small.inc b/asm_arm_small.inc
deleted file mode 100644
index 8e285fa..0000000
--- a/asm_arm_small.inc
+++ /dev/null
@@ -1,422 +0,0 @@
-/* Copyright 2015, Kenneth MacKay. Licensed under the BSD 2-clause license. */
-
-#ifndef _UECC_ASM_ARM_SMALL_H_
-#define _UECC_ASM_ARM_SMALL_H_
-
-#if (uECC_PLATFORM == uECC_arm_thumb)
- #define REG_RW "+l"
- #define REG_WRITE "=l"
-#else
- #define REG_RW "+r"
- #define REG_WRITE "=r"
-#endif
-
-#if (uECC_PLATFORM == uECC_arm_thumb2)
- #define RESUME_SYNTAX
-#else
- #define RESUME_SYNTAX ".syntax divided \n\t"
-#endif
-
-#if !asm_add
-static uECC_word_t vli_add(uECC_word_t *result,
- const uECC_word_t *left,
- const uECC_word_t *right,
- wordcount_t num_words) {
- uint32_t carry = 0;
- uint32_t left_word;
- uint32_t right_word;
-
- __asm__ volatile (
- ".syntax unified \n\t"
- "1: \n\t"
- "ldmia %[lptr]!, {%[left]} \n\t" /* Load left word. */
- "ldmia %[rptr]!, {%[right]} \n\t" /* Load right word. */
- "lsrs %[carry], #1 \n\t" /* Set up carry flag (carry = 0 after this). */
- "adcs %[left], %[right] \n\t" /* Add with carry. */
- "adcs %[carry], %[carry] \n\t" /* Store carry bit. */
- "stmia %[dptr]!, {%[left]} \n\t" /* Store result word. */
- "subs %[ctr], #1 \n\t" /* Decrement counter. */
- "bne 1b \n\t" /* Loop until counter == 0. */
- RESUME_SYNTAX
- : [dptr] REG_RW (result), [lptr] REG_RW (left), [rptr] REG_RW (right),
- [ctr] REG_RW (num_words), [carry] REG_RW (carry),
- [left] REG_WRITE (left_word), [right] REG_WRITE (right_word)
- :
- : "cc", "memory"
- );
- return carry;
-}
-#define asm_add 1
-#endif
-
-#if !asm_sub
-static uECC_word_t vli_sub(uECC_word_t *result,
- const uECC_word_t *left,
- const uECC_word_t *right,
- wordcount_t num_words) {
- uint32_t carry = 1; /* carry = 1 initially (means don't borrow) */
- uint32_t left_word;
- uint32_t right_word;
-
- __asm__ volatile (
- ".syntax unified \n\t"
- "1: \n\t"
- "ldmia %[lptr]!, {%[left]} \n\t" /* Load left word. */
- "ldmia %[rptr]!, {%[right]} \n\t" /* Load right word. */
- "lsrs %[carry], #1 \n\t" /* Set up carry flag (carry = 0 after this). */
- "sbcs %[left], %[right] \n\t" /* Subtract with borrow. */
- "adcs %[carry], %[carry] \n\t" /* Store carry bit. */
- "stmia %[dptr]!, {%[left]} \n\t" /* Store result word. */
- "subs %[ctr], #1 \n\t" /* Decrement counter. */
- "bne 1b \n\t" /* Loop until counter == 0. */
- RESUME_SYNTAX
- : [dptr] REG_RW (result), [lptr] REG_RW (left), [rptr] REG_RW (right),
- [ctr] REG_RW (num_words), [carry] REG_RW (carry),
- [left] REG_WRITE (left_word), [right] REG_WRITE (right_word)
- :
- : "cc", "memory"
- );
- return !carry;
-}
-#define asm_sub 1
-#endif
-
-#if !asm_mult
-static void vli_mult(uECC_word_t *result,
- const uECC_word_t *left,
- const uECC_word_t *right,
- wordcount_t num_words) {
-#if (uECC_PLATFORM != uECC_arm_thumb)
- uint32_t c0 = 0;
- uint32_t c1 = 0;
- uint32_t c2 = 0;
- uint32_t k = 0;
- uint32_t i;
- uint32_t t0, t1;
-
- __asm__ volatile (
- ".syntax unified \n\t"
-
- "1: \n\t" /* outer loop (k < num_words) */
- "movs %[i], #0 \n\t" /* i = 0 */
- "b 3f \n\t"
-
- "2: \n\t" /* outer loop (k >= num_words) */
- "movs %[i], %[k] \n\t" /* i = k */
- "subs %[i], %[last_word] \n\t" /* i = k - (num_words - 1) (times 4) */
-
- "3: \n\t" /* inner loop */
- "subs %[t0], %[k], %[i] \n\t" /* t0 = k-i */
-
- "ldr %[t1], [%[right], %[t0]] \n\t" /* t1 = right[k - i] */
- "ldr %[t0], [%[left], %[i]] \n\t" /* t0 = left[i] */
-
- "umull %[t0], %[t1], %[t0], %[t1] \n\t" /* (t0, t1) = left[i] * right[k - i] */
-
- "adds %[c0], %[t0] \n\t" /* add low word to c0 */
- "adcs %[c1], %[t1] \n\t" /* add high word to c1, including carry */
- "adcs %[c2], #0 \n\t" /* add carry to c2 */
-
- "adds %[i], #4 \n\t" /* i += 4 */
- "cmp %[i], %[last_word] \n\t" /* i > (num_words - 1) (times 4)? */
- "bgt 4f \n\t" /* if so, exit the loop */
- "cmp %[i], %[k] \n\t" /* i <= k? */
- "ble 3b \n\t" /* if so, continue looping */
-
- "4: \n\t" /* end inner loop */
-
- "str %[c0], [%[result], %[k]] \n\t" /* result[k] = c0 */
- "mov %[c0], %[c1] \n\t" /* c0 = c1 */
- "mov %[c1], %[c2] \n\t" /* c1 = c2 */
- "movs %[c2], #0 \n\t" /* c2 = 0 */
- "adds %[k], #4 \n\t" /* k += 4 */
- "cmp %[k], %[last_word] \n\t" /* k <= (num_words - 1) (times 4) ? */
- "ble 1b \n\t" /* if so, loop back, start with i = 0 */
- "cmp %[k], %[last_word], lsl #1 \n\t" /* k <= (num_words * 2 - 2) (times 4) ? */
- "ble 2b \n\t" /* if so, loop back, start with i = (k + 1) - num_words */
- /* end outer loop */
-
- "str %[c0], [%[result], %[k]] \n\t" /* result[num_words * 2 - 1] = c0 */
- RESUME_SYNTAX
- : [c0] "+r" (c0), [c1] "+r" (c1), [c2] "+r" (c2),
- [k] "+r" (k), [i] "=&r" (i), [t0] "=&r" (t0), [t1] "=&r" (t1)
- : [result] "r" (result), [left] "r" (left), [right] "r" (right),
- [last_word] "r" ((num_words - 1) * 4)
- : "cc", "memory"
- );
-
-#else /* Thumb-1 */
-
- register uint32_t *r0 __asm__("r0") = result;
- register const uint32_t *r1 __asm__("r1") = left;
- register const uint32_t *r2 __asm__("r2") = right;
-
- __asm__ volatile (
- ".syntax unified \n\t"
- "movs r3, #0 \n\t" /* c0 = 0 */
- "movs r4, #0 \n\t" /* c1 = 0 */
- "movs r5, #0 \n\t" /* c2 = 0 */
- "movs r6, #0 \n\t" /* k = 0 */
-
- "push {r0} \n\t" /* keep result on the stack */
-
- "1: \n\t" /* outer loop (k < num_words) */
- "movs r7, #0 \n\t" /* r7 = i = 0 */
- "b 3f \n\t"
-
- "2: \n\t" /* outer loop (k >= num_words) */
- "movs r7, r6 \n\t" /* r7 = k */
- "subs r7, %[last_word] \n\t" /* r7 = i = k - (num_words - 1) (times 4) */
-
- "3: \n\t" /* inner loop */
- "push {r3, r4, r5, r6} \n\t" /* push things, r3 (c0) is at the top of stack. */
- "subs r0, r6, r7 \n\t" /* r0 = k - i */
-
- "ldr r4, [r2, r0] \n\t" /* r4 = right[k - i] */
- "ldr r0, [r1, r7] \n\t" /* r0 = left[i] */
-
- "lsrs r3, r0, #16 \n\t" /* r3 = a1 */
- "uxth r0, r0 \n\t" /* r0 = a0 */
-
- "lsrs r5, r4, #16 \n\t" /* r5 = b1 */
- "uxth r4, r4 \n\t" /* r4 = b0 */
-
- "movs r6, r3 \n\t" /* r6 = a1 */
- "muls r6, r5, r6 \n\t" /* r6 = a1 * b1 */
- "muls r3, r4, r3 \n\t" /* r3 = b0 * a1 */
- "muls r5, r0, r5 \n\t" /* r5 = a0 * b1 */
- "muls r0, r4, r0 \n\t" /* r0 = a0 * b0 */
-
- "movs r4, #0 \n\t" /* r4 = 0 */
- "adds r3, r5 \n\t" /* r3 = b0 * a1 + a0 * b1 */
- "adcs r4, r4 \n\t" /* r4 = carry */
- "lsls r4, #16 \n\t" /* r4 = carry << 16 */
- "adds r6, r4 \n\t" /* r6 = a1 * b1 + carry */
-
- "lsls r4, r3, #16 \n\t" /* r4 = (b0 * a1 + a0 * b1) << 16 */
- "lsrs r3, #16 \n\t" /* r3 = (b0 * a1 + a0 * b1) >> 16 */
- "adds r0, r4 \n\t" /* r0 = low word = a0 * b0 + ((b0 * a1 + a0 * b1) << 16) */
- "adcs r6, r3 \n\t" /* r6 = high word = a1 * b1 + carry + ((b0 * a1 + a0 * b1) >> 16) */
-
- "pop {r3, r4, r5} \n\t" /* r3 = c0, r4 = c1, r5 = c2 */
- "adds r3, r0 \n\t" /* add low word to c0 */
- "adcs r4, r6 \n\t" /* add high word to c1, including carry */
- "movs r0, #0 \n\t" /* r0 = 0 (does not affect carry bit) */
- "adcs r5, r0 \n\t" /* add carry to c2 */
-
- "pop {r6} \n\t" /* r6 = k */
-
- "adds r7, #4 \n\t" /* i += 4 */
- "cmp r7, %[last_word] \n\t" /* i > (num_words - 1) (times 4)? */
- "bgt 4f \n\t" /* if so, exit the loop */
- "cmp r7, r6 \n\t" /* i <= k? */
- "ble 3b \n\t" /* if so, continue looping */
-
- "4: \n\t" /* end inner loop */
-
- "ldr r0, [sp, #0] \n\t" /* r0 = result */
-
- "str r3, [r0, r6] \n\t" /* result[k] = c0 */
- "mov r3, r4 \n\t" /* c0 = c1 */
- "mov r4, r5 \n\t" /* c1 = c2 */
- "movs r5, #0 \n\t" /* c2 = 0 */
- "adds r6, #4 \n\t" /* k += 4 */
- "cmp r6, %[last_word] \n\t" /* k <= (num_words - 1) (times 4) ? */
- "ble 1b \n\t" /* if so, loop back, start with i = 0 */
- "cmp r6, %[lw2] \n\t" /* k <= (num_words * 2 - 2) (times 4) ? */
- "ble 2b \n\t" /* if so, loop back, start with i = (k + 1) - num_words */
- /* end outer loop */
-
- "str r3, [r0, r6] \n\t" /* result[num_words * 2 - 1] = c0 */
- "pop {r0} \n\t" /* pop result off the stack */
-
- ".syntax divided \n\t"
- :
- : [r0] "l" (r0), [r1] "l" (r1), [r2] "l" (r2),
- [last_word] "r" ((num_words - 1) * 4), [lw2] "r" ((num_words - 1) * 4 * 2)
- : "r3", "r4", "r5", "r6", "r7", "cc", "memory"
- );
-#endif
-}
-#define asm_mult 1
-#endif
-
-#if uECC_SQUARE_FUNC
-#if !asm_square
-static void vli_square(uECC_word_t *result, const uECC_word_t *left, wordcount_t num_words) {
-#if (uECC_PLATFORM != uECC_arm_thumb)
- uint32_t c0 = 0;
- uint32_t c1 = 0;
- uint32_t c2 = 0;
- uint32_t k = 0;
- uint32_t i, tt;
- uint32_t t0, t1;
-
- __asm__ volatile (
- ".syntax unified \n\t"
-
- "1: \n\t" /* outer loop (k < num_words) */
- "movs %[i], #0 \n\t" /* i = 0 */
- "b 3f \n\t"
-
- "2: \n\t" /* outer loop (k >= num_words) */
- "movs %[i], %[k] \n\t" /* i = k */
- "subs %[i], %[last_word] \n\t" /* i = k - (num_words - 1) (times 4) */
-
- "3: \n\t" /* inner loop */
- "subs %[tt], %[k], %[i] \n\t" /* tt = k-i */
-
- "ldr %[t1], [%[left], %[tt]] \n\t" /* t1 = left[k - i] */
- "ldr %[t0], [%[left], %[i]] \n\t" /* t0 = left[i] */
-
- "umull %[t0], %[t1], %[t0], %[t1] \n\t" /* (t0, t1) = left[i] * right[k - i] */
-
- "cmp %[i], %[tt] \n\t" /* (i < k - i) ? */
- "bge 4f \n\t" /* if i >= k - i, skip */
- "lsls %[t1], #1 \n\t" /* high word << 1 */
- "adc %[c2], #0 \n\t" /* add carry bit to c2 */
- "lsls %[t0], #1 \n\t" /* low word << 1 */
- "adc %[t1], #0 \n\t" /* add carry bit to high word */
-
- "4: \n\t"
-
- "adds %[c0], %[t0] \n\t" /* add low word to c0 */
- "adcs %[c1], %[t1] \n\t" /* add high word to c1, including carry */
- "adc %[c2], #0 \n\t" /* add carry to c2 */
-
- "adds %[i], #4 \n\t" /* i += 4 */
- "cmp %[i], %[k] \n\t" /* i >= k? */
- "bge 5f \n\t" /* if so, exit the loop */
- "subs %[tt], %[k], %[i] \n\t" /* tt = k - i */
- "cmp %[i], %[tt] \n\t" /* i <= k - i? */
- "ble 3b \n\t" /* if so, continue looping */
-
- "5: \n\t" /* end inner loop */
-
- "str %[c0], [%[result], %[k]] \n\t" /* result[k] = c0 */
- "mov %[c0], %[c1] \n\t" /* c0 = c1 */
- "mov %[c1], %[c2] \n\t" /* c1 = c2 */
- "movs %[c2], #0 \n\t" /* c2 = 0 */
- "adds %[k], #4 \n\t" /* k += 4 */
- "cmp %[k], %[last_word] \n\t" /* k <= (num_words - 1) (times 4) ? */
- "ble 1b \n\t" /* if so, loop back, start with i = 0 */
- "cmp %[k], %[last_word], lsl #1 \n\t" /* k <= (num_words * 2 - 2) (times 4) ? */
- "ble 2b \n\t" /* if so, loop back, start with i = (k + 1) - num_words */
- /* end outer loop */
-
- "str %[c0], [%[result], %[k]] \n\t" /* result[num_words * 2 - 1] = c0 */
- RESUME_SYNTAX
- : [c0] "+r" (c0), [c1] "+r" (c1), [c2] "+r" (c2),
- [k] "+r" (k), [i] "=&r" (i), [tt] "=&r" (tt), [t0] "=&r" (t0), [t1] "=&r" (t1)
- : [result] "r" (result), [left] "r" (left), [last_word] "r" ((num_words - 1) * 4)
- : "cc", "memory"
- );
-
-#else
-
- register uint32_t *r0 __asm__("r0") = result;
- register const uint32_t *r1 __asm__("r1") = left;
-
- __asm__ volatile (
- ".syntax unified \n\t"
- "movs r2, #0 \n\t" /* c0 = 0 */
- "movs r3, #0 \n\t" /* c1 = 0 */
- "movs r4, #0 \n\t" /* c2 = 0 */
- "movs r5, #0 \n\t" /* k = 0 */
-
- "push {r0} \n\t" /* keep result on the stack */
-
- "1: \n\t" /* outer loop (k < num_words) */
- "movs r6, #0 \n\t" /* r6 = i = 0 */
- "b 3f \n\t"
-
- "2: \n\t" /* outer loop (k >= num_words) */
- "movs r6, r5 \n\t" /* r6 = k */
- "subs r6, %[last_word] \n\t" /* r6 = i = k - (num_words - 1) (times 4) */
-
- "3: \n\t" /* inner loop */
- "push {r2, r3, r4, r5} \n\t" /* push things, r2 (c0) is at the top of stack. */
- "subs r7, r5, r6 \n\t" /* r7 = k - i */
-
- "ldr r3, [r1, r7] \n\t" /* r3 = left[k - i] */
- "ldr r0, [r1, r6] \n\t" /* r0 = left[i] */
-
- "lsrs r2, r0, #16 \n\t" /* r2 = a1 */
- "uxth r0, r0 \n\t" /* r0 = a0 */
-
- "lsrs r4, r3, #16 \n\t" /* r4 = b1 */
- "uxth r3, r3 \n\t" /* r3 = b0 */
-
- "movs r5, r2 \n\t" /* r5 = a1 */
- "muls r5, r4, r5 \n\t" /* r5 = a1 * b1 */
- "muls r2, r3, r2 \n\t" /* r2 = b0 * a1 */
- "muls r4, r0, r4 \n\t" /* r4 = a0 * b1 */
- "muls r0, r3, r0 \n\t" /* r0 = a0 * b0 */
-
- "movs r3, #0 \n\t" /* r3 = 0 */
- "adds r2, r4 \n\t" /* r2 = b0 * a1 + a0 * b1 */
- "adcs r3, r3 \n\t" /* r3 = carry */
- "lsls r3, #16 \n\t" /* r3 = carry << 16 */
- "adds r5, r3 \n\t" /* r5 = a1 * b1 + carry */
-
- "lsls r3, r2, #16 \n\t" /* r3 = (b0 * a1 + a0 * b1) << 16 */
- "lsrs r2, #16 \n\t" /* r2 = (b0 * a1 + a0 * b1) >> 16 */
- "adds r0, r3 \n\t" /* r0 = low word = a0 * b0 + ((b0 * a1 + a0 * b1) << 16) */
- "adcs r5, r2 \n\t" /* r5 = high word = a1 * b1 + carry + ((b0 * a1 + a0 * b1) >> 16) */
-
- "movs r3, #0 \n\t" /* r3 = 0 */
- "cmp r6, r7 \n\t" /* (i < k - i) ? */
- "mov r7, r3 \n\t" /* r7 = 0 (does not affect condition)*/
- "bge 4f \n\t" /* if i >= k - i, skip */
- "lsls r5, #1 \n\t" /* high word << 1 */
- "adcs r7, r3 \n\t" /* r7 = carry bit for c2 */
- "lsls r0, #1 \n\t" /* low word << 1 */
- "adcs r5, r3 \n\t" /* add carry from shift to high word */
-
- "4: \n\t"
- "pop {r2, r3, r4} \n\t" /* r2 = c0, r3 = c1, r4 = c2 */
- "adds r2, r0 \n\t" /* add low word to c0 */
- "adcs r3, r5 \n\t" /* add high word to c1, including carry */
- "movs r0, #0 \n\t" /* r0 = 0 (does not affect carry bit) */
- "adcs r4, r0 \n\t" /* add carry to c2 */
- "adds r4, r7 \n\t" /* add carry from doubling (if any) */
-
- "pop {r5} \n\t" /* r5 = k */
-
- "adds r6, #4 \n\t" /* i += 4 */
- "cmp r6, r5 \n\t" /* i >= k? */
- "bge 5f \n\t" /* if so, exit the loop */
- "subs r7, r5, r6 \n\t" /* r7 = k - i */
- "cmp r6, r7 \n\t" /* i <= k - i? */
- "ble 3b \n\t" /* if so, continue looping */
-
- "5: \n\t" /* end inner loop */
-
- "ldr r0, [sp, #0] \n\t" /* r0 = result */
-
- "str r2, [r0, r5] \n\t" /* result[k] = c0 */
- "mov r2, r3 \n\t" /* c0 = c1 */
- "mov r3, r4 \n\t" /* c1 = c2 */
- "movs r4, #0 \n\t" /* c2 = 0 */
- "adds r5, #4 \n\t" /* k += 4 */
- "cmp r5, %[last_word] \n\t" /* k <= (num_words - 1) (times 4) ? */
- "ble 1b \n\t" /* if so, loop back, start with i = 0 */
- "cmp r5, %[lw2] \n\t" /* k <= (num_words * 2 - 2) (times 4) ? */
- "ble 2b \n\t" /* if so, loop back, start with i = (k + 1) - num_words */
- /* end outer loop */
-
- "str r2, [r0, r5] \n\t" /* result[num_words * 2 - 1] = c0 */
- "pop {r0} \n\t" /* pop result off the stack */
-
- ".syntax divided \n\t"
- : [r0] "+l" (r0), [r1] "+l" (r1)
- : [last_word] "r" ((num_words - 1) * 4), [lw2] "r" ((num_words - 1) * 4 * 2)
- : "r2", "r3", "r4", "r5", "r6", "r7", "cc", "memory"
- );
-#endif
-}
-#define asm_square 1
-#endif
-#endif /* uECC_SQUARE_FUNC */
-
-#endif /* _UECC_ASM_ARM_SMALL_H_ */
diff --git a/uECC.c b/uECC.c
index 5b29b6f..8d8ad9f 100644
--- a/uECC.c
+++ b/uECC.c
@@ -49,19 +49,9 @@
#endif
};
-#if (uECC_OPTIMIZATION_LEVEL > 0) && (uECC_OPTIMIZATION_LEVEL % 2 == 0)
- #define uECC_SQUARE_FUNC 1
-#else
- #define uECC_SQUARE_FUNC 0
-#endif
-
#if (uECC_PLATFORM == uECC_arm || uECC_PLATFORM == uECC_arm_thumb || \
uECC_PLATFORM == uECC_arm_thumb2)
- #if (uECC_OPTIMIZATION_LEVEL > 2)
- #include "asm_arm_mult_square.inc"
- #include "asm_arm_fast.inc"
- #endif
- #include "asm_arm_small.inc"
+ #include "asm_arm.inc"
#endif
#if default_RNG_defined
diff --git a/uECC.h b/uECC.h
index 0c4b092..20d26d8 100644
--- a/uECC.h
+++ b/uECC.h
@@ -23,11 +23,15 @@
/* Optimization level; trade speed for code size.
Larger values produce code that is faster but larger.
- Currently supported values are 0 - 4; 0 is unusably slow for most applications. */
+ Currently supported values are 0 - 3; 0 is unusably slow for most applications. */
#ifndef uECC_OPTIMIZATION_LEVEL
#define uECC_OPTIMIZATION_LEVEL 1
#endif
+#ifndef uECC_SQUARE_FUNC
+ #define uECC_SQUARE_FUNC 0
+#endif
+
/* Curve support selection. Set to 0 to remove that curve. */
#ifndef uECC_SUPPORTS_secp160r1
#define uECC_SUPPORTS_secp160r1 1
