Squashed 'src/secp256k1/' changes from b0210a9..bccaf86

bccaf86 Merge pull request #150
2a53a47 Merge pull request #151
5f5a31f Merge pull request #149
3907277 Merge pull request #142
a3e0611 Enable tests in x86 travis builds
45da235 x86 builder
8bb0e93 Merge pull request #155
971fe81 build: fix openssl detection for cross builds
f22d73e Explicitly access %0..%2 as 64-bit so we use the right registers for x32 ABI
e66d4d6 Avoid the stack in assembly and use explicit registers
cf7b2b4 Fix ECDSA message hashes to 32 bytes
056ad31 Really compile with -O3 by default
74ad63a Merge pull request #146
9000458 Merge pull request #145
1f46b00 build: fix __builtin_expect detection for clang
aaba2e0 Merge pull request #136
8a0775c Merge pull request #144
ee1eaa7 Merge pull request #141
c88e2b8 Compile with -O3 by default
6558a26 Make the benchmarks print out stats
000bdf6 Rename bench_verify to bench_recovery
7c6fed2 Add a few more additional tests.
992e03b travis: add clang to the test matrix
b43b79a Merge pull request #143
e06a924 Include time.h header for time().
8d11164 Add some additional tests.
3545627 Merge pull request #118
6a9901e Merge pull request #137
376b28b Merge pull request #128
1728806 Merge pull request #138
a5759c5 Check return value of malloc
39bd94d Variable time normalize
ad86bdf Merge pull request #140
54b768c Another redundant secp256k1_fe_normalize
69dcaab Merge pull request #139
1c29f2e Remove redundant secp256k1_fe_normalize from secp256k1_gej_add_ge_var.
2b9388b Remove unused secp256k1_fe_inv_all
f461b76 Allocate precomputation arrays on the heap
b2c9681 Make {mul,sqr}_inner use the same argument order as {mul,sqr}
6793505 Convert YASM code into inline assembly
f048615 Rewrite field assembly to match the C version
3ce74b1 Tweak precomputed table size for G

git-subtree-dir: src/secp256k1
git-subtree-split: bccaf86caa9c44166e5a66600b742c516e03c3f0

1 files changed, 492 insertions, 3 deletions

diff --git a/src/field_5x52_asm_impl.h b/src/field_5x52_asm_impl.h
index f29605b11b..98cc004bf0 100644
--- a/src/field_5x52_asm_impl.h
+++ b/src/field_5x52_asm_impl.h
@@ -1,13 +1,502 @@
 /**********************************************************************
- * Copyright (c) 2013 Pieter Wuille                                   *
+ * Copyright (c) 2013-2014 Diederik Huys, Pieter Wuille               *
  * Distributed under the MIT software license, see the accompanying   *
  * file COPYING or http://www.opensource.org/licenses/mit-license.php.*
  **********************************************************************/
 
+/**
+ * Changelog:
+ * - March 2013, Diederik Huys:    original version
+ * - November 2014, Pieter Wuille: updated to use Peter Dettman's parallel multiplication algorithm
+ * - December 2014, Pieter Wuille: converted from YASM to GCC inline assembly
+ */
+
 #ifndef _SECP256K1_FIELD_INNER5X52_IMPL_H_
 #define _SECP256K1_FIELD_INNER5X52_IMPL_H_
 
-void __attribute__ ((sysv_abi)) secp256k1_fe_mul_inner(const uint64_t *a, const uint64_t *b, uint64_t *r);
-void __attribute__ ((sysv_abi)) secp256k1_fe_sqr_inner(const uint64_t *a, uint64_t *r);
+SECP256K1_INLINE static void secp256k1_fe_mul_inner(uint64_t *r, const uint64_t *a, const uint64_t * SECP256K1_RESTRICT b) {
+/**
+ * Registers: rdx:rax = multiplication accumulator
+ *            r9:r8   = c
+ *            r15:rcx = d
+ *            r10-r14 = a0-a4
+ *            rbx     = b
+ *            rdi     = r
+ *            rsi     = a / t?
+ */
+  uint64_t tmp1, tmp2, tmp3;
+__asm__ __volatile__(
+    "movq 0(%%rsi),%%r10\n"
+    "movq 8(%%rsi),%%r11\n"
+    "movq 16(%%rsi),%%r12\n"
+    "movq 24(%%rsi),%%r13\n"
+    "movq 32(%%rsi),%%r14\n"
+
+    /* d += a3 * b0 */
+    "movq 0(%%rbx),%%rax\n"
+    "mulq %%r13\n"
+    "movq %%rax,%%rcx\n"
+    "movq %%rdx,%%r15\n"
+    /* d += a2 * b1 */
+    "movq 8(%%rbx),%%rax\n"
+    "mulq %%r12\n"
+    "addq %%rax,%%rcx\n"
+    "adcq %%rdx,%%r15\n"
+    /* d += a1 * b2 */
+    "movq 16(%%rbx),%%rax\n"
+    "mulq %%r11\n"
+    "addq %%rax,%%rcx\n"
+    "adcq %%rdx,%%r15\n"
+    /* d = a0 * b3 */
+    "movq 24(%%rbx),%%rax\n"
+    "mulq %%r10\n"
+    "addq %%rax,%%rcx\n"
+    "adcq %%rdx,%%r15\n"
+    /* c = a4 * b4 */
+    "movq 32(%%rbx),%%rax\n"
+    "mulq %%r14\n"
+    "movq %%rax,%%r8\n"
+    "movq %%rdx,%%r9\n"
+    /* d += (c & M) * R */
+    "movq $0xfffffffffffff,%%rdx\n"
+    "andq %%rdx,%%rax\n"
+    "movq $0x1000003d10,%%rdx\n"
+    "mulq %%rdx\n"
+    "addq %%rax,%%rcx\n"
+    "adcq %%rdx,%%r15\n"
+    /* c >>= 52 (%%r8 only) */
+    "shrdq $52,%%r9,%%r8\n"
+    /* t3 (tmp1) = d & M */
+    "movq %%rcx,%%rsi\n"
+    "movq $0xfffffffffffff,%%rdx\n"
+    "andq %%rdx,%%rsi\n"
+    "movq %%rsi,%q1\n"
+    /* d >>= 52 */
+    "shrdq $52,%%r15,%%rcx\n"
+    "xorq %%r15,%%r15\n"
+    /* d += a4 * b0 */
+    "movq 0(%%rbx),%%rax\n"
+    "mulq %%r14\n"
+    "addq %%rax,%%rcx\n"
+    "adcq %%rdx,%%r15\n"
+    /* d += a3 * b1 */
+    "movq 8(%%rbx),%%rax\n"
+    "mulq %%r13\n"
+    "addq %%rax,%%rcx\n"
+    "adcq %%rdx,%%r15\n"
+    /* d += a2 * b2 */
+    "movq 16(%%rbx),%%rax\n"
+    "mulq %%r12\n"
+    "addq %%rax,%%rcx\n"
+    "adcq %%rdx,%%r15\n"
+    /* d += a1 * b3 */
+    "movq 24(%%rbx),%%rax\n"
+    "mulq %%r11\n"
+    "addq %%rax,%%rcx\n"
+    "adcq %%rdx,%%r15\n"
+    /* d += a0 * b4 */
+    "movq 32(%%rbx),%%rax\n"
+    "mulq %%r10\n"
+    "addq %%rax,%%rcx\n"
+    "adcq %%rdx,%%r15\n"
+    /* d += c * R */
+    "movq %%r8,%%rax\n"
+    "movq $0x1000003d10,%%rdx\n"
+    "mulq %%rdx\n"
+    "addq %%rax,%%rcx\n"
+    "adcq %%rdx,%%r15\n"
+    /* t4 = d & M (%%rsi) */
+    "movq %%rcx,%%rsi\n"
+    "movq $0xfffffffffffff,%%rdx\n"
+    "andq %%rdx,%%rsi\n"
+    /* d >>= 52 */
+    "shrdq $52,%%r15,%%rcx\n"
+    "xorq %%r15,%%r15\n"
+    /* tx = t4 >> 48 (tmp3) */
+    "movq %%rsi,%%rax\n"
+    "shrq $48,%%rax\n"
+    "movq %%rax,%q3\n"
+    /* t4 &= (M >> 4) (tmp2) */
+    "movq $0xffffffffffff,%%rax\n"
+    "andq %%rax,%%rsi\n"
+    "movq %%rsi,%q2\n"
+    /* c = a0 * b0 */
+    "movq 0(%%rbx),%%rax\n"
+    "mulq %%r10\n"
+    "movq %%rax,%%r8\n"
+    "movq %%rdx,%%r9\n"
+    /* d += a4 * b1 */
+    "movq 8(%%rbx),%%rax\n"
+    "mulq %%r14\n"
+    "addq %%rax,%%rcx\n"
+    "adcq %%rdx,%%r15\n"
+    /* d += a3 * b2 */
+    "movq 16(%%rbx),%%rax\n"
+    "mulq %%r13\n"
+    "addq %%rax,%%rcx\n"
+    "adcq %%rdx,%%r15\n"
+    /* d += a2 * b3 */
+    "movq 24(%%rbx),%%rax\n"
+    "mulq %%r12\n"
+    "addq %%rax,%%rcx\n"
+    "adcq %%rdx,%%r15\n"
+    /* d += a1 * b4 */
+    "movq 32(%%rbx),%%rax\n"
+    "mulq %%r11\n"
+    "addq %%rax,%%rcx\n"
+    "adcq %%rdx,%%r15\n"
+    /* u0 = d & M (%%rsi) */
+    "movq %%rcx,%%rsi\n"
+    "movq $0xfffffffffffff,%%rdx\n"
+    "andq %%rdx,%%rsi\n"
+    /* d >>= 52 */
+    "shrdq $52,%%r15,%%rcx\n"
+    "xorq %%r15,%%r15\n"
+    /* u0 = (u0 << 4) | tx (%%rsi) */
+    "shlq $4,%%rsi\n"
+    "movq %q3,%%rax\n"
+    "orq %%rax,%%rsi\n"
+    /* c += u0 * (R >> 4) */
+    "movq $0x1000003d1,%%rax\n"
+    "mulq %%rsi\n"
+    "addq %%rax,%%r8\n"
+    "adcq %%rdx,%%r9\n"
+    /* r[0] = c & M */
+    "movq %%r8,%%rax\n"
+    "movq $0xfffffffffffff,%%rdx\n"
+    "andq %%rdx,%%rax\n"
+    "movq %%rax,0(%%rdi)\n"
+    /* c >>= 52 */
+    "shrdq $52,%%r9,%%r8\n"
+    "xorq %%r9,%%r9\n"
+    /* c += a1 * b0 */
+    "movq 0(%%rbx),%%rax\n"
+    "mulq %%r11\n"
+    "addq %%rax,%%r8\n"
+    "adcq %%rdx,%%r9\n"
+    /* c += a0 * b1 */
+    "movq 8(%%rbx),%%rax\n"
+    "mulq %%r10\n"
+    "addq %%rax,%%r8\n"
+    "adcq %%rdx,%%r9\n"
+    /* d += a4 * b2 */
+    "movq 16(%%rbx),%%rax\n"
+    "mulq %%r14\n"
+    "addq %%rax,%%rcx\n"
+    "adcq %%rdx,%%r15\n"
+    /* d += a3 * b3 */
+    "movq 24(%%rbx),%%rax\n"
+    "mulq %%r13\n"
+    "addq %%rax,%%rcx\n"
+    "adcq %%rdx,%%r15\n"
+    /* d += a2 * b4 */
+    "movq 32(%%rbx),%%rax\n"
+    "mulq %%r12\n"
+    "addq %%rax,%%rcx\n"
+    "adcq %%rdx,%%r15\n"
+    /* c += (d & M) * R */
+    "movq %%rcx,%%rax\n"
+    "movq $0xfffffffffffff,%%rdx\n"
+    "andq %%rdx,%%rax\n"
+    "movq $0x1000003d10,%%rdx\n"
+    "mulq %%rdx\n"
+    "addq %%rax,%%r8\n"
+    "adcq %%rdx,%%r9\n"
+    /* d >>= 52 */
+    "shrdq $52,%%r15,%%rcx\n"
+    "xorq %%r15,%%r15\n"
+    /* r[1] = c & M */
+    "movq %%r8,%%rax\n"
+    "movq $0xfffffffffffff,%%rdx\n"
+    "andq %%rdx,%%rax\n"
+    "movq %%rax,8(%%rdi)\n"
+    /* c >>= 52 */
+    "shrdq $52,%%r9,%%r8\n"
+    "xorq %%r9,%%r9\n"
+    /* c += a2 * b0 */
+    "movq 0(%%rbx),%%rax\n"
+    "mulq %%r12\n"
+    "addq %%rax,%%r8\n"
+    "adcq %%rdx,%%r9\n"
+    /* c += a1 * b1 */
+    "movq 8(%%rbx),%%rax\n"
+    "mulq %%r11\n"
+    "addq %%rax,%%r8\n"
+    "adcq %%rdx,%%r9\n"
+    /* c += a0 * b2 (last use of %%r10 = a0) */
+    "movq 16(%%rbx),%%rax\n"
+    "mulq %%r10\n"
+    "addq %%rax,%%r8\n"
+    "adcq %%rdx,%%r9\n"
+    /* fetch t3 (%%r10, overwrites a0), t4 (%%rsi) */
+    "movq %q2,%%rsi\n"
+    "movq %q1,%%r10\n"
+    /* d += a4 * b3 */
+    "movq 24(%%rbx),%%rax\n"
+    "mulq %%r14\n"
+    "addq %%rax,%%rcx\n"
+    "adcq %%rdx,%%r15\n"
+    /* d += a3 * b4 */
+    "movq 32(%%rbx),%%rax\n"
+    "mulq %%r13\n"
+    "addq %%rax,%%rcx\n"
+    "adcq %%rdx,%%r15\n"
+    /* c += (d & M) * R */
+    "movq %%rcx,%%rax\n"
+    "movq $0xfffffffffffff,%%rdx\n"
+    "andq %%rdx,%%rax\n"
+    "movq $0x1000003d10,%%rdx\n"
+    "mulq %%rdx\n"
+    "addq %%rax,%%r8\n"
+    "adcq %%rdx,%%r9\n"
+    /* d >>= 52 (%%rcx only) */
+    "shrdq $52,%%r15,%%rcx\n"
+    /* r[2] = c & M */
+    "movq %%r8,%%rax\n"
+    "movq $0xfffffffffffff,%%rdx\n"
+    "andq %%rdx,%%rax\n"
+    "movq %%rax,16(%%rdi)\n"
+    /* c >>= 52 */
+    "shrdq $52,%%r9,%%r8\n"
+    "xorq %%r9,%%r9\n"
+    /* c += t3 */
+    "addq %%r10,%%r8\n"
+    /* c += d * R */
+    "movq %%rcx,%%rax\n"
+    "movq $0x1000003d10,%%rdx\n"
+    "mulq %%rdx\n"
+    "addq %%rax,%%r8\n"
+    "adcq %%rdx,%%r9\n"
+    /* r[3] = c & M */
+    "movq %%r8,%%rax\n"
+    "movq $0xfffffffffffff,%%rdx\n"
+    "andq %%rdx,%%rax\n"
+    "movq %%rax,24(%%rdi)\n"
+    /* c >>= 52 (%%r8 only) */
+    "shrdq $52,%%r9,%%r8\n"
+    /* c += t4 (%%r8 only) */
+    "addq %%rsi,%%r8\n"
+    /* r[4] = c */
+    "movq %%r8,32(%%rdi)\n"
+: "+S"(a), "=m"(tmp1), "=m"(tmp2), "=m"(tmp3)
+: "b"(b), "D"(r)
+: "%rax", "%rcx", "%rdx", "%r8", "%r9", "%r10", "%r11", "%r12", "%r13", "%r14", "%r15", "cc", "memory"
+);
+}
+
+SECP256K1_INLINE static void secp256k1_fe_sqr_inner(uint64_t *r, const uint64_t *a) {
+/**
+ * Registers: rdx:rax = multiplication accumulator
+ *            r9:r8   = c
+ *            rcx:rbx = d
+ *            r10-r14 = a0-a4
+ *            r15     = M (0xfffffffffffff)
+ *            rdi     = r
+ *            rsi     = a / t?
+ */
+  uint64_t tmp1, tmp2, tmp3;
+__asm__ __volatile__(
+    "movq 0(%%rsi),%%r10\n"
+    "movq 8(%%rsi),%%r11\n"
+    "movq 16(%%rsi),%%r12\n"
+    "movq 24(%%rsi),%%r13\n"
+    "movq 32(%%rsi),%%r14\n"
+    "movq $0xfffffffffffff,%%r15\n"
+
+    /* d = (a0*2) * a3 */
+    "leaq (%%r10,%%r10,1),%%rax\n"
+    "mulq %%r13\n"
+    "movq %%rax,%%rbx\n"
+    "movq %%rdx,%%rcx\n"
+    /* d += (a1*2) * a2 */
+    "leaq (%%r11,%%r11,1),%%rax\n"
+    "mulq %%r12\n"
+    "addq %%rax,%%rbx\n"
+    "adcq %%rdx,%%rcx\n"
+    /* c = a4 * a4 */
+    "movq %%r14,%%rax\n"
+    "mulq %%r14\n"
+    "movq %%rax,%%r8\n"
+    "movq %%rdx,%%r9\n"
+    /* d += (c & M) * R */
+    "andq %%r15,%%rax\n"
+    "movq $0x1000003d10,%%rdx\n"
+    "mulq %%rdx\n"
+    "addq %%rax,%%rbx\n"
+    "adcq %%rdx,%%rcx\n"
+    /* c >>= 52 (%%r8 only) */
+    "shrdq $52,%%r9,%%r8\n"
+    /* t3 (tmp1) = d & M */
+    "movq %%rbx,%%rsi\n"
+    "andq %%r15,%%rsi\n"
+    "movq %%rsi,%q1\n"
+    /* d >>= 52 */
+    "shrdq $52,%%rcx,%%rbx\n"
+    "xorq %%rcx,%%rcx\n"
+    /* a4 *= 2 */
+    "addq %%r14,%%r14\n"
+    /* d += a0 * a4 */
+    "movq %%r10,%%rax\n"
+    "mulq %%r14\n"
+    "addq %%rax,%%rbx\n"
+    "adcq %%rdx,%%rcx\n"
+    /* d+= (a1*2) * a3 */
+    "leaq (%%r11,%%r11,1),%%rax\n"
+    "mulq %%r13\n"
+    "addq %%rax,%%rbx\n"
+    "adcq %%rdx,%%rcx\n"
+    /* d += a2 * a2 */
+    "movq %%r12,%%rax\n"
+    "mulq %%r12\n"
+    "addq %%rax,%%rbx\n"
+    "adcq %%rdx,%%rcx\n"
+    /* d += c * R */
+    "movq %%r8,%%rax\n"
+    "movq $0x1000003d10,%%rdx\n"
+    "mulq %%rdx\n"
+    "addq %%rax,%%rbx\n"
+    "adcq %%rdx,%%rcx\n"
+    /* t4 = d & M (%%rsi) */
+    "movq %%rbx,%%rsi\n"
+    "andq %%r15,%%rsi\n"
+    /* d >>= 52 */
+    "shrdq $52,%%rcx,%%rbx\n"
+    "xorq %%rcx,%%rcx\n"
+    /* tx = t4 >> 48 (tmp3) */
+    "movq %%rsi,%%rax\n"
+    "shrq $48,%%rax\n"
+    "movq %%rax,%q3\n"
+    /* t4 &= (M >> 4) (tmp2) */
+    "movq $0xffffffffffff,%%rax\n"
+    "andq %%rax,%%rsi\n"
+    "movq %%rsi,%q2\n"
+    /* c = a0 * a0 */
+    "movq %%r10,%%rax\n"
+    "mulq %%r10\n"
+    "movq %%rax,%%r8\n"
+    "movq %%rdx,%%r9\n"
+    /* d += a1 * a4 */
+    "movq %%r11,%%rax\n"
+    "mulq %%r14\n"
+    "addq %%rax,%%rbx\n"
+    "adcq %%rdx,%%rcx\n"
+    /* d += (a2*2) * a3 */
+    "leaq (%%r12,%%r12,1),%%rax\n"
+    "mulq %%r13\n"
+    "addq %%rax,%%rbx\n"
+    "adcq %%rdx,%%rcx\n"
+    /* u0 = d & M (%%rsi) */
+    "movq %%rbx,%%rsi\n"
+    "andq %%r15,%%rsi\n"
+    /* d >>= 52 */
+    "shrdq $52,%%rcx,%%rbx\n"
+    "xorq %%rcx,%%rcx\n"
+    /* u0 = (u0 << 4) | tx (%%rsi) */
+    "shlq $4,%%rsi\n"
+    "movq %q3,%%rax\n"
+    "orq %%rax,%%rsi\n"
+    /* c += u0 * (R >> 4) */
+    "movq $0x1000003d1,%%rax\n"
+    "mulq %%rsi\n"
+    "addq %%rax,%%r8\n"
+    "adcq %%rdx,%%r9\n"
+    /* r[0] = c & M */
+    "movq %%r8,%%rax\n"
+    "andq %%r15,%%rax\n"
+    "movq %%rax,0(%%rdi)\n"
+    /* c >>= 52 */
+    "shrdq $52,%%r9,%%r8\n"
+    "xorq %%r9,%%r9\n"
+    /* a0 *= 2 */
+    "addq %%r10,%%r10\n"
+    /* c += a0 * a1 */
+    "movq %%r10,%%rax\n"
+    "mulq %%r11\n"
+    "addq %%rax,%%r8\n"
+    "adcq %%rdx,%%r9\n"
+    /* d += a2 * a4 */
+    "movq %%r12,%%rax\n"
+    "mulq %%r14\n"
+    "addq %%rax,%%rbx\n"
+    "adcq %%rdx,%%rcx\n"
+    /* d += a3 * a3 */
+    "movq %%r13,%%rax\n"
+    "mulq %%r13\n"
+    "addq %%rax,%%rbx\n"
+    "adcq %%rdx,%%rcx\n"
+    /* c += (d & M) * R */
+    "movq %%rbx,%%rax\n"
+    "andq %%r15,%%rax\n"
+    "movq $0x1000003d10,%%rdx\n"
+    "mulq %%rdx\n"
+    "addq %%rax,%%r8\n"
+    "adcq %%rdx,%%r9\n"
+    /* d >>= 52 */
+    "shrdq $52,%%rcx,%%rbx\n"
+    "xorq %%rcx,%%rcx\n"
+    /* r[1] = c & M */
+    "movq %%r8,%%rax\n"
+    "andq %%r15,%%rax\n"
+    "movq %%rax,8(%%rdi)\n"
+    /* c >>= 52 */
+    "shrdq $52,%%r9,%%r8\n"
+    "xorq %%r9,%%r9\n"
+    /* c += a0 * a2 (last use of %%r10) */
+    "movq %%r10,%%rax\n"
+    "mulq %%r12\n"
+    "addq %%rax,%%r8\n"
+    "adcq %%rdx,%%r9\n"
+    /* fetch t3 (%%r10, overwrites a0),t4 (%%rsi) */
+    "movq %q2,%%rsi\n"
+    "movq %q1,%%r10\n"
+    /* c += a1 * a1 */
+    "movq %%r11,%%rax\n"
+    "mulq %%r11\n"
+    "addq %%rax,%%r8\n"
+    "adcq %%rdx,%%r9\n"
+    /* d += a3 * a4 */
+    "movq %%r13,%%rax\n"
+    "mulq %%r14\n"
+    "addq %%rax,%%rbx\n"
+    "adcq %%rdx,%%rcx\n"
+    /* c += (d & M) * R */
+    "movq %%rbx,%%rax\n"
+    "andq %%r15,%%rax\n"
+    "movq $0x1000003d10,%%rdx\n"
+    "mulq %%rdx\n"
+    "addq %%rax,%%r8\n"
+    "adcq %%rdx,%%r9\n"
+    /* d >>= 52 (%%rbx only) */
+    "shrdq $52,%%rcx,%%rbx\n"
+    /* r[2] = c & M */
+    "movq %%r8,%%rax\n"
+    "andq %%r15,%%rax\n"
+    "movq %%rax,16(%%rdi)\n"
+    /* c >>= 52 */
+    "shrdq $52,%%r9,%%r8\n"
+    "xorq %%r9,%%r9\n"
+    /* c += t3 */
+    "addq %%r10,%%r8\n"
+    /* c += d * R */
+    "movq %%rbx,%%rax\n"
+    "movq $0x1000003d10,%%rdx\n"
+    "mulq %%rdx\n"
+    "addq %%rax,%%r8\n"
+    "adcq %%rdx,%%r9\n"
+    /* r[3] = c & M */
+    "movq %%r8,%%rax\n"
+    "andq %%r15,%%rax\n"
+    "movq %%rax,24(%%rdi)\n"
+    /* c >>= 52 (%%r8 only) */
+    "shrdq $52,%%r9,%%r8\n"
+    /* c += t4 (%%r8 only) */
+    "addq %%rsi,%%r8\n"
+    /* r[4] = c */
+    "movq %%r8,32(%%rdi)\n"
+: "+S"(a), "=m"(tmp1), "=m"(tmp2), "=m"(tmp3)
+: "D"(r)
+: "%rax", "%rbx", "%rcx", "%rdx", "%r8", "%r9", "%r10", "%r11", "%r12", "%r13", "%r14", "%r15", "cc", "memory"
+);
+}
 
 #endif