diff options
| author | MITSUNARI Shigeo <herumi@nifty.com> | 2018-08-29 13:01:45 +0800 |
|---|---|---|
| committer | MITSUNARI Shigeo <herumi@nifty.com> | 2018-08-29 14:05:05 +0800 |
| commit | b7c2d17e99a2b95a8eb941ccdd03c559d56a830f (patch) | |
| tree | 2a29098e24405be86d8688a11dc922b33052d8bd | |
| parent | 8919b73aa9bad85e431d4485378875454e757322 (diff) | |
| download | dexon-mcl-b7c2d17e99a2b95a8eb941ccdd03c559d56a830f.tar.gz dexon-mcl-b7c2d17e99a2b95a8eb941ccdd03c559d56a830f.tar.zst dexon-mcl-b7c2d17e99a2b95a8eb941ccdd03c559d56a830f.zip | |
add fp2Dbl_sqrPreA_
| -rw-r--r-- | include/mcl/op.hpp | 2 | ||||
| -rw-r--r-- | src/fp_generator.hpp | 42 | ||||
| -rw-r--r-- | test/bn_test.cpp | 2 |
3 files changed, 43 insertions, 3 deletions
diff --git a/include/mcl/op.hpp b/include/mcl/op.hpp index 96771ea..57ceb52 100644 --- a/include/mcl/op.hpp +++ b/include/mcl/op.hpp @@ -197,7 +197,7 @@ struct Op { void2u fpDbl_sqrPreA_; void2u fpDbl_modA_; void3u fp2Dbl_mulPreA_; - void3u fp2Dbl_sqrPreA_; + void2u fp2Dbl_sqrPreA_; size_t maxN; size_t N; size_t bitSize; diff --git a/src/fp_generator.hpp b/src/fp_generator.hpp index d955341..2e5f769 100644 --- a/src/fp_generator.hpp +++ b/src/fp_generator.hpp @@ -387,6 +387,9 @@ private: op.fp2Dbl_mulPreA_ = getCurr<void3u>(); gen_fp2Dbl_mulPre(mulPreL); align(16); + op.fp2Dbl_sqrPreA_ = getCurr<void2u>(); + gen_fp2Dbl_sqrPre(mulPreL); + align(16); op.fp2_mulA_ = getCurr<void3u>(); gen_fp2_mul4(fpDbl_modL); align(16); @@ -2867,7 +2870,44 @@ private: gen_raw_sub(gp0, gp1, gp2, rax, 4); gen_raw_fp_sub(gp0 + 8 * 4, gp1 + 8 * 4, gp2 + 8 * 4, Pack(gt0, gt1, gt2, gt3, gt4, gt5, gt6, gt7), true); } - + void gen_fp2Dbl_sqrPre(Label& mulPreL) + { + assert(!isFullBit_); + const RegExp y = rsp + 0 * 8; + const RegExp x = rsp + 1 * 8; + const Ext1 t1(FpByte_, rsp, 2 * 8); + const Ext1 t2(FpByte_, rsp, t1.next); + // use mulPreL then use 3 + StackFrame sf(this, 3 /* not 2 */, 10 | UseRDX, t2.next); + mov(ptr [y], gp0); + mov(ptr [x], gp1); + const Pack a = sf.t.sub(0, 4); + const Pack b = sf.t.sub(4, 4); + load_rm(b, gp1 + FpByte_); + for (int i = 0; i < 4; i++) { + mov(rax, b[i]); + if (i == 0) { + add(rax, rax); + } else { + adc(rax, rax); + } + mov(ptr [(const RegExp&)t1 + i * 8], rax); + } + load_rm(a, gp1); + add_rr(a, b); + store_mr(t2, a); + mov(gp0, ptr [y]); + add(gp0, FpByte_ * 2); + lea(gp1, ptr [t1]); + mov(gp2, ptr [x]); + call(mulPreL); + mov(gp0, ptr [x]); + gen_raw_fp_sub(t1, gp0, gp0 + FpByte_, sf.t, false); + mov(gp0, ptr [y]); + lea(gp1, ptr [t1]); + lea(gp2, ptr [t2]); + call(mulPreL); + } void gen_fp2_add4() { assert(!isFullBit_); diff --git a/test/bn_test.cpp b/test/bn_test.cpp index af57309..929e235 100644 --- a/test/bn_test.cpp +++ b/test/bn_test.cpp @@ -358,7 +358,7 @@ CYBOZU_TEST_AUTO(naive) #ifdef ONLY_BENCH { Fp12 e; - for (int i = 0; i < 1000; i++) pairing(e, P, Q); + for (int i = 0; i < 10000; i++) { clk.begin(); pairing(e, P, Q); clk.end(); } } clk.put(); return; |