From 3f7ac71a3c95c90dfd1ec1fc3d2061d8176d5e18 Mon Sep 17 00:00:00 2001
From: MITSUNARI Shigeo <herumi@nifty.com>
Date: Mon, 11 Jul 2016 16:40:25 +0900
Subject: [PATCH] use llvm for large prime

---
 include/mcl/fp.hpp | 9 +++++++--
 sample/large.cpp   | 4 ++++
 src/fp_proto.hpp   | 3 +++
 src/gen.cpp        | 8 +++++---
 4 files changed, 19 insertions(+), 5 deletions(-)

diff --git a/include/mcl/fp.hpp b/include/mcl/fp.hpp
index 236820b..aadf8c4 100644
--- a/include/mcl/fp.hpp
+++ b/include/mcl/fp.hpp
@@ -119,8 +119,13 @@ public:
 #else
 		if (mode == fp::FP_LLVM || mode == fp::FP_LLVM_MONT) mode = fp::FP_AUTO;
 #endif
-		if (mode == fp::FP_AUTO) mode = fp::FP_GMP_MONT;
-		if (maxBitSize > 576 && mode != fp::FP_GMP_MONT) mode = fp::FP_GMP;
+		if (mode == fp::FP_AUTO) {
+			if (maxBitSize > 576) {
+				mode = fp::FP_GMP; // QQQ : slower than FP_GMP_MONT if maxBitSize == 768
+			} else {
+				mode = fp::FP_GMP_MONT;
+			}
+		}
 
 		op_.isMont = mode == fp::FP_GMP_MONT || mode == fp::FP_LLVM_MONT || mode == fp::FP_XBYAK;
 		if (mode == fp::FP_GMP_MONT || mode == fp::FP_LLVM_MONT) {
diff --git a/sample/large.cpp b/sample/large.cpp
index f7fbd86..fcfd93b 100644
--- a/sample/large.cpp
+++ b/sample/large.cpp
@@ -23,6 +23,10 @@ int main()
 {
 	test(mcl::fp::FP_GMP);
 	test(mcl::fp::FP_GMP_MONT);
+#ifdef MCL_USE_LLVM
+	test(mcl::fp::FP_LLVM);
+	test(mcl::fp::FP_LLVM_MONT);
+#endif
 } catch (std::exception& e) {
 	printf("err %s\n", e.what());
 	puts("make clean");
diff --git a/src/fp_proto.hpp b/src/fp_proto.hpp
index 87c728e..e68d19b 100644
--- a/src/fp_proto.hpp
+++ b/src/fp_proto.hpp
@@ -50,6 +50,9 @@ void mcl_fpDbl_add224(mcl::fp::Unit*, const mcl::fp::Unit*, const mcl::fp::Unit*
 void mcl_fpDbl_sub224(mcl::fp::Unit*, const mcl::fp::Unit*, const mcl::fp::Unit*, const mcl::fp::Unit*);
 #else
 MCL_FP_DEF_FUNC(576)
+MCL_FP_DEF_FUNC(640)
+MCL_FP_DEF_FUNC(704)
+MCL_FP_DEF_FUNC(768)
 #endif
 
 #undef MCL_FP_DEF_FUNC
diff --git a/src/gen.cpp b/src/gen.cpp
index abb128c..f453004 100644
--- a/src/gen.cpp
+++ b/src/gen.cpp
@@ -1,5 +1,6 @@
 #include "llvm_gen.hpp"
 #include <cybozu/option.hpp>
+#include <mcl/op.hpp>
 #include <map>
 #include <set>
 #include <fstream>
@@ -810,11 +811,11 @@ struct Code : public mcl::Generator {
 		unit2 = unit * 2;
 		unitStr = cybozu::itoa(unit);
 	}
-	void gen(const StrSet& privateFuncList)
+	void gen(const StrSet& privateFuncList, uint32_t maxBitSize)
 	{
 		this->privateFuncList = &privateFuncList;
 		gen_once();
-		uint32_t end = ((576 + unit - 1) / unit) * unit;
+		uint32_t end = ((maxBitSize + unit - 1) / unit) * unit;
 		for (uint32_t i = 64; i <= end; i += unit) {
 			setBit(i);
 			gen_all();
@@ -854,7 +855,8 @@ int main(int argc, char *argv[])
 		c.setOldLLVM();
 	}
 	c.setUnit(unit);
-	c.gen(privateFuncList);
+	uint32_t maxBitSize = MCL_MAX_OP_BIT_SIZE;
+	c.gen(privateFuncList, maxBitSize);
 } catch (std::exception& e) {
 	printf("ERR %s\n", e.what());
 	return 1;