diff --git a/Makefile b/Makefile
index 1bc145a..1f1ed6c 100644
--- a/Makefile
+++ b/Makefile
@@ -10,12 +10,16 @@ NOASM        ?=
 TEST_PATH    ?= ./...
 GOCACHE      ?= off
 BENCH_OPTS   ?= -v -bench=. -run="NonExistingTest"
-
+TEST_PATH    ?= ./...
 
 ifeq ($(NOASM),1)
 	OPTS+=$(OPTS_TAGS)
 endif
 
+ifeq ($(PPROF),1)
+	BENCH_OPTS+= -cpuprofile=cpu.out -memprofile=mem0.out
+endif
+
 TARGETS= \
 	dh   \
 	drbg \
@@ -32,16 +36,16 @@ make_dirs:
 	cp -rf etc $(GOPATH_LOCAL)/$(GOPATH_DIR)
 
 test: clean make_dirs $(addprefix prep-,$(TARGETS))
-	cd $(GOPATH_LOCAL); GOPATH=$(GOPATH_LOCAL) go test $(OPTS) -v ./...
+	cd $(GOPATH_LOCAL); GOPATH=$(GOPATH_LOCAL) go test $(OPTS) -v $(TEST_PATH)
 
 cover:
 	cd $(GOPATH_LOCAL); GOPATH=$(GOPATH_LOCAL) go test \
-		-race -coverprofile=coverage_$(NOASM).txt -covermode=atomic $(OPTS) -v ./...
+		-race -coverprofile=coverage_$(NOASM).txt -covermode=atomic $(OPTS) -v $(TEST_PATH)
 	cat $(GOPATH_LOCAL)/coverage_$(NOASM).txt >> coverage.txt
 
 bench: clean $(addprefix prep-,$(TARGETS))
 	cd $(GOPATH_LOCAL); GOCACHE=$(GOCACHE) GOPATH=$(GOPATH_LOCAL) $(GO) test \
-		$(BENCH_OPTS) ./...
+		$(BENCH_OPTS) $(TEST_PATH)
 
 clean:
 	rm -rf $(GOPATH_LOCAL)
@@ -51,3 +55,9 @@ vendor-sidh-for-tls: clean
 	mkdir -p $(VENDOR_DIR)/github_com/henrydcase/nobs/
 	rsync -a . $(VENDOR_DIR)/github_com/henrydcase/nobs/ --exclude=$(VENDOR_DIR) --exclude=.git --exclude=.travis.yml --exclude=README.md
 	find $(VENDOR_DIR) -type f -print0 -name "*.go" | xargs -0 sed -i 's/github\.com/github_com/g'
+
+pprof-cpu:
+	$(GO) tool pprof $(GOPATH_LOCAL)/cpu.out
+
+pprof-mem:
+	$(GO) tool pprof $(GOPATH_LOCAL)/mem0.out
diff --git a/dh/sidh/p503/arith_amd64.s b/dh/sidh/p503/arith_amd64.s
index 0808e88..5968754 100644
--- a/dh/sidh/p503/arith_amd64.s
+++ b/dh/sidh/p503/arith_amd64.s
@@ -90,31 +90,42 @@ TEXT ·fp503ConditionalSwap(SB),NOSPLIT,$0-17
 
 	MOVQ	x+0(FP), REG_P1
 	MOVQ	y+8(FP), REG_P2
-	MOVB	choice+16(FP), AL	// AL = 0 or 1
-	MOVBLZX	AL, AX				// AX = 0 or 1
-	NEGQ	AX					// RAX = 0x00..00 or 0xff..ff
+	MOVBLZX	choice+16(FP), AX	// AL = 0 or 1
+
+	// Make AX, so that either all bits are set or non
+	// AX = 0 or 1
+	NEGQ	AX
+
+	// Fill xmm15. After this step first half of XMM15 is
+	// just zeros and second half is whatever in AX
+	MOVQ    AX, X15
+
+	// Copy lower double word everywhere else. So that
+	// XMM15=AL|AL|AL|AL. As AX has either all bits set
+	// or non result will be that XMM15 has also either
+	// all bits set or non of them.
+	PSHUFD $0, X15, X15
 
 #ifndef CSWAP_BLOCK
 #define CSWAP_BLOCK(idx) 	\
-	MOVQ	(idx*8)(REG_P1), BX	\ // BX = x[idx]
-	MOVQ 	(idx*8)(REG_P2), CX	\ // CX = y[idx]
-	MOVQ	CX, DX				\ // DX = y[idx]
-	XORQ	BX, DX				\ // DX = y[idx] ^ x[idx]
-	ANDQ	AX, DX				\ // DX = (y[idx] ^ x[idx]) & mask
-	XORQ	DX, BX				\ // BX = (y[idx] ^ x[idx]) & mask) ^ x[idx] = x[idx] or y[idx]
-	XORQ	DX, CX				\ // CX = (y[idx] ^ x[idx]) & mask) ^ y[idx] = y[idx] or x[idx]
-	MOVQ	BX, (idx*8)(REG_P1)	\
-	MOVQ	CX, (idx*8)(REG_P2)
+	MOVOU	(idx*16)(REG_P1), X0 \
+	MOVOU 	(idx*16)(REG_P2), X1 \
+	\ // X2 = mask & (X0 ^ X1)
+	MOVO	X1, X2 \
+	PXOR	X0, X2 \
+	PAND	X15, X2 \
+	\
+	PXOR	X2, X0 \
+	PXOR	X2, X1 \
+	\
+	MOVOU	X0, (idx*16)(REG_P1) \
+	MOVOU	X1, (idx*16)(REG_P2)
 #endif
 
 	CSWAP_BLOCK(0)
 	CSWAP_BLOCK(1)
 	CSWAP_BLOCK(2)
 	CSWAP_BLOCK(3)
-	CSWAP_BLOCK(4)
-	CSWAP_BLOCK(5)
-	CSWAP_BLOCK(6)
-	CSWAP_BLOCK(7)
 
 #ifdef CSWAP_BLOCK
 #undef CSWAP_BLOCK
diff --git a/dh/sidh/p751/arith_amd64.s b/dh/sidh/p751/arith_amd64.s
index 4596d8f..aefb448 100644
--- a/dh/sidh/p751/arith_amd64.s
+++ b/dh/sidh/p751/arith_amd64.s
@@ -126,130 +126,48 @@ TEXT ·fp751ConditionalSwap(SB), NOSPLIT, $0-17
 
 	MOVQ	x+0(FP), REG_P1
 	MOVQ	y+8(FP), REG_P2
-	MOVB	choice+16(FP), AL	// AL = 0 or 1
-	MOVBLZX	AL, AX			// AX = 0 or 1
-	NEGQ	AX			// RAX = 0x00..00 or 0xff..ff
+	MOVBLZX	choice+16(FP), AX	// AL = 0 or 1
 
-	MOVQ	(0*8)(REG_P1), BX	// BX = x[0]
-	MOVQ 	(0*8)(REG_P2), CX	// CX = y[0]
-	MOVQ	CX, DX			// DX = y[0]
-	XORQ	BX, DX			// DX = y[0] ^ x[0]
-	ANDQ	AX, DX			// DX = (y[0] ^ x[0]) & mask
-	XORQ	DX, BX			// BX = (y[0] ^ x[0]) & mask) ^ x[0] = x[0] or y[0]
-	XORQ	DX, CX			// CX = (y[0] ^ x[0]) & mask) ^ y[0] = y[0] or x[0]
-	MOVQ	BX, (0*8)(REG_P1)
-	MOVQ	CX, (0*8)(REG_P2)
+	// Make AX, so that either all bits are set or non
+	// AX = 0 or 1
+	NEGQ	AX
 
-	MOVQ	(1*8)(REG_P1), BX
-	MOVQ 	(1*8)(REG_P2), CX
-	MOVQ	CX, DX
-	XORQ	BX, DX
-	ANDQ	AX, DX
-	XORQ	DX, BX
-	XORQ	DX, CX
-	MOVQ	BX, (1*8)(REG_P1)
-	MOVQ	CX, (1*8)(REG_P2)
+	// Fill xmm15. After this step first half of XMM15 is
+	// just zeros and second half is whatever in AX
+	MOVQ    AX, X15
 
-	MOVQ	(2*8)(REG_P1), BX
-	MOVQ 	(2*8)(REG_P2), CX
-	MOVQ	CX, DX
-	XORQ	BX, DX
-	ANDQ	AX, DX
-	XORQ	DX, BX
-	XORQ	DX, CX
-	MOVQ	BX, (2*8)(REG_P1)
-	MOVQ	CX, (2*8)(REG_P2)
+	// Copy lower double word everywhere else. So that
+	// XMM15=AL|AL|AL|AL. As AX has either all bits set
+	// or non result will be that XMM15 has also either
+	// all bits set or non of them.
+	PSHUFD $0, X15, X15
 
-	MOVQ	(3*8)(REG_P1), BX
-	MOVQ 	(3*8)(REG_P2), CX
-	MOVQ	CX, DX
-	XORQ	BX, DX
-	ANDQ	AX, DX
-	XORQ	DX, BX
-	XORQ	DX, CX
-	MOVQ	BX, (3*8)(REG_P1)
-	MOVQ	CX, (3*8)(REG_P2)
+#ifndef CSWAP_BLOCK
+#define CSWAP_BLOCK(idx) 	\
+	MOVOU	(idx*16)(REG_P1), X0 \
+	MOVOU 	(idx*16)(REG_P2), X1 \
+	\ // X2 = mask & (X0 ^ X1)
+	MOVO	X1, X2 \
+	PXOR	X0, X2 \
+	PAND	X15, X2 \
+	\
+	PXOR	X2, X0 \
+	PXOR	X2, X1 \
+	\
+	MOVOU	X0, (idx*16)(REG_P1)	\
+	MOVOU	X1, (idx*16)(REG_P2)
+#endif
 
-	MOVQ	(4*8)(REG_P1), BX
-	MOVQ 	(4*8)(REG_P2), CX
-	MOVQ	CX, DX
-	XORQ	BX, DX
-	ANDQ	AX, DX
-	XORQ	DX, BX
-	XORQ	DX, CX
-	MOVQ	BX, (4*8)(REG_P1)
-	MOVQ	CX, (4*8)(REG_P2)
-
-	MOVQ	(5*8)(REG_P1), BX
-	MOVQ 	(5*8)(REG_P2), CX
-	MOVQ	CX, DX
-	XORQ	BX, DX
-	ANDQ	AX, DX
-	XORQ	DX, BX
-	XORQ	DX, CX
-	MOVQ	BX, (5*8)(REG_P1)
-	MOVQ	CX, (5*8)(REG_P2)
-
-	MOVQ	(6*8)(REG_P1), BX
-	MOVQ 	(6*8)(REG_P2), CX
-	MOVQ	CX, DX
-	XORQ	BX, DX
-	ANDQ	AX, DX
-	XORQ	DX, BX
-	XORQ	DX, CX
-	MOVQ	BX, (6*8)(REG_P1)
-	MOVQ	CX, (6*8)(REG_P2)
-
-	MOVQ	(7*8)(REG_P1), BX
-	MOVQ 	(7*8)(REG_P2), CX
-	MOVQ	CX, DX
-	XORQ	BX, DX
-	ANDQ	AX, DX
-	XORQ	DX, BX
-	XORQ	DX, CX
-	MOVQ	BX, (7*8)(REG_P1)
-	MOVQ	CX, (7*8)(REG_P2)
-
-	MOVQ	(8*8)(REG_P1), BX
-	MOVQ 	(8*8)(REG_P2), CX
-	MOVQ	CX, DX
-	XORQ	BX, DX
-	ANDQ	AX, DX
-	XORQ	DX, BX
-	XORQ	DX, CX
-	MOVQ	BX, (8*8)(REG_P1)
-	MOVQ	CX, (8*8)(REG_P2)
-
-	MOVQ	(9*8)(REG_P1), BX
-	MOVQ 	(9*8)(REG_P2), CX
-	MOVQ	CX, DX
-	XORQ	BX, DX
-	ANDQ	AX, DX
-	XORQ	DX, BX
-	XORQ	DX, CX
-	MOVQ	BX, (9*8)(REG_P1)
-	MOVQ	CX, (9*8)(REG_P2)
-
-	MOVQ	(10*8)(REG_P1), BX
-	MOVQ 	(10*8)(REG_P2), CX
-	MOVQ	CX, DX
-	XORQ	BX, DX
-	ANDQ	AX, DX
-	XORQ	DX, BX
-	XORQ	DX, CX
-	MOVQ	BX, (10*8)(REG_P1)
-	MOVQ	CX, (10*8)(REG_P2)
-
-	MOVQ	(11*8)(REG_P1), BX
-	MOVQ 	(11*8)(REG_P2), CX
-	MOVQ	CX, DX
-	XORQ	BX, DX
-	ANDQ	AX, DX
-	XORQ	DX, BX
-	XORQ	DX, CX
-	MOVQ	BX, (11*8)(REG_P1)
-	MOVQ	CX, (11*8)(REG_P2)
+	CSWAP_BLOCK(0)
+	CSWAP_BLOCK(1)
+	CSWAP_BLOCK(2)
+	CSWAP_BLOCK(3)
+	CSWAP_BLOCK(4)
+	CSWAP_BLOCK(5)
 
+#ifdef CSWAP_BLOCK
+#undef CSWAP_BLOCK
+#endif
 	RET
 
 TEXT ·fp751AddReduced(SB), NOSPLIT, $0-24