SIKE: enable optimized version

.gitmodules

@@ -7,3 +7,6 @@
 [submodule "3rd/gbench"]
 	path = 3rd/gbench
 	url = https://github.com/henrydcase/benchmark.git
+[submodule "3rd/cpu_features"]
+	path = 3rd/cpu_features
+	url = https://github.com/google/cpu_features.git

3rd/cpu_features

@@ -0,0 +1 @@
+Subproject commit 3e8243b7d9951c078259c3186c039a6e8f036055

CMakeLists.txt

@@ -34,6 +34,7 @@ else()
 endif()
 
 add_subdirectory(3rd/gtest)
+add_subdirectory(3rd/cpu_features)
 
 # Arch settings
 
@@ -82,6 +83,7 @@ include_directories(
   public
   src/common/
   src
+  3rd/cpu_features/include
 )
 
 set_property(GLOBAL PROPERTY obj_libs "")
@@ -243,12 +245,14 @@ get_property(OBJ_LIBS GLOBAL PROPERTY obj_libs)
 target_link_libraries(
   pqc
   common
+  cpu_features
   ${OBJ_LIBS}
 )
 
 target_link_libraries(
   pqc_s
   common
+  cpu_features
   ${OBJ_LIBS}
 )
 

src/capi/pqapi.c

@@ -1,6 +1,7 @@
 #include <stdint.h>
 #include <stdbool.h>
 #include <pqc/pqc.h>
+#include <cpuinfo_x86.h>
 
 #include "schemes.h"
 
@@ -126,3 +127,9 @@ bool pqc_sig_verify(const params_t *p,
     const uint8_t *pk) {
     return !((sig_params_t *)p)->verify(sig, siglen, m, mlen, pk);
 }
+
+X86Features CPU_CAPS;
+void static_initialization(void) __attribute__((constructor));
+void static_initialization(void) {
+    CPU_CAPS = GetX86Info().features;
+}

src/kem/sike/CMakeLists.txt

@@ -1,12 +1,13 @@
 set(
   SRC_CLEAN_SIKE_P434
   p434/fpx.c
-  p434/isogeny.c
   p434/fp_generic.c
+  p434/fp_glue.c
+  p434/fp-x86_64.S
+  p434/isogeny.c
   p434/params.c
   p434/sike.c)
 
 define_kem_alg(
   sike_p434_clean
   PQC_SIKEP434_CLEAN "${SRC_CLEAN_SIKE_P434}" "${CMAKE_CURRENT_SOURCE_DIR}")
-

src/kem/sike/p434/fp-x86_64.S

@@ -15,10 +15,10 @@
 .quad	0x6CFC5FD681C52056
 .quad	0x0002341F27177344
 
-.globl	sike_fpadd
+.globl	sike_fpadd_asm
-.hidden sike_fpadd
+.hidden sike_fpadd_asm
-.type	sike_fpadd,@function
+.type	sike_fpadd_asm,@function
-sike_fpadd:
+sike_fpadd_asm:
 .cfi_startproc
 	pushq	%r12
 .cfi_adjust_cfa_offset	8
@@ -107,14 +107,7 @@ sike_fpadd:
 .hidden sike_cswap_asm
 .type	sike_cswap_asm,@function
 sike_cswap_asm:
-
-
 	movq	%rdx,%xmm3
-
-
-
-
-
 	pshufd	$68,%xmm3,%xmm3
 
 	movdqu	0(%rdi),%xmm0
@@ -258,10 +251,10 @@ sike_cswap_asm:
 	movdqu	%xmm1,208(%rsi)
 
 	.byte	0xf3,0xc3
-.globl	sike_fpsub
+.globl	sike_fpsub_asm
-.hidden sike_fpsub
+.hidden sike_fpsub_asm
-.type	sike_fpsub,@function
+.type	sike_fpsub_asm,@function
-sike_fpsub:
+sike_fpsub_asm:
 .cfi_startproc
 	pushq	%r12
 .cfi_adjust_cfa_offset	8
@@ -508,10 +501,10 @@ sike_mpdblsubx2_asm:
 	.byte	0xf3,0xc3
 .cfi_endproc
 
-.globl	sike_fprdc
+.globl	sike_fprdc_asm
-.hidden sike_fprdc
+.hidden sike_fprdc_asm
-.type	sike_fprdc,@function
+.type	sike_fprdc_asm,@function
-sike_fprdc:
+sike_fprdc_asm:
 .cfi_startproc
 	pushq	%r12
 .cfi_adjust_cfa_offset	8
@@ -723,10 +716,10 @@ sike_fprdc:
 .cfi_adjust_cfa_offset	-8
 	.byte	0xf3,0xc3
 .cfi_endproc
-.globl	sike_mpmul
+.globl	sike_mpmul_asm
-.hidden sike_mpmul
+.hidden sike_mpmul_asm
-.type	sike_mpmul,@function
+.type	sike_mpmul_asm,@function
-sike_mpmul:
+sike_mpmul_asm:
 .cfi_startproc
 	pushq	%r12
 .cfi_adjust_cfa_offset	8

src/kem/sike/p434/fp_generic.c

@@ -5,12 +5,16 @@
 *********************************************************************************************/
 #include "utils.h"
 #include "fpx.h"
+#include <cpuinfo_x86.h>
+
+extern X86Features CPU_CAPS;
 
 // Global constants
 extern const struct params_t params;
 
+// Digit multiplication, digit * digit -> 2-digit result
 static void digit_x_digit(const crypto_word_t a, const crypto_word_t b, crypto_word_t* c)
-{ // Digit multiplication, digit * digit -> 2-digit result
+{
     crypto_word_t al, ah, bl, bh, temp;
     crypto_word_t albl, albh, ahbl, ahbh, res1, res2, res3, carry;
     crypto_word_t mask_low = (crypto_word_t)(-1) >> (sizeof(crypto_word_t)*4);
@@ -43,10 +47,11 @@ static void digit_x_digit(const crypto_word_t a, const crypto_word_t b, crypto_w
     c[1] ^= (ahbh & mask_high) + carry;             // C11
 }
 
-void sike_fpadd(const felm_t a, const felm_t b, felm_t c)
+// Modular addition, c = a+b mod p434.
-{ // Modular addition, c = a+b mod p434.
 // Inputs: a, b in [0, 2*p434-1]
 // Output: c in [0, 2*p434-1]
+void sike_fpadd(const felm_t a, const felm_t b, felm_t c)
+{
     unsigned int i, carry = 0;
     crypto_word_t mask;
 
@@ -84,12 +89,20 @@ void sike_fpsub(const felm_t a, const felm_t b, felm_t c)
     }
 }
 
+// Multiprecision comba multiply, c = a*b, where lng(a) = lng(b) = NWORDS_FIELD.
+void sike_mpmul_asm(const felm_t a, const felm_t b, dfelm_t c);
 void sike_mpmul(const felm_t a, const felm_t b, dfelm_t c)
-{ // Multiprecision comba multiply, c = a*b, where lng(a) = lng(b) = NWORDS_FIELD.
+{
     unsigned int i, j;
     crypto_word_t t = 0, u = 0, v = 0, UV[2];
     unsigned int carry = 0;
 
+    // TODO: it actually needs BMI2 & ADOX. cpu_features needs to be updated
+    if (CPU_CAPS.bmi2) {
+        sike_mpmul_asm(a,b,c);
+        return;
+    }
+
     for (i = 0; i < NWORDS_FIELD; i++) {
         for (j = 0; j <= i; j++) {
             MUL(a[j], b[i-j], UV+1, UV[0]);
@@ -118,11 +131,18 @@ void sike_mpmul(const felm_t a, const felm_t b, dfelm_t c)
     c[2*NWORDS_FIELD-1] = v;
 }
 
-void sike_fprdc(const felm_t ma, felm_t mc)
+// Efficient Montgomery reduction using comba and exploiting the special form of the prime p434.
-{ // Efficient Montgomery reduction using comba and exploiting the special form of the prime p434.
 // mc = ma*R^-1 mod p434x2, where R = 2^448.
 // If ma < 2^448*p434, the output mc is in the range [0, 2*p434-1].
 // ma is assumed to be in Montgomery representation.
+void sike_fprdc_asm(const felm_t ma, felm_t mc);
+void sike_fprdc(const felm_t ma, felm_t mc)
+{
+    if (CPU_CAPS.bmi2) {
+        sike_fprdc_asm(ma, mc);
+        return;
+    }
+
     unsigned int i, j, carry, count = ZERO_WORDS;
     crypto_word_t UV[2], t = 0, u = 0, v = 0;
 

src/kem/sike/p434/fp_glue.c

@@ -0,0 +1,4 @@
+#include "fpx.h"
+#include "utils.h"
+
+void sike_mpmul_asm_X(const felm_t a, const felm_t b, dfelm_t c);