fix mulx argument order using sed, test feadd, femul (fails due to #234)

3 files changed, 95 insertions, 1 deletions

diff --git a/src/Specific/NISTP256/AMD64/measurements.txt b/src/Specific/NISTP256/AMD64/measurements.txt
index 1ec583018..d4dca8e54 100644
--- a/src/Specific/NISTP256/AMD64/measurements.txt
+++ b/src/Specific/NISTP256/AMD64/measurements.txt
@@ -1,2 +1,2 @@
 1576	ashryn-noht-notb-noac-broadwell	2.60ghz	7.1.1	cac291e0
-1688	ashryn-noht-notb-ac-broadwell	2.50ghz	7.1.1	d7ad9528
+1640	ashryn-noht-notb-ac-broadwell	2.50ghz	7.1.1	6c558bd2
diff --git a/src/Specific/NISTP256/AMD64/test/feadd_test.c b/src/Specific/NISTP256/AMD64/test/feadd_test.c
new file mode 100644
index 000000000..3cdd1fd60
--- /dev/null
+++ b/src/Specific/NISTP256/AMD64/test/feadd_test.c
@@ -0,0 +1,46 @@
+#include <stdint.h>
+#include <inttypes.h>
+#include <stdio.h>
+#include "feadd.h"
+
+static int check(uint64_t out[4], uint64_t ref[4]) {
+  return out[0] == ref[0] && out[1] == ref[1] && out[2] == ref[2] && out[3] == ref[3];
+}
+
+int main() {
+  {
+    uint64_t out[4] = {0};
+    uint64_t in1[4] = {0, 0, 0, 1};
+    uint64_t in2[4] = {0, 0, 0, 1};
+    uint64_t ref[4] = {0, 0, 0, 2};
+    feadd(out, in1[0], in1[1], in1[2], in1[3], in2[0], in2[1], in2[2], in2[3]);
+    if (!check(out, ref)) return 1;
+  }
+  {
+    uint64_t out[4] = {0};
+    uint64_t in1[4] = {0, 0, 0, 0};
+    uint64_t in2[4] = {0, 0, 0, 0};
+    uint64_t ref[4] = {0, 0, 0, 0};
+    feadd(out, in1[0], in1[1], in1[2], in1[3], in2[0], in2[1], in2[2], in2[3]);
+    if (!check(out, ref)) return 2;
+  }
+  {
+    uint64_t out[4] = {0};
+    uint64_t in1[4] = {0xffffffff00000001, 0x0000000000000000, 0x00000000ffffffff, 0xfffffffffffffffe}; // p256-1
+    uint64_t in2[4] = {0, 0, 0, 1};
+    uint64_t ref[4] = {0, 0, 0, 0};
+    feadd(out, in1[0], in1[1], in1[2], in1[3], in2[0], in2[1], in2[2], in2[3]);
+    if (!check(out, ref)) return 3;
+  }
+  {
+    uint64_t out[4] = {0};
+    uint64_t in1[4] = {0xffffffff00000001, 0x0000000000000000, 0x00000000ffffffff, 0xfffffffffffffffe}; // p256-1
+    uint64_t in2[4] = {0, 0, 0, 7};
+    uint64_t ref[4] = {0, 0, 0, 6};
+    feadd(out, in1[0], in1[1], in1[2], in1[3], in2[0], in2[1], in2[2], in2[3]);
+    if (!check(out, ref)) return 4;
+  }
+  
+  //printf("0x%016" PRIx64 " 0x%016" PRIx64 " 0x%016" PRIx64 " 0x%016" PRIx64 "\n", out[0], out[1], out[2], out[3]);
+  //printf("((((((0x%016" PRIx64 "<<64)+ 0x%016" PRIx64 ")<<64)+ 0x%016" PRIx64 ")<<64)+ 0x%016" PRIx64 ")\n", out[0], out[1], out[2], out[3]);
+}
diff --git a/src/Specific/NISTP256/AMD64/test/femul_test.c b/src/Specific/NISTP256/AMD64/test/femul_test.c
new file mode 100644
index 000000000..9519399e0
--- /dev/null
+++ b/src/Specific/NISTP256/AMD64/test/femul_test.c
@@ -0,0 +1,48 @@
+#include <stdint.h>
+#include <inttypes.h>
+#include <stdio.h>
+#include "femul.h"
+
+// all arrays big-endian
+static uint64_t Rmodm[4] = {0xfffffffe, 0xffffffffffffffff, 0xffffffff00000000, 0x0000000000000001};
+
+int main() {
+  uint64_t out[4] = {0};
+  if ( femul(out,
+        Rmodm[0], Rmodm[1], Rmodm[2], Rmodm[3],
+        Rmodm[0], Rmodm[1], Rmodm[2], Rmodm[3]), 
+      ! (out[0]   == Rmodm[0]
+        && out[1] == Rmodm[1]
+        && out[2] == Rmodm[2]
+        && out[3] == Rmodm[3]
+        )) { return 1; }
+  if ( femul(out,
+        0, 0, 0, 0,
+        Rmodm[0], Rmodm[1], Rmodm[2], Rmodm[3]), 
+      ! (out[0]   == 0
+        && out[1] == 0
+        && out[2] == 0
+        && out[3] == 0
+        )) { return 2; }
+  if ( femul(out,
+        0, 0, 0, 1,
+        0, 0, 0, 1), 
+      ! (out[0]   == 0xfffffffe00000003 // R^-1
+        && out[1] == 0xfffffffd00000002
+        && out[2] == 0x00000001fffffffe
+        && out[3] == 0x0000000300000000
+        )) { return 3; }
+  if ( femul(out,
+        0, 0, 0, 1,
+        0x4fffffffd, 0xfffffffffffffffe, 0xfffffffbffffffff, 0x0000000000000003),  // R^2
+      ! (out[0]   == Rmodm[0]
+        && out[1] == Rmodm[1]
+        && out[2] == Rmodm[2]
+        && out[3] == Rmodm[3]
+        )) { return 4; }
+
+
+  //printf("0x%016" PRIx64 " 0x%016" PRIx64 " 0x%016" PRIx64 " 0x%016" PRIx64 "\n", out[0], out[1], out[2], out[3]);
+  //printf("((((((0x%016" PRIx64 "<<64)+ 0x%016" PRIx64 ")<<64)+ 0x%016" PRIx64 ")<<64)+ 0x%016" PRIx64 ")\n", out[0], out[1], out[2], out[3]);
+  return 0;
+}