From ca817fcb69ea8ff09f63416d169ce7b0c5bcf235 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Mamy=20Andr=C3=A9-Ratsimbazafy?= <mamy_github@numforge.co>
Date: Sat, 29 Feb 2020 18:27:20 +0100
Subject: [PATCH] Use Assembly cmov on x86

---
 benchmarks/bls12_381_fp.nim              |  8 +--
 constantine/primitives/constant_time.nim | 70 ++++++++++++++++++++----
 2 files changed, 62 insertions(+), 16 deletions(-)

diff --git a/benchmarks/bls12_381_fp.nim b/benchmarks/bls12_381_fp.nim
index 17bc11f..cf6bd40 100644
--- a/benchmarks/bls12_381_fp.nim
+++ b/benchmarks/bls12_381_fp.nim
@@ -43,8 +43,8 @@ proc addBench() =
     x += y
   let stop = getMonotime()
 
-  echo &"Time for {Iters} conditional additions (constant-time 381-bit): {inMilliseconds(stop-start)} ms"
-  echo &"Time for 1 conditional addition ==> {inNanoseconds((stop-start) div Iters)} ns"
+  echo &"Time for {Iters} additions in 𝔽p (constant-time 381-bit): {inMilliseconds(stop-start)} ms"
+  echo &"Time for 1 addition in 𝔽p ==> {inNanoseconds((stop-start) div Iters)} ns"
 
 addBench()
 
@@ -60,7 +60,7 @@ proc mulBench() =
     r.prod(x, y)
   let stop = getMonotime()
 
-  echo &"Time for {Iters} multiplications (constant-time 381-bit): {inMilliseconds(stop-start)} ms"
-  echo &"Time for 1 multiplication ==> {inNanoseconds((stop-start) div Iters)} ns"
+  echo &"Time for {Iters} multiplications 𝔽p (constant-time 381-bit): {inMilliseconds(stop-start)} ms"
+  echo &"Time for 1 multiplication 𝔽p ==> {inNanoseconds((stop-start) div Iters)} ns"
 
 mulBench()
diff --git a/constantine/primitives/constant_time.nim b/constantine/primitives/constant_time.nim
index 2cc2e22..41bd118 100644
--- a/constantine/primitives/constant_time.nim
+++ b/constantine/primitives/constant_time.nim
@@ -227,7 +227,7 @@ template `<=`*[T: Ct](x, y: T): CTBool[T] =
 template `xor`*[T: Ct](x, y: CTBool[T]): CTBool[T] =
   CTBool[T](noteq(T(x), T(y)))
 
-template mux*[T: Ct](ctl: CTBool[T], x, y: T): T =
+func mux*[T](ctl: CTBool[T], x, y: T): T {.inline.}=
   ## Multiplexer / selector
   ## Returns x if ctl is true
   ## else returns y
@@ -237,22 +237,68 @@ template mux*[T: Ct](ctl: CTBool[T], x, y: T): T =
   # Alternatives:
   # - https://cryptocoding.net/index.php/Coding_rules
   # - https://www.cl.cam.ac.uk/~rja14/Papers/whatyouc.pdf
-  #
-  # TODO: assembly fastpath for conditional mov
-  let # Templates duplicate input params code
-    x_Mux = x
-    y_Mux = y
-  y_Mux xor (-T(ctl) and (x_Mux xor y_Mux))
+  when defined(amd64) or defined(i386):
+    when sizeof(T) == 8:
+      var muxed = x
+      asm """
+        testq %[ctl], %[ctl]
+        cmovzq %[y], %[muxed]
+        : [muxed] "+r" (`muxed`)
+        : [ctl] "r" (`ctl`), [y] "r" (`y`)
+        : "cc"
+      """
+      muxed
+    elif sizeof(T) == 4:
+      var muxed = x
+      asm """
+        testl %[ctl], %[ctl]
+        cmovzl %[y], %[muxed]
+        : [muxed] "+r" (`muxed`)
+        : [ctl] "r" (`ctl`), [y] "r" (`y`)
+        : "cc"
+      """
+      muxed
+    else:
+      {.error: "Unsupported word size".}
+  else:
+    let # Templates duplicate input params code
+      x_Mux = x
+      y_Mux = y
+    y_Mux xor (-T(ctl) and (x_Mux xor y_Mux))
 
-template mux*[T: CTBool](ctl: CTBool, x, y: T): T =
+func mux*[T: CTBool](ctl: CTBool, x, y: T): T {.inline.}=
   ## Multiplexer / selector
   ## Returns x if ctl is true
   ## else returns y
   ## So equivalent to ctl? x: y
-  let # Templates duplicate input params code
-    x_Mux = x
-    y_Mux = y
-  T(T.T(y_Mux) xor (-T.T(ctl) and T.T(x_Mux xor y_Mux)))
+  when defined(amd64) or defined(i386):
+    when sizeof(T) == 8:
+      var muxed = x
+      asm """
+        testq %[ctl], %[ctl]
+        cmovzq %[y], %[muxed]
+        : [muxed] "+r" (`muxed`)
+        : [ctl] "r" (`ctl`), [y] "r" (`y`)
+        : "cc"
+      """
+      muxed
+    elif sizeof(T) == 4:
+      var muxed = x
+      asm """
+        testl %[ctl], %[ctl]
+        cmovzl %[y], %[muxed]
+        : [muxed] "+r" (`muxed`)
+        : [ctl] "r" (`ctl`), [y] "r" (`y`)
+        : "cc"
+      """
+      muxed
+    else:
+      {.error: "Unsupported word size".}
+  else:
+    let # Templates duplicate input params code
+      x_Mux = x
+      y_Mux = y
+    T(T.T(y_Mux) xor (-T.T(ctl) and T.T(x_Mux xor y_Mux)))
 
 # ############################################################
 #