more comprehensive benchmark of Fp

benchmarks/bls12_381_fp.nim

@@ -29,14 +29,18 @@ import
   ./timers
 
 const Iters = 1_000_000
+const InvIters = 1000
 
 randomize(1234)
 
 echo "\n⚠️ Measurements are approximate and use the CPU nominal clock: Turbo-Boost and overclocking will skew them."
 echo "==========================================================================================================\n"
 
+proc report(op, field: string, start, stop: MonoTime, startClk, stopClk: int64, iters: int) =
+  echo &"{op:<15} {field:<15} {inNanoseconds((stop-start) div iters):>9} ns {(stopClk - startClk) div iters:>9} cycles"
+
 proc addBench() =
-  var r, x, y: Fp[BLS12_381]
+  var x, y: Fp[BLS12_381]
   # BN254 field modulus
   x.fromHex("0x30644e72e131a029b85045b68181585d97816a916871ca8d3c208c16d87cfd47")
   # BLS12-381 prime - 2
@@ -48,13 +52,43 @@ proc addBench() =
     x += y
   let stopClk = getTicks()
   let stop = getMonotime()
+  report("Addition", "Fp[BLS12_381]", start, stop, startClk, stopClk, Iters)
 
-  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"
-  echo &"Cycles per addition 𝔽p ==> {(stopClk - startClk) div Iters} cycles"
 
 addBench()
 
+proc subBench() =
+  var x, y: Fp[BLS12_381]
+  # BN254 field modulus
+  x.fromHex("0x30644e72e131a029b85045b68181585d97816a916871ca8d3c208c16d87cfd47")
+  # BLS12-381 prime - 2
+  y.fromHex("0x1a0111ea397fe69a4b1ba7b6434bacd764774b84f38512bf6730d2a0f6b0f6241eabfffeb153ffffb9feffffffffaaa9")
+
+  let start = getMonotime()
+  let startClk = getTicks()
+  for _ in 0 ..< Iters:
+    x -= y
+  let stopClk = getTicks()
+  let stop = getMonotime()
+  report("Substraction", "Fp[BLS12_381]", start, stop, startClk, stopClk, Iters)
+
+# subBench()
+
+proc negBench() =
+  var r, x: Fp[BLS12_381]
+  # BN254 field modulus
+  x.fromHex("0x30644e72e131a029b85045b68181585d97816a916871ca8d3c208c16d87cfd47")
+
+  let start = getMonotime()
+  let startClk = getTicks()
+  for _ in 0 ..< Iters:
+    r.neg(x)
+  let stopClk = getTicks()
+  let stop = getMonotime()
+  report("Negation", "Fp[BLS12_381]", start, stop, startClk, stopClk, Iters)
+
+negBench()
+
 proc mulBench() =
   var r, x, y: Fp[BLS12_381]
   # BN254 field modulus
@@ -68,9 +102,39 @@ proc mulBench() =
     r.prod(x, y)
   let stopClk = getTicks()
   let stop = getMonotime()
-
-  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"
-  echo &"Cycles per multiplication 𝔽p ==> {(stopClk - startClk) div Iters} cycles"
+  report("Multiplication", "Fp[BLS12_381]", start, stop, startClk, stopClk, Iters)
 
 mulBench()
+
+proc sqrBench() =
+  var r, x: Fp[BLS12_381]
+  # BN254 field modulus
+  x.fromHex("0x30644e72e131a029b85045b68181585d97816a916871ca8d3c208c16d87cfd47")
+
+  let start = getMonotime()
+  let startClk = getTicks()
+  for _ in 0 ..< Iters:
+    r.square(x)
+  let stopClk = getTicks()
+  let stop = getMonotime()
+  report("Squaring", "Fp[BLS12_381]", start, stop, startClk, stopClk, Iters)
+
+sqrBench()
+
+proc invBench() =
+  # TODO: having x on the stack triggers stack smashing detection. To be investigated
+  var x: ref Fp[BLS12_381]
+  new x
+  # BN254 field modulus
+  x[].fromHex("0x30644e72e131a029b85045b68181585d97816a916871ca8d3c208c16d87cfd47")
+
+  let start = getMonotime()
+  let startClk = getTicks()
+  for _ in 0 ..< InvIters:
+    # Note: we don't copy the original x so x is alterning between x and x^-1
+    inv(x[])
+  let stopClk = getTicks()
+  let stop = getMonotime()
+  report("Inversion", "Fp[BLS12_381]", start, stop, startClk, stopClk, InvIters)
+
+invBench()