bench: add serial MSM benchmark

benchmarks/bls12381_msm_g1.nim

@@ -0,0 +1,81 @@
+# Nim-BLSCurve
+# Copyright (c) 2024 Status Research & Development GmbH
+# Licensed under either of
+#  * Apache License, version 2.0, ([LICENSE-APACHE](LICENSE-APACHE))
+#  * MIT license ([LICENSE-MIT](LICENSE-MIT))
+# at your option.
+# This file may not be copied, modified, or distributed except according to
+# those terms.
+
+import
+  system/ansi_c,
+  std/random,
+  ../blscurve,
+  ./bench_templates
+
+when BLS_BACKEND == BLST:
+  import
+    ../blscurve/blst/blst_abi
+
+var benchRNG = initRand(0xFACADE)
+
+proc benchECmsmG1*(numPoints, iters: int) =
+  var ps = newSeq[cblst_p1_affine](numPoints)
+  var coefs = newSeq[cblst_scalar](numPoints)
+  for i in 0 ..< ps.len:
+    # Fill with generator, BLST handles doubling without branching
+    var t: cblst_p1
+    t.addr.blst_p1_from_affine(toCC(BLS12_381_G1, cblst_p1_affine))
+
+    var scal{.noinit.}: array[32, byte]
+    for val in scal.mitems:
+      val = byte benchRNG.rand(0xFF)
+
+    # 1. Randomize the input point to avoid triggering the "double" path
+    var scalar{.noinit.}: cblst_scalar
+    scalar.addr.blst_scalar_from_bendian(scal)
+    t.addr.blst_p1_mult(t.addr, toCC(scalar, byte), 96) # go fast, 96 should be enough
+    ps[i].addr.blst_p1_to_affine(t.addr)
+
+    # 2. Randomize the coefficients
+    for val in scal.mitems:
+      val = byte benchRNG.rand(0xFF)
+
+    coefs[i].addr.blst_scalar_from_bendian(scal)
+
+  var r{.noinit.}: cblst_p1
+
+  proc msm(r: var cblst_p1, coefs: seq[cblst_scalar], ps: seq[cblst_p1_affine]) =
+    # Ensure no Nim GC interference by using c_malloc
+    let scratch = cast[ptr limb_t](c_malloc(csize_t blst_p1s_mult_pippenger_scratch_sizeof(ps.len.uint)))
+
+    let ps0 = ps[0].unsafeAddr
+    let ps00 = [ps0, nil] # Weird API with double indirection
+    let coefs0 = toCC(coefs[0], byte)
+    let coefs00 = [coefs0, nil]
+
+    blst_p1s_mult_pippenger(r.addr, ps00[0].unsafeAddr, ps.len.uint,
+                            coefs00[0].unsafeAddr, 255,
+                            scratch)
+
+
+  bench("EC MSM G1 - " & $numPoints, iters):
+    r.msm(coefs, ps)
+
+when isMainModule:
+  benchECmsmG1(8, 10)
+  benchECmsmG1(16, 10)
+  benchECmsmG1(32, 10)
+  benchECmsmG1(64, 10)
+  benchECmsmG1(128, 10)
+  benchECmsmG1(256, 10)
+  # benchECmsmG1(512, 10)
+  # benchECmsmG1(1024, 10)
+  # benchECmsmG1(2048, 1)
+  # benchECmsmG1(4096, 1)
+  # benchECmsmG1(8192, 1)
+  # benchECmsmG1(16384, 1)
+  # benchECmsmG1(32768, 1)
+  benchECmsmG1(65536, 1)
+  # benchECmsmG1(131071, 1)
+  # benchECmsmG1(262144, 1)

blscurve.nimble

@@ -73,5 +73,9 @@ task test, "Run all tests":
       run "--threads:on -d:BLS_FORCE_BACKEND=blst -d:danger --warnings:off",
           "benchmarks/bench_all.nim"
 
+      # Build but don't run MSM bench, it's slow
+      build "-d:BLS_FORCE_BACKEND=blst -d:danger --warnings:off",
+          "benchmarks/bls12381_msm_g1.nim"
+
 task bench, "Run benchmarks":
   run "--threads:on -d:danger --warnings:off", "benchmarks/bench_all.nim"