chore(ci): properly set os and architecture for nightly and release (#1780)

.github/workflows/nightly.yml

@@ -18,6 +18,11 @@ jobs:
       matrix:
         os: [ubuntu-latest, macos-latest]
         experimental: [true, false]
+        arch: [amd64]
+        include:
+        - os: macos-latest
+          experimental: false
+          arch: arm64
     runs-on: ${{ matrix.os }}
     steps:
       - name: Checkout code
@@ -34,20 +39,18 @@ jobs:
           path: |
             vendor/
             .git/modules
-          key: ${{ runner.os }}-vendor-modules-${{ steps.submodules.outputs.hash }}
+          key: ${{ runner.os }}-${{matrix.arch}}-submodules-${{ steps.submodules.outputs.hash }}
 
 
       - name: prep variables
         id: vars
         run: |
-          ARCH=$(uname -m) 
           EXPERIMENTAL=$([[ "${{ matrix.experimental }}" == "true" ]] && echo "-experimental" || echo "")
 
-          echo "arch=${ARCH}" >> $GITHUB_OUTPUT
           echo "experimental=${EXPERIMENTAL}" >> $GITHUB_OUTPUT
 
-          NWAKU_ARTIFACT_NAME=$(echo "nwaku${EXPERIMENTAL}-${ARCH}-${{runner.os}}.tar.gz" | tr "[:upper:]" "[:lower:]")
-          NWAKU_TOOLS_ARTIFACT_NAME=$(echo "nwaku-tools${EXPERIMENTAL}-${ARCH}-${{runner.os}}.tar.gz" | tr "[:upper:]" "[:lower:]")
+          NWAKU_ARTIFACT_NAME=$(echo "nwaku${EXPERIMENTAL}-${{matrix.arch}}-${{runner.os}}.tar.gz" | tr "[:upper:]" "[:lower:]")
+          NWAKU_TOOLS_ARTIFACT_NAME=$(echo "nwaku-tools${EXPERIMENTAL}-${{matrix.arch}}-${{runner.os}}.tar.gz" | tr "[:upper:]" "[:lower:]")
 
           echo "nwaku=${NWAKU_ARTIFACT_NAME}" >> $GITHUB_OUTPUT
           echo "nwakutools=${NWAKU_TOOLS_ARTIFACT_NAME}" >> $GITHUB_OUTPUT
@@ -55,7 +58,9 @@ jobs:
       - name: build artifacts
         id: build
         run: |
-          make QUICK_AND_DIRTY_COMPILER=1 V=1 CI=false EXPERIMENTAL=${{matrix.experimental}} wakunode2 tools
+          OS=$([[ "${{runner.os}}" == "macOS" ]] && echo "macosx" || echo "linux")
+
+          make QUICK_AND_DIRTY_COMPILER=1 V=1 CI=false EXPERIMENTAL=${{matrix.experimental}} NIMFLAGS="-d:disableMarchNative --os:${OS} --cpu:${{matrix.arch}}" wakunode2 tools
 
           tar -cvzf ${{steps.vars.outputs.nwaku}} ./build/wakunode2
           tar -cvzf ${{steps.vars.outputs.nwakutools}} ./build/wakucanary ./build/networkmonitor

.github/workflows/release-assets.yml

@@ -13,12 +13,17 @@ jobs:
       matrix:
         env:
           - { NPROC: 2 }
-        platform: [ubuntu-latest, macos-latest]
-    runs-on: ${{ matrix.platform }}
+        os: [ubuntu-latest, macos-latest]
+        arch: [amd64]
+        include:
+        - os: macos-latest
+          experimental: false
+          arch: arm64
+    runs-on: ${{ matrix.os }}
     env: ${{ matrix.env }}
     timeout-minutes: 60
 
-    name: ${{ matrix.platform }} - ${{ matrix.env.NPROC }} processes
+    name: ${{ matrix.os }} - ${{ matrix.env.NPROC }} processes
 
     steps:
       - name: Checkout code
@@ -35,19 +40,28 @@ jobs:
           path: |
             vendor/
             .git/modules
-          key: ${{ runner.os }}-vendor-modules-${{ steps.submodules.outputs.hash }}
+          key: ${{ runner.os }}-${{matrix.arch}}-submodules-${{ steps.submodules.outputs.hash }}
+
+      - name: prep variables
+        id: vars
+        run: |
+          NWAKU_ARTIFACT_NAME=$(echo "nwaku${EXPERIMENTAL}-${{matrix.arch}}-${{runner.os}}.tar.gz" | tr "[:upper:]" "[:lower:]")
+
+          echo "nwaku=${NWAKU_ARTIFACT_NAME}" >> $GITHUB_OUTPUT
 
       - name: Install dependencies
         run: |
-          make -j${NPROC} NIMFLAGS="--parallelBuild:${NPROC}" V=1 update 
-          make -j${NPROC} NIMFLAGS="--parallelBuild:${NPROC}" CI=false wakunode1
-          make -j${NPROC} NIMFLAGS="--parallelBuild:${NPROC}" CI=false wakunode2
-          make -j${NPROC} NIMFLAGS="--parallelBuild:${NPROC}" CI=false chat2
-          tar -cvzf nim-waku-${{ matrix.platform }}.tar.gz ./build/
+          OS=$([[ "${{runner.os}}" == "macOS" ]] && echo "macosx" || echo "linux")
+
+          make -j${NPROC} NIMFLAGS="--parallelBuild:${NPROC} -d:disableMarchNative --os:${OS} --cpu:${{matrix.arch}}" V=1 update
+          make -j${NPROC} NIMFLAGS="--parallelBuild:${NPROC} -d:disableMarchNative --os:${OS} --cpu:${{matrix.arch}}" CI=false wakunode1
+          make -j${NPROC} NIMFLAGS="--parallelBuild:${NPROC} -d:disableMarchNative --os:${OS} --cpu:${{matrix.arch}}" CI=false wakunode2
+          make -j${NPROC} NIMFLAGS="--parallelBuild:${NPROC} -d:disableMarchNative --os:${OS} --cpu:${{matrix.arch}}" CI=false chat2
+          tar -cvzf ${{steps.vars.outputs.nwaku}} ./build/
 
       - name: Upload asset
         uses: actions/upload-artifact@v2.2.3
         with:
-          name: nim-waku-${{ matrix.platform }}.tar.gz
-          path: nim-waku-${{ matrix.platform }}.tar.gz
+          name: ${{steps.vars.outputs.nwaku}}
+          path: ${{steps.vars.outputs.nwaku}}
           if-no-files-found: error

Makefile

@@ -64,10 +64,6 @@ TARGET ?= prod
 GIT_VERSION ?= $(shell git describe --abbrev=6 --always --tags)
 NIM_PARAMS := $(NIM_PARAMS) -d:git_version=\"$(GIT_VERSION)\"
 
-## Pass CPU architecture to C compiler, use basic x86-64 instruction set by default
-ARCHITECTURE ?= "x86-64"
-NIM_PARAMS := $(NIM_PARAMS) --passC:\"-march=$(ARCHITECTURE)\"
-
 ## Heaptracker options
 HEAPTRACKER ?= 0
 HEAPTRACKER_INJECT ?= 0

config.nims

@@ -18,19 +18,34 @@ if defined(windows):
   # because these require direct manipulations of the stdout File object.
   switch("define", "chronicles_colors=off")
 
-# This helps especially for 32-bit x86, which sans SSE2 and newer instructions
-# requires quite roundabout code generation for cryptography, and other 64-bit
-# and larger arithmetic use cases, along with register starvation issues. When
-# engineering a more portable binary release, this should be tweaked but still
-# use at least -msse2 or -msse3.
+# https://github.com/status-im/nimbus-eth2/blob/stable/docs/cpu_features.md#ssse3-supplemental-sse3
+# suggests that SHA256 hashing with SSSE3 is 20% faster than without SSSE3, so
+# given its near-ubiquity in the x86 installed base, it renders a distribution
+# build more viable on an overall broader range of hardware.
+#
 if defined(disableMarchNative):
   if defined(i386) or defined(amd64):
     if defined(macosx):
+      # macOS Catalina is EOL as of 2022-09
+      # https://support.apple.com/kb/sp833
+      # "macOS Big Sur - Technical Specifications" lists current oldest
+      # supported models: MacBook (2015 or later), MacBook Air (2013 or later),
+      # MacBook Pro (Late 2013 or later), Mac mini (2014 or later), iMac (2014
+      # or later), iMac Pro (2017 or later), Mac Pro (2013 or later).
+      #
+      # These all have Haswell or newer CPUs.
+      #
+      # This ensures AVX2, AES-NI, PCLMUL, BMI1, and BMI2 instruction set support.
       switch("passC", "-march=haswell -mtune=generic")
       switch("passL", "-march=haswell -mtune=generic")
     else:
-      switch("passC", "-msse3")
-      switch("passL", "-msse3")
+      if defined(marchOptimized):
+        # https://github.com/status-im/nimbus-eth2/blob/stable/docs/cpu_features.md#bmi2--adx
+        switch("passC", "-march=broadwell -mtune=generic")
+        switch("passL", "-march=broadwell -mtune=generic")
+      else:
+        switch("passC", "-mssse3")
+        switch("passL", "-mssse3")
 elif defined(macosx) and defined(arm64):
   # Apple's Clang can't handle "-march=native" on M1: https://github.com/status-im/nimbus-eth2/issues/2758
   switch("passC", "-mcpu=apple-m1")
@@ -41,8 +56,9 @@ else:
   if defined(windows):
     # https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65782
     # ("-fno-asynchronous-unwind-tables" breaks Nim's exception raising, sometimes)
-    switch("passC", "-mno-avx512vl")
-    switch("passL", "-mno-avx512vl")
+    switch("passC", "-mno-avx512f")
+    switch("passL", "-mno-avx512f")
+
 
 --threads:on
 --opt:speed