Merge pull request #3111 from autonomys/improve-gpu-support

Improve GPU support
autonomys · Oct 9, 2024 · 6c0c789 · 6c0c789
2 parents 3cda331 + 8f4a907
commit 6c0c789
Show file tree

Hide file tree

Showing 10 changed files with 58 additions and 89 deletions.
diff --git a/.github/workflows/rust.yml b/.github/workflows/rust.yml
@@ -104,14 +104,13 @@ jobs:
           sub-packages: '["nvcc", "cudart"]'
         if: runner.os == 'Linux' || runner.os == 'Windows'
 
-      # TODO: ROCm compilation doesn't work in CI right now, good luck fixing it
-      # - name: Configure ROCm cache (Windows)
-      #   uses: actions/cache@0c45773b623bea8c8e75f6c82b208c3cf94ea4f9 # v4.0.2
-      #   id: rocm-cache
-      #   with:
-      #     path: C:\Program Files\AMD\ROCm
-      #     key: ${{ runner.os }}-rocm
-      #   if: runner.os == 'Windows'
+      - name: Configure ROCm cache (Windows)
+        uses: actions/cache@0c45773b623bea8c8e75f6c82b208c3cf94ea4f9 # v4.0.2
+        id: rocm-cache
+        with:
+          path: C:\Program Files\AMD\ROCm
+          key: ${{ runner.os }}-rocm
+        if: runner.os == 'Windows'
 
       - name: ROCm toolchain
         run: |
@@ -126,19 +125,13 @@ jobs:
           sudo ldconfig
         if: runner.os == 'Linux'
 
-      # TODO: ROCm compilation doesn't work in CI right now, good luck fixing it
-      # - name: ROCm toolchain
-      #   run: |
-      #     $ErrorActionPreference = "Stop"
-      #     Invoke-WebRequest -Uri https://download.amd.com/developer/eula/rocm-hub/AMD-Software-PRO-Edition-24.Q3-WinSvr2022-For-HIP.exe -OutFile "${env:RUNNER_TEMP}\HIP-SDK-Installer.exe"
-      #     Start-Process "${env:RUNNER_TEMP}\HIP-SDK-Installer.exe" -ArgumentList '-install' -NoNewWindow -Wait
-      #     Remove-Item "${env:RUNNER_TEMP}\HIP-SDK-Installer.exe"
-      #   if: runner.os == 'Windows' && steps.rocm-cache.outputs.cache-hit != 'true'
-      #
-      # - name: ROCm toolchain environment (Windows)
-      #   run: |
-      #     Add-Content $env:GITHUB_PATH "C:\Program Files\AMD\ROCm\6.1\bin"
-      #   if: runner.os == 'Windows'
+      - name: ROCm toolchain
+        run: |
+          $ErrorActionPreference = "Stop"
+          Invoke-WebRequest -Uri https://download.amd.com/developer/eula/rocm-hub/AMD-Software-PRO-Edition-24.Q3-WinSvr2022-For-HIP.exe -OutFile "${env:RUNNER_TEMP}\HIP-SDK-Installer.exe"
+          Start-Process "${env:RUNNER_TEMP}\HIP-SDK-Installer.exe" -ArgumentList '-install' -NoNewWindow -Wait
+          Remove-Item "${env:RUNNER_TEMP}\HIP-SDK-Installer.exe"
+        if: runner.os == 'Windows' && steps.rocm-cache.outputs.cache-hit != 'true'
 
       - name: Configure cache
         uses: actions/cache@0c45773b623bea8c8e75f6c82b208c3cf94ea4f9 # v4.0.2
@@ -161,20 +154,17 @@ jobs:
         if: runner.os == 'Linux' || runner.os == 'Windows'
 
       - name: cargo clippy (ROCm)
-        env:
-          NVCC: off
         run: |
           cargo -Zgitoxide -Zgit clippy --locked --all-targets --features rocm -- -D warnings
         if: runner.os == 'Linux'
 
-      # TODO: ROCm compilation doesn't work in CI right now, good luck fixing it
-      # - name: cargo clippy (ROCm)
-      #   env:
-      #     NVCC: off
-      #     HIPCC: hipcc.bin.exe
-      #   run: |
-      #     cargo -Zgitoxide -Zgit clippy --locked --all-targets --features rocm -- -D warnings
-      #   if: runner.os == 'Windows'
+      - name: cargo clippy (ROCm)
+        env:
+          # Why `PROGRA~1` instead of `Program Files`? Because Windows!
+          HIPCC: C:\PROGRA~1\AMD\ROCm\6.1\bin\hipcc.bin.exe
+        run: |
+          cargo -Zgitoxide -Zgit clippy --locked --all-targets --features rocm -- -D warnings
+        if: runner.os == 'Windows'
 
   cargo-docs:
     runs-on: ${{ fromJson(github.repository_owner == 'autonomys' && '["self-hosted", "ubuntu-20.04-x86-64"]' || '"ubuntu-22.04"') }}

diff --git a/.github/workflows/snapshot-build.yml b/.github/workflows/snapshot-build.yml
@@ -170,14 +170,13 @@ jobs:
           sub-packages: '["nvcc", "cudart"]'
         if: runner.os == 'Linux' || runner.os == 'Windows'
 
-      # TODO: ROCm compilation doesn't work in CI right now, good luck fixing it
-      # - name: Configure ROCm cache (Windows)
-      #   uses: actions/cache@0c45773b623bea8c8e75f6c82b208c3cf94ea4f9 # v4.0.2
-      #   id: rocm-cache
-      #   with:
-      #     path: C:\Program Files\AMD\ROCm
-      #     key: ${{ runner.os }}-rocm
-      #   if: runner.os == 'Windows'
+      - name: Configure ROCm cache (Windows)
+        uses: actions/cache@0c45773b623bea8c8e75f6c82b208c3cf94ea4f9 # v4.0.2
+        id: rocm-cache
+        with:
+          path: C:\Program Files\AMD\ROCm
+          key: ${{ runner.os }}-rocm
+        if: runner.os == 'Windows'
 
       - name: ROCm toolchain
         run: |
@@ -193,19 +192,13 @@ jobs:
         # TODO: ROCm packages are only available for x86-64 for now
         if: runner.os == 'Linux' && startsWith(matrix.build.target, 'x86_64')
 
-      # TODO: ROCm compilation doesn't work in CI right now, good luck fixing it
-      # - name: ROCm toolchain
-      #   run: |
-      #     $ErrorActionPreference = "Stop"
-      #     Invoke-WebRequest -Uri https://download.amd.com/developer/eula/rocm-hub/AMD-Software-PRO-Edition-24.Q3-WinSvr2022-For-HIP.exe -OutFile "${env:RUNNER_TEMP}\HIP-SDK-Installer.exe"
-      #     Start-Process "${env:RUNNER_TEMP}\HIP-SDK-Installer.exe" -ArgumentList '-install' -NoNewWindow -Wait
-      #     Remove-Item "${env:RUNNER_TEMP}\HIP-SDK-Installer.exe"
-      #   if: runner.os == 'Windows' && steps.rocm-cache.outputs.cache-hit != 'true'
-      #
-      # - name: ROCm toolchain environment (Windows)
-      #   run: |
-      #     Add-Content $env:GITHUB_PATH "C:\Program Files\AMD\ROCm\6.1\bin"
-      #   if: runner.os == 'Windows'
+      - name: ROCm toolchain
+        run: |
+          $ErrorActionPreference = "Stop"
+          Invoke-WebRequest -Uri https://download.amd.com/developer/eula/rocm-hub/AMD-Software-PRO-Edition-24.Q3-WinSvr2022-For-HIP.exe -OutFile "${env:RUNNER_TEMP}\HIP-SDK-Installer.exe"
+          Start-Process "${env:RUNNER_TEMP}\HIP-SDK-Installer.exe" -ArgumentList '-install' -NoNewWindow -Wait
+          Remove-Item "${env:RUNNER_TEMP}\HIP-SDK-Installer.exe"
+        if: runner.os == 'Windows' && steps.rocm-cache.outputs.cache-hit != 'true'
 
       - name: Configure cache
         uses: actions/cache@0c45773b623bea8c8e75f6c82b208c3cf94ea4f9 # v4.0.2
@@ -222,26 +215,21 @@ jobs:
           cargo -Zgitoxide -Zgit build --locked -Z build-std --target ${{ matrix.build.target }} --profile production --bin subspace-farmer
         if: runner.os == 'macOS' || !startsWith(matrix.build.target, 'x86_64')
 
-      # TODO: ROCm compilation doesn't work in CI right now, good luck fixing it
-      # # ROCm can't be enabled together with CUDA for now
-      # - name: Build farmer (ROCm, Windows)
-      #   env:
-      #     NVCC: off
-      #     HIPCC: hipcc.bin.exe
-      #   run: |
-      #     cargo -Zgitoxide -Zgit build --locked -Z build-std --target ${{ matrix.build.target }} --profile production --bin subspace-farmer --features rocm
-      #     move ${{ env.PRODUCTION_TARGET }}/subspace-farmer.exe ${{ env.PRODUCTION_TARGET }}/subspace-farmer-rocm.exe
-      #   # TODO: ROCm packages are only available for x86-64 for now
-      #   if: runner.os == 'Windows' && startsWith(matrix.build.target, 'x86_64')
+      # ROCm can't be enabled together with CUDA for now
+      - name: Build farmer (ROCm, Windows)
+        env:
+          # Why `PROGRA~1` instead of `Program Files`? Because Windows!
+          HIPCC: C:\PROGRA~1\AMD\ROCm\6.1\bin\hipcc.bin.exe
+        run: |
+          cargo -Zgitoxide -Zgit build --locked -Z build-std --target ${{ matrix.build.target }} --profile production --bin subspace-farmer --features rocm
+          move ${{ env.PRODUCTION_TARGET }}/subspace-farmer.exe ${{ env.PRODUCTION_TARGET }}/subspace-farmer-rocm.exe
+        if: runner.os == 'Windows' && startsWith(matrix.build.target, 'x86_64')
 
       # ROCm can't be enabled together with CUDA for now
       - name: Build farmer (ROCm, Ubuntu)
-        env:
-          NVCC: off
         run: |
           cargo -Zgitoxide -Zgit build --locked -Z build-std --target ${{ matrix.build.target }} --profile production --bin subspace-farmer --features rocm
           mv ${{ env.PRODUCTION_TARGET }}/subspace-farmer ${{ env.PRODUCTION_TARGET }}/subspace-farmer-rocm
-        # TODO: ROCm packages are only available for x86-64 for now
         if: runner.os == 'Linux' && startsWith(matrix.build.target, 'x86_64')
 
       - name: Build farmer
@@ -291,8 +279,7 @@ jobs:
       - name: Sign Application (Windows)
         run: |
           AzureSignTool sign --azure-key-vault-url "${{ secrets.AZURE_KEY_VAULT_URI }}" --azure-key-vault-client-id "${{ secrets.AZURE_CLIENT_ID }}" --azure-key-vault-client-secret "${{ secrets.AZURE_CLIENT_SECRET }}" --azure-key-vault-tenant-id "${{ secrets.AZURE_TENANT_ID }}" --azure-key-vault-certificate "${{ secrets.AZURE_CERT_NAME }}" --file-digest sha512 --timestamp-rfc3161 http://timestamp.digicert.com -v "${{ env.PRODUCTION_TARGET }}/subspace-farmer.exe"
-          # TODO: ROCm compilation doesn't work in CI right now, good luck fixing it
-          # AzureSignTool sign --azure-key-vault-url "${{ secrets.AZURE_KEY_VAULT_URI }}" --azure-key-vault-client-id "${{ secrets.AZURE_CLIENT_ID }}" --azure-key-vault-client-secret "${{ secrets.AZURE_CLIENT_SECRET }}" --azure-key-vault-tenant-id "${{ secrets.AZURE_TENANT_ID }}" --azure-key-vault-certificate "${{ secrets.AZURE_CERT_NAME }}" --file-digest sha512 --timestamp-rfc3161 http://timestamp.digicert.com -v "${{ env.PRODUCTION_TARGET }}/subspace-farmer-rocm.exe"
+          AzureSignTool sign --azure-key-vault-url "${{ secrets.AZURE_KEY_VAULT_URI }}" --azure-key-vault-client-id "${{ secrets.AZURE_CLIENT_ID }}" --azure-key-vault-client-secret "${{ secrets.AZURE_CLIENT_SECRET }}" --azure-key-vault-tenant-id "${{ secrets.AZURE_TENANT_ID }}" --azure-key-vault-certificate "${{ secrets.AZURE_CERT_NAME }}" --file-digest sha512 --timestamp-rfc3161 http://timestamp.digicert.com -v "${{ env.PRODUCTION_TARGET }}/subspace-farmer-rocm.exe"
           AzureSignTool sign --azure-key-vault-url "${{ secrets.AZURE_KEY_VAULT_URI }}" --azure-key-vault-client-id "${{ secrets.AZURE_CLIENT_ID }}" --azure-key-vault-client-secret "${{ secrets.AZURE_CLIENT_SECRET }}" --azure-key-vault-tenant-id "${{ secrets.AZURE_TENANT_ID }}" --azure-key-vault-certificate "${{ secrets.AZURE_CERT_NAME }}" --file-digest sha512 --timestamp-rfc3161 http://timestamp.digicert.com -v "${{ env.PRODUCTION_TARGET }}/subspace-node.exe"
         # Allow code signing to fail on non-release builds and in non-subspace repos (forks)
         continue-on-error: ${{ github.repository_owner != 'autonomys' || github.event_name != 'push' || github.ref_type != 'tag' }}
@@ -308,7 +295,6 @@ jobs:
       - name: Prepare executables for uploading (Ubuntu, ROCm)
         run: |
           mv ${{ env.PRODUCTION_TARGET }}/subspace-farmer-rocm executables/subspace-farmer-rocm-${{ matrix.build.suffix }}
-        # TODO: ROCm packages are only available for x86-64 for now
         if: runner.os == 'Linux' && startsWith(matrix.build.target, 'x86_64')
 
       - name: Prepare executables for uploading (macOS)
@@ -327,8 +313,7 @@ jobs:
         run: |
           mkdir executables
           move ${{ env.PRODUCTION_TARGET }}/subspace-farmer.exe executables/subspace-farmer-${{ matrix.build.suffix }}.exe
-          # TODO: ROCm compilation doesn't work in CI right now, good luck fixing it
-          # move ${{ env.PRODUCTION_TARGET }}/subspace-farmer-rocm.exe executables/subspace-farmer-rocm-${{ matrix.build.suffix }}.exe
+          move ${{ env.PRODUCTION_TARGET }}/subspace-farmer-rocm.exe executables/subspace-farmer-rocm-${{ matrix.build.suffix }}.exe
           move ${{ env.PRODUCTION_TARGET }}/subspace-node.exe executables/subspace-node-${{ matrix.build.suffix }}.exe
         if: runner.os == 'Windows'
 

diff --git a/Cargo.lock b/Cargo.lock
diff --git a/Dockerfile-bootstrap-node b/Dockerfile-bootstrap-node
@@ -65,7 +65,7 @@ RUN \
     if [ $BUILDARCH != "riscv64" ] && [ $TARGETARCH = "riscv64" ]; then \
       export RUSTFLAGS="$RUSTFLAGS -C linker=riscv64-linux-gnu-gcc" \
     ; fi && \
-    if [ $TARGETARCH = "amd64" ] && [ $RUSTFLAGS = ""]; then \
+    if [ $TARGETARCH = "amd64" ] && [ "$RUSTFLAGS" = ""]; then \
       export RUSTFLAGS="-C target-cpu=skylake" \
     ; fi && \
     RUSTC_TARGET_ARCH=$(echo $TARGETARCH | sed "s/amd64/x86_64/g" | sed "s/arm64/aarch64/g" | sed "s/riscv64/riscv64gc/g") && \

diff --git a/Dockerfile-farmer b/Dockerfile-farmer
@@ -98,7 +98,6 @@ RUN \
       ldconfig \
     ; fi
 
-# TODO: Remove `NVCC=off` hack once `sppark` has proper features for CUDA and ROCm
 # ROCm is only used on x86-64 since they don't have other packages
 RUN \
     if [ $BUILDARCH != "arm64" ] && [ $TARGETARCH = "arm64" ]; then \
@@ -107,13 +106,13 @@ RUN \
     if [ $BUILDARCH != "riscv64" ] && [ $TARGETARCH = "riscv64" ]; then \
       export RUSTFLAGS="$RUSTFLAGS -C linker=riscv64-linux-gnu-gcc" \
     ; fi && \
-    if [ $TARGETARCH = "amd64" ] && [ $RUSTFLAGS = "" ]; then \
+    if [ $TARGETARCH = "amd64" ] && [ "$RUSTFLAGS" = "" ]; then \
       export RUSTFLAGS="-C target-cpu=skylake" \
     ; fi && \
     export PATH=/usr/local/cuda/bin${PATH:+:${PATH}} && \
     RUSTC_TARGET_ARCH=$(echo $TARGETARCH | sed "s/amd64/x86_64/g" | sed "s/arm64/aarch64/g" | sed "s/riscv64/riscv64gc/g") && \
     if [ $BUILDARCH = "amd64" ] && [ $TARGETARCH = "amd64" ]; then \
-      NVCC=off /root/.cargo/bin/cargo -Zgitoxide -Zgit build \
+      /root/.cargo/bin/cargo -Zgitoxide -Zgit build \
           --locked \
           -Z build-std \
           --profile $PROFILE \

diff --git a/Dockerfile-node b/Dockerfile-node
@@ -67,7 +67,7 @@ RUN \
     if [ $BUILDARCH != "riscv64" ] && [ $TARGETARCH = "riscv64" ]; then \
       export RUSTFLAGS="$RUSTFLAGS -C linker=riscv64-linux-gnu-gcc" \
     ; fi && \
-    if [ $TARGETARCH = "amd64" ] && [ $RUSTFLAGS = ""]; then \
+    if [ $TARGETARCH = "amd64" ] && [ "$RUSTFLAGS" = ""]; then \
       export RUSTFLAGS="-C target-cpu=skylake" \
     ; fi && \
     RUSTC_TARGET_ARCH=$(echo $TARGETARCH | sed "s/amd64/x86_64/g" | sed "s/arm64/aarch64/g" | sed "s/riscv64/riscv64gc/g") && \

diff --git a/Dockerfile-runtime b/Dockerfile-runtime
@@ -65,7 +65,7 @@ RUN \
     if [ $BUILDARCH != "riscv64" ] && [ $TARGETARCH = "riscv64" ]; then \
       export RUSTFLAGS="$RUSTFLAGS -C linker=riscv64-linux-gnu-gcc" \
     ; fi && \
-    if [ $TARGETARCH = "amd64" ] && [ $RUSTFLAGS = ""]; then \
+    if [ $TARGETARCH = "amd64" ] && [ "$RUSTFLAGS" = ""]; then \
       export RUSTFLAGS="-C target-cpu=skylake" \
     ; fi && \
     RUSTC_TARGET_ARCH=$(echo $TARGETARCH | sed "s/amd64/x86_64/g" | sed "s/arm64/aarch64/g" | sed "s/riscv64/riscv64gc/g") && \

diff --git a/shared/subspace-proof-of-space-gpu/Cargo.toml b/shared/subspace-proof-of-space-gpu/Cargo.toml
@@ -15,8 +15,8 @@ include = [
 [dependencies]
 blst = { version = "0.3.13", optional = true }
 rust-kzg-blst = { git = "https://github.com/grandinetech/rust-kzg", rev = "6c8fcc623df3d7e8c0f30951a49bfea764f90bf4", default-features = false, optional = true }
-# TODO: Fork with ROCm support, switch to upstream once `rocm` branch from `https://github.com/dot-asm/sppark` is upstreamed
-sppark = { version = "0.1.8", git = "https://github.com/autonomys/sppark", rev = "71c49160d7aa24f92c20592d2d26ef16f5400a04", optional = true }
+# TODO: Fork with ROCm support, switch to upstream once `rocm` branch from `https://github.com/dot-asm/sppark` + https://github.com/dot-asm/sppark/pull/2 are upstreamed
+sppark = { version = "0.1.8", git = "https://github.com/autonomys/sppark", rev = "b2a181eb99c8200f1a604f04122551ea39fbf63f", optional = true }
 subspace-core-primitives = { version = "0.1.0", path = "../../crates/subspace-core-primitives", default-features = false, optional = true }
 subspace-kzg = { version = "0.1.0", path = "../subspace-kzg", optional = true }
 
@@ -30,15 +30,14 @@ cc = "1.1.23"
 
 [features]
 # Only Volta+ architectures are supported (GeForce RTX 16xx consumer GPUs and newer)
-cuda = ["_gpu"]
+cuda = ["_gpu", "sppark/cuda"]
 # TODO: ROCm can't be enabled at the same time as `cuda` feature at the moment
 # Seems to support RDNA 2+, at least on Linux
-rocm = ["_gpu"]
+rocm = ["_gpu", "sppark/rocm"]
 # Internal feature, shouldn't be used directly
 _gpu = [
     "dep:blst",
     "dep:rust-kzg-blst",
-    "dep:sppark",
     "dep:subspace-core-primitives",
     "dep:subspace-kzg",
 ]
diff --git a/shared/subspace-proof-of-space-gpu/README.md b/shared/subspace-proof-of-space-gpu/README.md
@@ -24,8 +24,3 @@ For other operating systems/platforms check official documentation: <https://doc
 ### ROCm
 
 For AMD/ROCm support follow their official documentation: <https://rocm.docs.amd.com/en/latest/>
-
-For compilation `NVCC=off` environment variable must be additionally used:
-```bash
-NVCC=off cargo build
-```
diff --git a/shared/subspace-proof-of-space-gpu/build.rs b/shared/subspace-proof-of-space-gpu/build.rs
@@ -18,6 +18,8 @@ fn main() {
     }
 
     if cfg!(feature = "rocm") {
+        println!("cargo::rerun-if-env-changed=HIPCC");
+
         let mut hipcc = cc::Build::new();
         hipcc.compiler(env::var("HIPCC").unwrap_or("hipcc".to_string()));
         hipcc.cpp(true);
@@ -88,5 +90,4 @@ fn main() {
     }
 
     println!("cargo::rerun-if-changed=src");
-    println!("cargo::rerun-if-env-changed=CXXFLAGS");
 }