Unified Snap

Makefile

@@ -115,13 +115,13 @@ batchdep: build/.supraseal-install
 batchdep: $(BUILD_DEPS)
 .PHONY: batchdep
 
-batch: CURIO_TAGS+= supraseal
+batch: CURIO_TAGS+= supraseal_nvme
 batch: CGO_LDFLAGS_ALLOW='.*'
 batch: batchdep batch-build
 .PHONY: batch
 
 
-batch-calibnet: CURIO_TAGS+= supraseal
+batch-calibnet: CURIO_TAGS+= supraseal_nvme
 batch-calibnet: CURIO_TAGS+= calibnet
 batch-calibnet: CGO_LDFLAGS_ALLOW='.*'
 batch-calibnet: batchdep batch-build

cmd/curio/batch.go

@@ -0,0 +1,69 @@
+package main
+
+import (
+	"fmt"
+
+	"github.com/urfave/cli/v2"
+	"golang.org/x/xerrors"
+
+	"github.com/filecoin-project/curio/cmd/curio/internal/translations"
+	"github.com/filecoin-project/curio/lib/supraffi"
+)
+
+var batchCmd = &cli.Command{
+	Name:  "batch",
+	Usage: translations.T("Manage batch sealing operations"),
+	Subcommands: []*cli.Command{
+		batchSetupCmd,
+	},
+}
+
+var batchSetupCmd = &cli.Command{
+	Name:  "setup",
+	Usage: translations.T("Setup SPDK for batch sealing (configures hugepages and binds NVMe devices)"),
+	Description: translations.T(`Setup SPDK for batch sealing operations.
+
+This command automatically:
+- Downloads SPDK if not already available
+- Configures 1GB hugepages (36 pages minimum)
+- Binds NVMe devices for use with SupraSeal
+
+Requires root/sudo access for SPDK setup operations.`),
+	Flags: []cli.Flag{
+		&cli.IntFlag{
+			Name:  "hugepages",
+			Usage: translations.T("Number of 1GB hugepages to configure"),
+			Value: 36,
+		},
+		&cli.IntFlag{
+			Name:  "min-pages",
+			Usage: translations.T("Minimum number of hugepages required"),
+			Value: 36,
+		},
+	},
+	Action: func(cctx *cli.Context) error {
+		nrHuge := cctx.Int("hugepages")
+		minPages := cctx.Int("min-pages")
+
+		fmt.Println("Setting up SPDK for batch sealing...")
+		fmt.Printf("Configuring %d hugepages (minimum required: %d)\n", nrHuge, minPages)
+
+		err := supraffi.CheckAndSetupSPDK(nrHuge, minPages)
+		if err != nil {
+			return xerrors.Errorf("SPDK setup failed: %w\n\n"+
+				"Please ensure you have:\n"+
+				"1. Root/sudo access for SPDK setup\n"+
+				"2. Raw NVMe devices available (no filesystems on them)\n"+
+				"3. Sufficient hugepages configured (see documentation)", err)
+		}
+
+		fmt.Println("✓ SPDK setup completed successfully")
+		fmt.Println("\nNext steps:")
+		fmt.Println("1. Verify hugepages: cat /proc/meminfo | grep Huge")
+		fmt.Println("2. Configure your batch sealing layer (see documentation)")
+		fmt.Println("3. Start batch sealing operations")
+
+		return nil
+	},
+}
+

cmd/curio/main.go

@@ -72,6 +72,7 @@ func main() {
 		ffiCmd,
 		calcCmd,
 		toolboxCmd,
+		batchCmd,
 	}
 
 	jaeger := tracing.SetupJaegerTracing("curio")

documentation/en/supraseal.md

@@ -166,13 +166,33 @@ The build should be run on the target machine. Binaries won't be portable betwee
 
 ### Setup NVMe devices for SPDK:
 
-{% hint style="info" %}
-This is only needed while batch sealing is in beta, future versions of Curio will handle this automatically.
+{% hint style="success" %}
+SPDK setup can be done automatically using the Curio CLI command:
 {% endhint %}
 
+```bash
+sudo curio batch setup
+```
+
+This command will:
+- Download SPDK if not already available
+- Configure 1GB hugepages (36 pages by default)
+- Bind NVMe devices for use with SupraSeal
+
+You can customize the number of hugepages:
+
+```bash
+sudo curio batch setup --hugepages 36 --min-pages 36
+```
+
+Alternatively, if you need to manually check SPDK status or unbind devices, you can use:
+
 ```bash
 cd extern/supraseal/deps/spdk-v24.05/
-env NRHUGE=36 ./scripts/setup.sh
+# Check status
+sudo ./scripts/setup.sh status
+# Manually run setup (not normally needed)
+sudo env NRHUGE=36 ./scripts/setup.sh
 ```
 
 ### Benchmark NVME IOPS

documentation/zh/supraseal.md

@@ -307,13 +307,33 @@ Hugepagesize:    1048576 kB
 ### Setup NVMe devices for SPDK:
 ### 为SPDK设置NVMe设备：
 
-{% hint style="info" %}
-这只在批量密封处于测试阶段时需要，Curio的未来版本将自动处理这个问题。
+{% hint style="success" %}
+可以使用 Curio CLI 命令自动完成 SPDK 设置：
 {% endhint %}
 
+```bash
+sudo curio batch setup
+```
+
+此命令将：
+- 如果尚未可用，则下载 SPDK
+- 配置 1GB 大页面（默认 36 页）
+- 绑定 NVMe 设备以供 SupraSeal 使用
+
+您可以自定义大页面数量：
+
+```bash
+sudo curio batch setup --hugepages 36 --min-pages 36
+```
+
+或者，如果您需要手动检查 SPDK 状态或解绑设备，可以使用：
+
 ```bash
 cd extern/supraseal/deps/spdk-v24.05/
-env NRHUGE=36 ./scripts/setup.sh
+# 检查状态
+sudo ./scripts/setup.sh status
+# 手动运行设置（通常不需要）
+sudo env NRHUGE=36 ./scripts/setup.sh
 ```
 
 

extern/supraseal/README.md

@@ -225,7 +225,7 @@ During the build process it will clone and build SPDK, sppark, and blst.
 ./build.sh
 ```
 
-SPDK must be setup after every reboot:
+SPDK setup is automatic when using Curio's batch sealer. 
 ```
 cd deps/spdk-v24.05
 sudo env NRHUGE=36 ./scripts/setup.sh

extern/supraseal/build.sh

@@ -144,7 +144,9 @@ CXXSTD=`$CXX -dM -E -x c++ /dev/null | \
 
 INCLUDE="-I$SPDK/include -I$SPDK/isa-l/.. -I$SPDK/dpdk/build/include"
 CFLAGS="$SECTOR_SIZE $INCLUDE -g -O2"
-CXXFLAGS="$CFLAGS -march=native $CXXSTD \
+# Use x86-64-v3 CPU flags to match GOAMD64=v3 requirement (equivalent to Intel Haswell/AMD Excavator+)
+# GCC multiversioning will provide CPU-specific optimizations at runtime
+CXXFLAGS="$CFLAGS -march=x86-64-v3 -mtune=generic $CXXSTD \
           -fPIC -fno-omit-frame-pointer -fno-strict-aliasing \
           -fstack-protector -fno-common \
           -D_GNU_SOURCE -U_FORTIFY_SOURCE -D_FORTIFY_SOURCE=2 \
@@ -275,33 +277,38 @@ if [ ! -d "deps/blst" ]; then
     git clone https://github.com/supranational/blst.git deps/blst
     (cd deps/blst
     git checkout bef14ca512ea575aff6f661fdad794263938795d
-     ./build.sh -march=native)
+     ./build.sh -march=x86-64-v3)
 fi
 
-$CC -c sha/sha_ext_mbx2.S -o obj/sha_ext_mbx2.o
-
-# Generate .h files for the Poseidon constants
-xxd -i poseidon/constants/constants_2  > obj/constants_2.h
-xxd -i poseidon/constants/constants_4  > obj/constants_4.h
-xxd -i poseidon/constants/constants_8  > obj/constants_8.h
-xxd -i poseidon/constants/constants_11 > obj/constants_11.h
-xxd -i poseidon/constants/constants_16 > obj/constants_16.h
-xxd -i poseidon/constants/constants_24 > obj/constants_24.h
-xxd -i poseidon/constants/constants_36 > obj/constants_36.h
+# Generate .h files for the Poseidon constants (needed for tree_r binaries)
+# These are fast and can run in parallel, but must complete before tree_r binaries compile
+xxd -i poseidon/constants/constants_2  > obj/constants_2.h &
+xxd -i poseidon/constants/constants_4  > obj/constants_4.h &
+xxd -i poseidon/constants/constants_8  > obj/constants_8.h &
+xxd -i poseidon/constants/constants_11 > obj/constants_11.h &
+xxd -i poseidon/constants/constants_16 > obj/constants_16.h &
+xxd -i poseidon/constants/constants_24 > obj/constants_24.h &
+xxd -i poseidon/constants/constants_36 > obj/constants_36.h &
+
+# Compile all object files in parallel - these are independent of constants headers
+# SHA extension code with x86-64-v3 flags to match GOAMD64=v3
+# The code uses SHA-NI instructions which are available on Intel Haswell+ and AMD Zen+
+# Runtime will detect CPU capabilities and use appropriate version
+$CC -c -march=x86-64-v3 -mtune=generic sha/sha_ext_mbx2.S -o obj/sha_ext_mbx2.o &
+# Compile multiversion wrapper for CPU-specific optimizations
+$CXX $CXXFLAGS -Ideps/blst/src -c sha/sha_ext_mbx2_wrapper.cpp -o obj/sha_ext_mbx2_wrapper.o &
 
 # PC1
 $CXX $CXXFLAGS -Ideps/sppark/util -o obj/pc1.o -c pc1/pc1.cpp &
 
-# PC2
+# PC2 - compile once with unified interface supporting both NVMe and FileReader
 $CXX $CXXFLAGS -o obj/streaming_node_reader_nvme.o -c nvme/streaming_node_reader_nvme.cpp &
 $CXX $CXXFLAGS -o obj/ring_t.o -c nvme/ring_t.cpp &
-$NVCC $CFLAGS $CUDA_ARCH -std=c++17 -DNO_SPDK -Xcompiler -march=native \
+# Single compilation of pc2.cu - works with both reader types via template interface
+$NVCC $CFLAGS $CUDA_ARCH -std=c++17 -DNO_SPDK \
+      -Xcompiler -march=x86-64-v3,-mtune=generic \
       -Xcompiler -Wall,-Wextra,-Wno-subobject-linkage,-Wno-unused-parameter \
       -Ideps/sppark -Ideps/sppark/util -Ideps/blst/src -c pc2/cuda/pc2.cu -o obj/pc2.o &
-# File-reader variant of pc2 for tree_r_file
-$NVCC $CFLAGS $CUDA_ARCH -std=c++17 -DNO_SPDK -DSTREAMING_NODE_READER_FILES -DRENAME_PC2_HASH_FILES -Xcompiler -march=native \
-      -Xcompiler -Wall,-Wextra,-Wno-subobject-linkage,-Wno-unused-parameter \
-      -Ideps/sppark -Ideps/sppark/util -Ideps/blst/src -c pc2/cuda/pc2.cu -o obj/pc2_files.o &
 
 $CXX $CXXFLAGS $INCLUDE -Iposeidon -Ideps/sppark -Ideps/sppark/util -Ideps/blst/src \
     -c sealing/supra_seal.cpp -o obj/supra_seal.o -Wno-subobject-linkage &
@@ -311,74 +318,48 @@ $CXX $CXXFLAGS $INCLUDE -DSTREAMING_NODE_READER_FILES -Iposeidon -Ideps/sppark -
 
 wait
 
-# Sppark object dedupe
-nm obj/pc2.o | grep -E 'select_gpu|all_gpus|cuda_available|gpu_props|ngpus|drop_gpu_ptr_t|clone_gpu_ptr_t' | awk '{print $3 " supra_" $3}' > symbol_rename.txt
-nm obj/pc2_files.o | grep -E 'select_gpu|all_gpus|cuda_available|gpu_props|ngpus|drop_gpu_ptr_t|clone_gpu_ptr_t' | awk '{print $3 " supra_" $3}' >> symbol_rename.txt
-# Deduplicate symbol rename entries
-sort -u -o symbol_rename.txt symbol_rename.txt
-
-for obj in obj/pc1.o obj/pc2.o obj/pc2_files.o obj/ring_t.o obj/streaming_node_reader_nvme.o obj/supra_seal.o obj/supra_tree_r_file.o obj/sha_ext_mbx2.o; do
-  objcopy --redefine-syms=symbol_rename.txt $obj
-done
-
-# Weaken duplicate symbols between pc2.o and pc2_files.o to avoid multiple-definition at link time
-nm -g --defined-only obj/pc2.o | awk '{print $3}' | sort -u > obj/syms_pc2.txt
-nm -g --defined-only obj/pc2_files.o | awk '{print $3}' | sort -u > obj/syms_pc2_files.txt
-comm -12 obj/syms_pc2.txt obj/syms_pc2_files.txt | grep -v '^pc2_hash_files' > obj/syms_dups.txt
-if [ -s obj/syms_dups.txt ]; then
-  while read -r sym; do
-    objcopy --weaken-symbol="$sym" obj/pc2_files.o
-  done < obj/syms_dups.txt
-fi
-
-rm symbol_rename.txt
-
+# All object files and constants headers are now ready
 ar rvs obj/libsupraseal.a \
    obj/pc1.o \
    obj/pc2.o \
-   obj/pc2_files.o \
    obj/ring_t.o \
    obj/streaming_node_reader_nvme.o \
    obj/supra_seal.o \
    obj/supra_tree_r_file.o \
-   obj/sha_ext_mbx2.o
+   obj/sha_ext_mbx2.o \
+   obj/sha_ext_mbx2_wrapper.o
 
+# Build binaries in parallel
+# Note: tree_r binaries compile poseidon.cpp which includes constants_*.h files
+# These are guaranteed to be ready after the wait above
 $CXX $CXXFLAGS -Ideps/sppark -Ideps/sppark/util -Ideps/blst/src \
     -o bin/seal demos/main.cpp \
     -Lobj -lsupraseal \
     $LDFLAGS -Ldeps/blst -lblst -L$CUDA/lib64 -lcudart_static -lgmp -lconfig++ &
 
-# tree-r CPU only
-$CXX $SECTOR_SIZE $CXXSTD -pthread -g -O3 -march=native \
+# tree-r CPU only (uses poseidon.cpp which includes constants_*.h)
+$CXX $SECTOR_SIZE $CXXSTD -pthread -g -O3 -march=x86-64-v3 -mtune=generic \
     -Wall -Wextra -Werror -Wno-subobject-linkage \
     tools/tree_r.cpp poseidon/poseidon.cpp \
     -o bin/tree_r_cpu -Iposeidon -Ideps/sppark -Ideps/blst/src -L deps/blst -lblst &
 
 # tree-r CPU + GPU
 $NVCC $SECTOR_SIZE -DNO_SPDK -DSTREAMING_NODE_READER_FILES \
-     $CUDA_ARCH -std=c++17 -g -O3 -Xcompiler -march=native \
+     $CUDA_ARCH -std=c++17 -g -O3 -Xcompiler -march=x86-64-v3,-mtune=generic \
      -Xcompiler -Wall,-Wextra,-Werror \
      -Xcompiler -Wno-subobject-linkage,-Wno-unused-parameter \
      -x cu tools/tree_r.cpp -o bin/tree_r \
      -Iposeidon -Ideps/sppark -Ideps/sppark/util -Ideps/blst/src -L deps/blst -lblst -lconfig++ &
 
 # tree-d CPU only
-$CXX -DRUNTIME_SECTOR_SIZE $CXXSTD -g -O3 -march=native \
+$CXX -DRUNTIME_SECTOR_SIZE $CXXSTD -g -O3 -march=x86-64-v3 -mtune=generic \
     -Wall -Wextra -Werror -Wno-subobject-linkage \
     tools/tree_d.cpp \
     -o bin/tree_d_cpu -Ipc1 -L deps/blst -lblst &
 
 # Standalone GPU pc2
 $NVCC $SECTOR_SIZE -DNO_SPDK -DSTREAMING_NODE_READER_FILES \
-     $CUDA_ARCH -std=c++17 -g -O3 -Xcompiler -march=native \
-     -Xcompiler -Wall,-Wextra,-Werror \
-     -Xcompiler -Wno-subobject-linkage,-Wno-unused-parameter \
-     -x cu tools/tree_r.cpp -o bin/tree_r \
-     -Iposeidon -Ideps/sppark -Ideps/sppark/util -Ideps/blst/src -L deps/blst -lblst -lconfig++ &
-
-# Standalone GPU pc2
-$NVCC $SECTOR_SIZE -DNO_SPDK -DSTREAMING_NODE_READER_FILES \
-     $CUDA_ARCH -std=c++17 -g -O3 -Xcompiler -march=native \
+     $CUDA_ARCH -std=c++17 -g -O3 -Xcompiler -march=x86-64-v3,-mtune=generic \
      -Xcompiler -Wall,-Wextra,-Werror \
      -Xcompiler -Wno-subobject-linkage,-Wno-unused-parameter \
      -x cu tools/pc2.cu -o bin/pc2 \