Update gotch to pytorch 2.10.0

Makefile

@@ -0,0 +1,151 @@
+# Makefile for gotch - PyTorch 2.10.0 Go bindings
+# Architecture-specific configuration is loaded from arch/ directory
+
+SHELL := /bin/bash
+
+# Detect OS and architecture
+OS    := $(shell uname -s | tr '[:upper:]' '[:lower:]')
+ARCH  := $(shell uname -m)
+
+# Normalize architecture names
+ifeq ($(ARCH),x86_64)
+	ARCH := amd64
+else ifeq ($(ARCH),aarch64)
+	ARCH := arm64
+endif
+
+# Allow override of architecture configuration
+# Examples:
+#   make ARCH_CONFIG=linux-amd64-cuda build
+#   make ARCH_CONFIG=darwin-arm64 test
+ARCH_CONFIG ?= $(OS)-$(ARCH)
+
+# Include architecture-specific configuration
+ARCH_FILE := arch/$(ARCH_CONFIG).mk
+ifeq ($(wildcard $(ARCH_FILE)),)
+	$(error Architecture config file not found: $(ARCH_FILE). Available: $(wildcard arch/*.mk))
+endif
+include $(ARCH_FILE)
+
+# Go test flags
+TEST_FLAGS := -v
+TEST_TIMEOUT := 5m
+
+.PHONY: all build test test-nn test-ts clean help ffi-validate
+
+# Default target
+all: build
+
+# Build all core packages
+build:
+	@echo "Building gotch core packages..."
+	@go build -v . ./ts ./nn ./vision
+
+# Run all tests
+test: test-nn test-ts ffi-validate
+	@echo "All tests completed"
+
+# Run nn package tests
+# Running with -p 1 to force sequential execution (PyTorch 2.10.0 thread-local gradient state)
+test-nn:
+	@echo "Running nn package tests..."
+	@go test $(TEST_FLAGS) -timeout $(TEST_TIMEOUT) -p 1 -parallel 1 ./nn
+
+# Run ts package tests
+test-ts:
+	@echo "Running ts package tests..."
+	@go test $(TEST_FLAGS) -timeout $(TEST_TIMEOUT) ./ts
+
+# Run specific test in nn package
+# Usage: make test-nn-specific TEST=TestInitTensor_Memcheck
+test-nn-specific:
+	@echo "Running specific nn test: $(TEST)..."
+	@go test $(TEST_FLAGS) -timeout $(TEST_TIMEOUT) -run $(TEST) ./nn
+
+# Run specific test in ts package
+# Usage: make test-ts-specific TEST=TestTensor
+test-ts-specific:
+	@echo "Running specific ts test: $(TEST)..."
+	@go test $(TEST_FLAGS) -timeout $(TEST_TIMEOUT) -run $(TEST) ./ts
+
+# Run tests with coverage
+test-nn-coverage:
+	@echo "Running nn tests with coverage..."
+	@go test -v -timeout $(TEST_TIMEOUT) -coverprofile=coverage-nn.out ./nn
+	@go tool cover -html=coverage-nn.out -o coverage-nn.html
+	@echo "Coverage report saved to coverage-nn.html"
+
+test-ts-coverage:
+	@echo "Running ts tests with coverage..."
+	@go test -v -timeout $(TEST_TIMEOUT) -coverprofile=coverage-ts.out ./ts
+	@go tool cover -html=coverage-ts.out -o coverage-ts.html
+	@echo "Coverage report saved to coverage-ts.html"
+
+# Clean build artifacts
+clean:
+	@echo "Cleaning build artifacts..."
+	@go clean -cache -testcache
+	@rm -f coverage-*.out coverage-*.html
+
+# Display build environment
+env:
+	@echo "Build Environment:"
+	@echo "  Platform:           $(PLATFORM_DESC)"
+	@echo "  Arch Config:        $(ARCH_CONFIG) ($(ARCH_FILE))"
+	@echo "  LIBTORCH_PATH:      $(LIBTORCH_PATH)"
+	@echo "  $(RUNTIME_LIB_VAR): $($(RUNTIME_LIB_VAR))"
+	@echo ""
+	@echo "CGO Flags:"
+	@echo "  CGO_CFLAGS:         $(CGO_CFLAGS)"
+	@echo "  CGO_LDFLAGS:        $(CGO_LDFLAGS)"
+	@echo "  CGO_CXXFLAGS:       $(CGO_CXXFLAGS)"
+	@echo ""
+	@echo "Go version:"
+	@go version
+	@echo ""
+	@echo "LibTorch version: 2.10.0"
+
+# Check if MPS is available
+check-mps:
+	@echo "Checking MPS availability..."
+	@go run -exec 'env DYLD_LIBRARY_PATH=$(DYLD_LIBRARY_PATH)' tools/check_device.go || echo "Create tools/check_device.go to test device availability"
+
+# Validate FFI type conversions
+ffi-validate:
+	@echo "Validating FFI type conversions..."
+	@go run tools/ffi-validation/main.go
+
+# Help target
+help:
+	@echo "Gotch Makefile - PyTorch 2.10.0 Go Bindings"
+	@echo ""
+	@echo "Current Configuration:"
+	@echo "  Platform:      $(PLATFORM_DESC)"
+	@echo "  Arch Config:   $(ARCH_CONFIG)"
+	@echo "  LibTorch:      $(LIBTORCH_PATH)"
+	@echo ""
+	@echo "Targets:"
+	@echo "  make build              - Build all core packages"
+	@echo "  make test               - Run all tests (nn + ts)"
+	@echo "  make test-nn            - Run nn package tests"
+	@echo "  make test-ts            - Run ts package tests"
+	@echo "  make test-nn-specific   - Run specific nn test (TEST=TestName)"
+	@echo "  make test-ts-specific   - Run specific ts test (TEST=TestName)"
+	@echo "  make test-nn-coverage   - Run nn tests with coverage report"
+	@echo "  make test-ts-coverage   - Run ts tests with coverage report"
+	@echo "  make clean              - Clean build artifacts and caches"
+	@echo "  make env                - Display build environment"
+	@echo "  make check-mps          - Check MPS device availability"
+	@echo "  make ffi-validate       - Validate FFI type conversions (C <-> Go)"
+	@echo "  make help               - Show this help message"
+	@echo ""
+	@echo "Architecture Configuration:"
+	@echo "  Default: auto-detected (current: $(ARCH_CONFIG))"
+	@echo "  Override: make ARCH_CONFIG=linux-amd64-cuda build"
+	@echo "  Available configs: $(notdir $(basename $(wildcard arch/*.mk)))"
+	@echo ""
+	@echo "Examples:"
+	@echo "  make test-nn"
+	@echo "  make test-nn-specific TEST=TestInitTensor_Memcheck"
+	@echo "  make ARCH_CONFIG=linux-amd64-cuda build"
+	@echo "  LIBTORCH_PATH=/custom/path make build"

README.md

@@ -20,13 +20,13 @@
 
 ## Dependencies
 
-- **Libtorch** C++ v2.1.0 library of [Pytorch](https://pytorch.org/)
+- **Libtorch** C++ v2.10.0 library of [Pytorch](https://pytorch.org/)
 - Clang-17/Clang++-17 compilers
 
 ## Installation
 
 - Default CUDA version is `11.8` if CUDA is available otherwise using CPU version.
-- Default Pytorch C++ API version is `2.1.0`
+- Default Pytorch C++ API version is `2.10.0`
 
 **NOTE**: `libtorch` will be installed at **`/usr/local/lib`**
 
@@ -266,6 +266,20 @@ func main() {
 
 - See [pkg.go.dev](https://pkg.go.dev/github.com/sugarme/gotch?tab=doc) for APIs detail.
 
+For unit tests use:
+- make test
+
+## PyTorch 2.10.0 Upgrade Notes
+
+This version includes critical fixes for PyTorch 2.10.0 compatibility:
+- Fixed gradient state management (thread-local in PyTorch 2.10.0)
+- Fixed FFI type conversion bugs (C.int → Go int)
+
+To validate FFI conversions:
+```bash
+make ffi-validate
+```
+
 ## License
 
 `gotch` is Apache 2.0 licensed.

device.go

@@ -12,9 +12,11 @@ type Device struct {
 }
 
 type Cuda Device
+type Mps Device
 
 var (
 	CPU  Device = Device{Name: "CPU", Value: -1}
+	MPS  Mps    = Mps{Name: "MPS", Value: -2}
 	CUDA Cuda   = Cuda{Name: "CUDA", Value: 0}
 )
 
@@ -34,6 +36,14 @@ func NewCuda() Device {
 	return CudaBuilder(0)
 }
 
+// MPS methods:
+// ============
+
+// IsAvailable returns true if MPS (Metal Performance Shaders) support is available
+func (m Mps) IsAvailable() bool {
+	return lib.AtcMpsIsAvailable()
+}
+
 // Cuda methods:
 // =============
 
@@ -74,6 +84,8 @@ func (d Device) CInt() CInt {
 	switch {
 	case d.Name == "CPU":
 		return -1
+	case d.Name == "MPS":
+		return -2
 	case d.Name == "CUDA":
 		// TODO: create a function to retrieve cuda_index
 		var deviceIndex int = d.Value
@@ -87,7 +99,9 @@ func (d Device) CInt() CInt {
 func (d Device) OfCInt(v CInt) Device {
 	switch {
 	case v == -1:
-		return Device{Name: "CPU", Value: 1}
+		return Device{Name: "CPU", Value: -1}
+	case v == -2:
+		return Device{Name: "MPS", Value: -2}
 	case v >= 0:
 		return CudaBuilder(uint(v))
 	default:
@@ -124,3 +138,25 @@ func CudaIfAvailable() Device {
 		return CPU
 	}
 }
+
+// MPSIfAvailable returns an MPS device if available, else CPU.
+func MPSIfAvailable() Device {
+	switch {
+	case MPS.IsAvailable():
+		return Device{Name: "MPS", Value: -2}
+	default:
+		return CPU
+	}
+}
+
+// BestAvailableDevice returns the best available device (CUDA > MPS > CPU).
+func BestAvailableDevice() Device {
+	switch {
+	case CUDA.IsAvailable():
+		return CudaBuilder(0)
+	case MPS.IsAvailable():
+		return Device{Name: "MPS", Value: -2}
+	default:
+		return CPU
+	}
+}

gen/gen.ml

@@ -73,6 +73,7 @@ let excluded_functions =
     ; "sym_size"
     ; "sym_stride"
     ; "sym_storage_offset"
+    ; "_sparse_semi_structured_addmm"
     ]
 
 let no_tensor_options =
@@ -1441,7 +1442,7 @@ let run ~yaml_filename ~cpp_filename ~ffi_filename ~must_wrapper_filename
 
 
 let () =
-  run ~yaml_filename:"gen/pytorch/Declarations-v2.1.0.yaml"
+  run ~yaml_filename:"gen/pytorch/Declarations-v2.10.0.yaml"
     ~cpp_filename:"libtch/torch_api_generated"
     ~ffi_filename:"libtch/c-generated.go"
     ~must_wrapper_filename:"ts/must-tensor-generated.go"