Merge pull request #1869 from Gustav-Simonsson/gpu_miner

all: Add GPU mining, disabled by default

Godeps/Godeps.json

@@ -16,8 +16,8 @@
 		},
 		{
 			"ImportPath": "github.com/ethereum/ethash",
-			"Comment": "v23.1-234-g062e40a",
-			"Rev": "062e40a1a1671f5a5102862b56e4c56f68a732f5"
+			"Comment": "v23.1-235-gb39e007",
+			"Rev": "b39e007d393ab5945b4c0748a7415b7e31c5db04"
 		},
 		{
 			"ImportPath": "github.com/fatih/color",

Godeps/_workspace/src/github.com/Gustav-Simonsson/go-opencl/cl/cl.go

+/*
+Package cl provides a binding to the OpenCL api. It's mostly a low-level
+wrapper that avoids adding functionality while still making the interface
+a little more friendly and easy to use.
+
+Resource life-cycle management:
+
+For any CL object that gets created (buffer, queue, kernel, etc..) you should
+call object.Release() when finished with it to free the CL resources. This
+explicitely calls the related clXXXRelease method for the type. However,
+as a fallback there is a finalizer set for every resource item that takes
+care of it (eventually) if Release isn't called. In this way you can have
+better control over the life cycle of resources while having a fall back
+to avoid leaks. This is similar to how file handles and such are handled
+in the Go standard packages.
+*/
+package cl
+
+// #include "headers/1.2/opencl.h"
+// #cgo CFLAGS: -Iheaders/1.2
+// #cgo darwin LDFLAGS: -framework OpenCL
+// #cgo linux LDFLAGS: -lOpenCL
+import "C"
+import "errors"
+
+var ErrUnsupported = errors.New("cl: unsupported")

Godeps/_workspace/src/github.com/Gustav-Simonsson/go-opencl/cl/cl_test.go

+package cl
+
+import (
+	"math/rand"
+	"reflect"
+	"strings"
+	"testing"
+)
+
+var kernelSource = `
+__kernel void square(
+   __global float* input,
+   __global float* output,
+   const unsigned int count)
+{
+   int i = get_global_id(0);
+   if(i < count)
+       output[i] = input[i] * input[i];
+}
+`
+
+func getObjectStrings(object interface{}) map[string]string {
+	v := reflect.ValueOf(object)
+	t := reflect.TypeOf(object)
+
+	strs := make(map[string]string)
+
+	numMethods := t.NumMethod()
+	for i := 0; i < numMethods; i++ {
+		method := t.Method(i)
+		if method.Type.NumIn() == 1 && method.Type.NumOut() == 1 && method.Type.Out(0).Kind() == reflect.String {
+			// this is a string-returning method with (presumably) only a pointer receiver parameter
+			// call it
+			outs := v.Method(i).Call([]reflect.Value{})
+			// put the result in our map
+			strs[method.Name] = (outs[0].Interface()).(string)
+		}
+	}
+
+	return strs
+}
+
+func TestPlatformStringsContainNoNULs(t *testing.T) {
+	platforms, err := GetPlatforms()
+	if err != nil {
+		t.Fatalf("Failed to get platforms: %+v", err)
+	}
+
+	for _, p := range platforms {
+		for key, value := range getObjectStrings(p) {
+			if strings.Contains(value, "\x00") {
+				t.Fatalf("platform string %q =  %+q contains NUL", key, value)
+			}
+		}
+	}
+}
+
+func TestDeviceStringsContainNoNULs(t *testing.T) {
+	platforms, err := GetPlatforms()
+	if err != nil {
+		t.Fatalf("Failed to get platforms: %+v", err)
+	}
+
+	for _, p := range platforms {
+		devs, err := p.GetDevices(DeviceTypeAll)
+		if err != nil {
+			t.Fatalf("Failed to get devices for platform %q: %+v", p.Name(), err)
+		}
+
+		for _, d := range devs {
+			for key, value := range getObjectStrings(d) {
+				if strings.Contains(value, "\x00") {
+					t.Fatalf("device string %q =  %+q contains NUL", key, value)
+				}
+			}
+		}
+	}
+}
+
+func TestHello(t *testing.T) {
+	var data [1024]float32
+	for i := 0; i < len(data); i++ {
+		data[i] = rand.Float32()
+	}
+
+	platforms, err := GetPlatforms()
+	if err != nil {
+		t.Fatalf("Failed to get platforms: %+v", err)
+	}
+	for i, p := range platforms {
+		t.Logf("Platform %d:", i)
+		t.Logf("  Name: %s", p.Name())
+		t.Logf("  Vendor: %s", p.Vendor())
+		t.Logf("  Profile: %s", p.Profile())
+		t.Logf("  Version: %s", p.Version())
+		t.Logf("  Extensions: %s", p.Extensions())
+	}
+	platform := platforms[0]
+
+	devices, err := platform.GetDevices(DeviceTypeAll)
+	if err != nil {
+		t.Fatalf("Failed to get devices: %+v", err)
+	}
+	if len(devices) == 0 {
+		t.Fatalf("GetDevices returned no devices")
+	}
+	deviceIndex := -1
+	for i, d := range devices {
+		if deviceIndex < 0 && d.Type() == DeviceTypeGPU {
+			deviceIndex = i
+		}
+		t.Logf("Device %d (%s): %s", i, d.Type(), d.Name())
+		t.Logf("  Address Bits: %d", d.AddressBits())
+		t.Logf("  Available: %+v", d.Available())
+		// t.Logf("  Built-In Kernels: %s", d.BuiltInKernels())
+		t.Logf("  Compiler Available: %+v", d.CompilerAvailable())
+		t.Logf("  Double FP Config: %s", d.DoubleFPConfig())
+		t.Logf("  Driver Version: %s", d.DriverVersion())
+		t.Logf("  Error Correction Supported: %+v", d.ErrorCorrectionSupport())
+		t.Logf("  Execution Capabilities: %s", d.ExecutionCapabilities())
+		t.Logf("  Extensions: %s", d.Extensions())
+		t.Logf("  Global Memory Cache Type: %s", d.GlobalMemCacheType())
+		t.Logf("  Global Memory Cacheline Size: %d KB", d.GlobalMemCachelineSize()/1024)
+		t.Logf("  Global Memory Size: %d MB", d.GlobalMemSize()/(1024*1024))
+		t.Logf("  Half FP Config: %s", d.HalfFPConfig())
+		t.Logf("  Host Unified Memory: %+v", d.HostUnifiedMemory())
+		t.Logf("  Image Support: %+v", d.ImageSupport())
+		t.Logf("  Image2D Max Dimensions: %d x %d", d.Image2DMaxWidth(), d.Image2DMaxHeight())
+		t.Logf("  Image3D Max Dimenionns: %d x %d x %d", d.Image3DMaxWidth(), d.Image3DMaxHeight(), d.Image3DMaxDepth())
+		// t.Logf("  Image Max Buffer Size: %d", d.ImageMaxBufferSize())
+		// t.Logf("  Image Max Array Size: %d", d.ImageMaxArraySize())
+		// t.Logf("  Linker Available: %+v", d.LinkerAvailable())
+		t.Logf("  Little Endian: %+v", d.EndianLittle())
+		t.Logf("  Local Mem Size Size: %d KB", d.LocalMemSize()/1024)
+		t.Logf("  Local Mem Type: %s", d.LocalMemType())
+		t.Logf("  Max Clock Frequency: %d", d.MaxClockFrequency())
+		t.Logf("  Max Compute Units: %d", d.MaxComputeUnits())
+		t.Logf("  Max Constant Args: %d", d.MaxConstantArgs())
+		t.Logf("  Max Constant Buffer Size: %d KB", d.MaxConstantBufferSize()/1024)
+		t.Logf("  Max Mem Alloc Size: %d KB", d.MaxMemAllocSize()/1024)
+		t.Logf("  Max Parameter Size: %d", d.MaxParameterSize())
+		t.Logf("  Max Read-Image Args: %d", d.MaxReadImageArgs())
+		t.Logf("  Max Samplers: %d", d.MaxSamplers())
+		t.Logf("  Max Work Group Size: %d", d.MaxWorkGroupSize())
+		t.Logf("  Max Work Item Dimensions: %d", d.MaxWorkItemDimensions())
+		t.Logf("  Max Work Item Sizes: %d", d.MaxWorkItemSizes())
+		t.Logf("  Max Write-Image Args: %d", d.MaxWriteImageArgs())
+		t.Logf("  Memory Base Address Alignment: %d", d.MemBaseAddrAlign())
+		t.Logf("  Native Vector Width Char: %d", d.NativeVectorWidthChar())
+		t.Logf("  Native Vector Width Short: %d", d.NativeVectorWidthShort())
+		t.Logf("  Native Vector Width Int: %d", d.NativeVectorWidthInt())
+		t.Logf("  Native Vector Width Long: %d", d.NativeVectorWidthLong())
+		t.Logf("  Native Vector Width Float: %d", d.NativeVectorWidthFloat())
+		t.Logf("  Native Vector Width Double: %d", d.NativeVectorWidthDouble())
+		t.Logf("  Native Vector Width Half: %d", d.NativeVectorWidthHalf())
+		t.Logf("  OpenCL C Version: %s", d.OpenCLCVersion())
+		// t.Logf("  Parent Device: %+v", d.ParentDevice())
+		t.Logf("  Profile: %s", d.Profile())
+		t.Logf("  Profiling Timer Resolution: %d", d.ProfilingTimerResolution())
+		t.Logf("  Vendor: %s", d.Vendor())
+		t.Logf("  Version: %s", d.Version())
+	}
+	if deviceIndex < 0 {
+		deviceIndex = 0
+	}
+	device := devices[deviceIndex]
+	t.Logf("Using device %d", deviceIndex)
+	context, err := CreateContext([]*Device{device})
+	if err != nil {
+		t.Fatalf("CreateContext failed: %+v", err)
+	}
+	// imageFormats, err := context.GetSupportedImageFormats(0, MemObjectTypeImage2D)
+	// if err != nil {
+	// 	t.Fatalf("GetSupportedImageFormats failed: %+v", err)
+	// }
+	// t.Logf("Supported image formats: %+v", imageFormats)
+	queue, err := context.CreateCommandQueue(device, 0)
+	if err != nil {
+		t.Fatalf("CreateCommandQueue failed: %+v", err)
+	}
+	program, err := context.CreateProgramWithSource([]string{kernelSource})
+	if err != nil {
+		t.Fatalf("CreateProgramWithSource failed: %+v", err)
+	}
+	if err := program.BuildProgram(nil, ""); err != nil {
+		t.Fatalf("BuildProgram failed: %+v", err)
+	}
+	kernel, err := program.CreateKernel("square")
+	if err != nil {
+		t.Fatalf("CreateKernel failed: %+v", err)
+	}
+	for i := 0; i < 3; i++ {
+		name, err := kernel.ArgName(i)
+		if err == ErrUnsupported {
+			break
+		} else if err != nil {
+			t.Errorf("GetKernelArgInfo for name failed: %+v", err)
+			break
+		} else {
+			t.Logf("Kernel arg %d: %s", i, name)
+		}
+	}
+	input, err := context.CreateEmptyBuffer(MemReadOnly, 4*len(data))
+	if err != nil {
+		t.Fatalf("CreateBuffer failed for input: %+v", err)
+	}
+	output, err := context.CreateEmptyBuffer(MemReadOnly, 4*len(data))
+	if err != nil {
+		t.Fatalf("CreateBuffer failed for output: %+v", err)
+	}
+	if _, err := queue.EnqueueWriteBufferFloat32(input, true, 0, data[:], nil); err != nil {
+		t.Fatalf("EnqueueWriteBufferFloat32 failed: %+v", err)
+	}
+	if err := kernel.SetArgs(input, output, uint32(len(data))); err != nil {
+		t.Fatalf("SetKernelArgs failed: %+v", err)
+	}
+
+	local, err := kernel.WorkGroupSize(device)
+	if err != nil {
+		t.Fatalf("WorkGroupSize failed: %+v", err)
+	}
+	t.Logf("Work group size: %d", local)
+	size, _ := kernel.PreferredWorkGroupSizeMultiple(nil)
+	t.Logf("Preferred Work Group Size Multiple: %d", size)
+
+	global := len(data)
+	d := len(data) % local
+	if d != 0 {
+		global += local - d
+	}
+	if _, err := queue.EnqueueNDRangeKernel(kernel, nil, []int{global}, []int{local}, nil); err != nil {
+		t.Fatalf("EnqueueNDRangeKernel failed: %+v", err)
+	}
+
+	if err := queue.Finish(); err != nil {
+		t.Fatalf("Finish failed: %+v", err)
+	}
+
+	results := make([]float32, len(data))
+	if _, err := queue.EnqueueReadBufferFloat32(output, true, 0, results, nil); err != nil {
+		t.Fatalf("EnqueueReadBufferFloat32 failed: %+v", err)
+	}
+
+	correct := 0
+	for i, v := range data {
+		if results[i] == v*v {
+			correct++
+		}
+	}
+
+	if correct != len(data) {
+		t.Fatalf("%d/%d correct values", correct, len(data))
+	}
+}

Godeps/_workspace/src/github.com/Gustav-Simonsson/go-opencl/cl/context.go

+package cl
+
+// #include <stdlib.h>
+// #ifdef __APPLE__
+// #include "OpenCL/opencl.h"
+// #else
+// #include "cl.h"
+// #endif
+import "C"
+
+import (
+	"runtime"
+	"unsafe"
+)
+
+const maxImageFormats = 256
+
+type Context struct {
+	clContext C.cl_context
+	devices   []*Device
+}
+
+type MemObject struct {
+	clMem C.cl_mem
+	size  int
+}
+
+func releaseContext(c *Context) {
+	if c.clContext != nil {
+		C.clReleaseContext(c.clContext)
+		c.clContext = nil
+	}
+}
+
+func releaseMemObject(b *MemObject) {
+	if b.clMem != nil {
+		C.clReleaseMemObject(b.clMem)
+		b.clMem = nil
+	}
+}
+
+func newMemObject(mo C.cl_mem, size int) *MemObject {
+	memObject := &MemObject{clMem: mo, size: size}
+	runtime.SetFinalizer(memObject, releaseMemObject)
+	return memObject
+}
+
+func (b *MemObject) Release() {
+	releaseMemObject(b)
+}
+
+// TODO: properties
+func CreateContext(devices []*Device) (*Context, error) {
+	deviceIds := buildDeviceIdList(devices)
+	var err C.cl_int
+	clContext := C.clCreateContext(nil, C.cl_uint(len(devices)), &deviceIds[0], nil, nil, &err)
+	if err != C.CL_SUCCESS {
+		return nil, toError(err)
+	}
+	if clContext == nil {
+		return nil, ErrUnknown
+	}
+	context := &Context{clContext: clContext, devices: devices}
+	runtime.SetFinalizer(context, releaseContext)
+	return context, nil
+}
+
+func (ctx *Context) GetSupportedImageFormats(flags MemFlag, imageType MemObjectType) ([]ImageFormat, error) {
+	var formats [maxImageFormats]C.cl_image_format
+	var nFormats C.cl_uint
+	if err := C.clGetSupportedImageFormats(ctx.clContext, C.cl_mem_flags(flags), C.cl_mem_object_type(imageType), maxImageFormats, &formats[0], &nFormats); err != C.CL_SUCCESS {
+		return nil, toError(err)
+	}
+	fmts := make([]ImageFormat, nFormats)
+	for i, f := range formats[:nFormats] {
+		fmts[i] = ImageFormat{
+			ChannelOrder:    ChannelOrder(f.image_channel_order),
+			ChannelDataType: ChannelDataType(f.image_channel_data_type),
+		}
+	}
+	return fmts, nil
+}
+
+func (ctx *Context) CreateCommandQueue(device *Device, properties CommandQueueProperty) (*CommandQueue, error) {
+	var err C.cl_int
+	clQueue := C.clCreateCommandQueue(ctx.clContext, device.id, C.cl_command_queue_properties(properties), &err)
+	if err != C.CL_SUCCESS {
+		return nil, toError(err)
+	}
+	if clQueue == nil {
+		return nil, ErrUnknown
+	}
+	commandQueue := &CommandQueue{clQueue: clQueue, device: device}
+	runtime.SetFinalizer(commandQueue, releaseCommandQueue)
+	return commandQueue, nil
+}
+
+func (ctx *Context) CreateProgramWithSource(sources []string) (*Program, error) {
+	cSources := make([]*C.char, len(sources))
+	for i, s := range sources {
+		cs := C.CString(s)
+		cSources[i] = cs
+		defer C.free(unsafe.Pointer(cs))
+	}
+	var err C.cl_int
+	clProgram := C.clCreateProgramWithSource(ctx.clContext, C.cl_uint(len(sources)), &cSources[0], nil, &err)
+	if err != C.CL_SUCCESS {
+		return nil, toError(err)
+	}
+	if clProgram == nil {
+		return nil, ErrUnknown
+	}
+	program := &Program{clProgram: clProgram, devices: ctx.devices}
+	runtime.SetFinalizer(program, releaseProgram)
+	return program, nil
+}
+
+func (ctx *Context) CreateBufferUnsafe(flags MemFlag, size int, dataPtr unsafe.Pointer) (*MemObject, error) {
+	var err C.cl_int
+	clBuffer := C.clCreateBuffer(ctx.clContext, C.cl_mem_flags(flags), C.size_t(size), dataPtr, &err)
+	if err != C.CL_SUCCESS {
+		return nil, toError(err)
+	}
+	if clBuffer == nil {
+		return nil, ErrUnknown
+	}
+	return newMemObject(clBuffer, size), nil
+}
+
+func (ctx *Context) CreateEmptyBuffer(flags MemFlag, size int) (*MemObject, error) {
+	return ctx.CreateBufferUnsafe(flags, size, nil)
+}
+
+func (ctx *Context) CreateEmptyBufferFloat32(flags MemFlag, size int) (*MemObject, error) {
+	return ctx.CreateBufferUnsafe(flags, 4*size, nil)
+}
+
+func (ctx *Context) CreateBuffer(flags MemFlag, data []byte) (*MemObject, error) {
+	return ctx.CreateBufferUnsafe(flags, len(data), unsafe.Pointer(&data[0]))
+}
+
+//float64
+func (ctx *Context) CreateBufferFloat32(flags MemFlag, data []float32) (*MemObject, error) {
+	return ctx.CreateBufferUnsafe(flags, 4*len(data), unsafe.Pointer(&data[0]))
+}
+
+func (ctx *Context) CreateUserEvent() (*Event, error) {
+	var err C.cl_int
+	clEvent := C.clCreateUserEvent(ctx.clContext, &err)
+	if err != C.CL_SUCCESS {
+		return nil, toError(err)
+	}
+	return newEvent(clEvent), nil
+}
+
+func (ctx *Context) Release() {
+	releaseContext(ctx)
+}
+
+// http://www.khronos.org/registry/cl/sdk/1.2/docs/man/xhtml/clCreateSubBuffer.html
+// func (memObject *MemObject) CreateSubBuffer(flags MemFlag, bufferCreateType BufferCreateType, )

Godeps/_workspace/src/github.com/Gustav-Simonsson/go-opencl/cl/device12.go

+// +build cl12
+
+package cl
+
+// #ifdef __APPLE__
+// #include "OpenCL/opencl.h"
+// #else
+// #include "cl.h"
+// #endif
+import "C"
+import "unsafe"
+
+const FPConfigCorrectlyRoundedDivideSqrt FPConfig = C.CL_FP_CORRECTLY_ROUNDED_DIVIDE_SQRT
+
+func init() {
+	fpConfigNameMap[FPConfigCorrectlyRoundedDivideSqrt] = "CorrectlyRoundedDivideSqrt"
+}
+
+func (d *Device) BuiltInKernels() string {
+	str, _ := d.getInfoString(C.CL_DEVICE_BUILT_IN_KERNELS, true)
+	return str
+}
+
+// Is CL_FALSE if the implementation does not have a linker available. Is CL_TRUE if the linker is available. This can be CL_FALSE for the embedded platform profile only. This must be CL_TRUE if CL_DEVICE_COMPILER_AVAILABLE is CL_TRUE
+func (d *Device) LinkerAvailable() bool {
+	val, _ := d.getInfoBool(C.CL_DEVICE_LINKER_AVAILABLE, true)
+	return val
+}
+
+func (d *Device) ParentDevice() *Device {
+	var deviceId C.cl_device_id
+	if err := C.clGetDeviceInfo(d.id, C.CL_DEVICE_PARENT_DEVICE, C.size_t(unsafe.Sizeof(deviceId)), unsafe.Pointer(&deviceId), nil); err != C.CL_SUCCESS {
+		panic("ParentDevice failed")
+	}
+	if deviceId == nil {
+		return nil
+	}
+	return &Device{id: deviceId}
+}
+
+// Max number of pixels for a 1D image created from a buffer object. The minimum value is 65536 if CL_DEVICE_IMAGE_SUPPORT is CL_TRUE.
+func (d *Device) ImageMaxBufferSize() int {
+	val, _ := d.getInfoSize(C.CL_DEVICE_IMAGE_MAX_BUFFER_SIZE, true)
+	return int(val)
+}
+
+// Max number of images in a 1D or 2D image array. The minimum value is 2048 if CL_DEVICE_IMAGE_SUPPORT is CL_TRUE
+func (d *Device) ImageMaxArraySize() int {
+	val, _ := d.getInfoSize(C.CL_DEVICE_IMAGE_MAX_ARRAY_SIZE, true)
+	return int(val)
+}

Godeps/_workspace/src/github.com/Gustav-Simonsson/go-opencl/cl/headers/1.2/cl_gl.h

+/**********************************************************************************
+ * Copyright (c) 2008 - 2012 The Khronos Group Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and/or associated documentation files (the
+ * "Materials"), to deal in the Materials without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Materials, and to
+ * permit persons to whom the Materials are furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Materials.
+ *
+ * THE MATERIALS ARE PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY
+ * CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * MATERIALS OR THE USE OR OTHER DEALINGS IN THE MATERIALS.
+ **********************************************************************************/
+
+#ifndef __OPENCL_CL_GL_H
+#define __OPENCL_CL_GL_H
+
+#include <cl.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+typedef cl_uint     cl_gl_object_type;
+typedef cl_uint     cl_gl_texture_info;
+typedef cl_uint     cl_gl_platform_info;
+typedef struct __GLsync *cl_GLsync;
+
+/* cl_gl_object_type = 0x2000 - 0x200F enum values are currently taken           */
+#define CL_GL_OBJECT_BUFFER                     0x2000
+#define CL_GL_OBJECT_TEXTURE2D                  0x2001
+#define CL_GL_OBJECT_TEXTURE3D                  0x2002
+#define CL_GL_OBJECT_RENDERBUFFER               0x2003
+#define CL_GL_OBJECT_TEXTURE2D_ARRAY            0x200E
+#define CL_GL_OBJECT_TEXTURE1D                  0x200F
+#define CL_GL_OBJECT_TEXTURE1D_ARRAY            0x2010
+#define CL_GL_OBJECT_TEXTURE_BUFFER             0x2011
+
+/* cl_gl_texture_info           */
+#define CL_GL_TEXTURE_TARGET                    0x2004
+#define CL_GL_MIPMAP_LEVEL                      0x2005
+#define CL_GL_NUM_SAMPLES                       0x2012
+
+
+extern CL_API_ENTRY cl_mem CL_API_CALL
+clCreateFromGLBuffer(cl_context     /* context */,
+                     cl_mem_flags   /* flags */,
+                     cl_GLuint      /* bufobj */,
+                     int *          /* errcode_ret */) CL_API_SUFFIX__VERSION_1_0;
+
+extern CL_API_ENTRY cl_mem CL_API_CALL
+clCreateFromGLTexture(cl_context      /* context */,
+                      cl_mem_flags    /* flags */,
+                      cl_GLenum       /* target */,
+                      cl_GLint        /* miplevel */,
+                      cl_GLuint       /* texture */,
+                      cl_int *        /* errcode_ret */) CL_API_SUFFIX__VERSION_1_2;
+    
+extern CL_API_ENTRY cl_mem CL_API_CALL
+clCreateFromGLRenderbuffer(cl_context   /* context */,
+                           cl_mem_flags /* flags */,
+                           cl_GLuint    /* renderbuffer */,
+                           cl_int *     /* errcode_ret */) CL_API_SUFFIX__VERSION_1_0;
+
+extern CL_API_ENTRY cl_int CL_API_CALL
+clGetGLObjectInfo(cl_mem                /* memobj */,
+                  cl_gl_object_type *   /* gl_object_type */,
+                  cl_GLuint *           /* gl_object_name */) CL_API_SUFFIX__VERSION_1_0;
+                  
+extern CL_API_ENTRY cl_int CL_API_CALL
+clGetGLTextureInfo(cl_mem               /* memobj */,
+                   cl_gl_texture_info   /* param_name */,
+                   size_t               /* param_value_size */,
+                   void *               /* param_value */,
+                   size_t *             /* param_value_size_ret */) CL_API_SUFFIX__VERSION_1_0;
+
+extern CL_API_ENTRY cl_int CL_API_CALL
+clEnqueueAcquireGLObjects(cl_command_queue      /* command_queue */,
+                          cl_uint               /* num_objects */,
+                          const cl_mem *        /* mem_objects */,
+                          cl_uint               /* num_events_in_wait_list */,
+                          const cl_event *      /* event_wait_list */,
+                          cl_event *            /* event */) CL_API_SUFFIX__VERSION_1_0;
+
+extern CL_API_ENTRY cl_int CL_API_CALL
+clEnqueueReleaseGLObjects(cl_command_queue      /* command_queue */,
+                          cl_uint               /* num_objects */,
+                          const cl_mem *        /* mem_objects */,
+                          cl_uint               /* num_events_in_wait_list */,
+                          const cl_event *      /* event_wait_list */,
+                          cl_event *            /* event */) CL_API_SUFFIX__VERSION_1_0;
+
+
+/* Deprecated OpenCL 1.1 APIs */
+extern CL_API_ENTRY CL_EXT_PREFIX__VERSION_1_1_DEPRECATED cl_mem CL_API_CALL
+clCreateFromGLTexture2D(cl_context      /* context */,
+                        cl_mem_flags    /* flags */,
+                        cl_GLenum       /* target */,
+                        cl_GLint        /* miplevel */,
+                        cl_GLuint       /* texture */,
+                        cl_int *        /* errcode_ret */) CL_EXT_SUFFIX__VERSION_1_1_DEPRECATED;
+    
+extern CL_API_ENTRY CL_EXT_PREFIX__VERSION_1_1_DEPRECATED cl_mem CL_API_CALL
+clCreateFromGLTexture3D(cl_context      /* context */,
+                        cl_mem_flags    /* flags */,
+                        cl_GLenum       /* target */,
+                        cl_GLint        /* miplevel */,
+                        cl_GLuint       /* texture */,
+                        cl_int *        /* errcode_ret */) CL_EXT_SUFFIX__VERSION_1_1_DEPRECATED;
+    
+/* cl_khr_gl_sharing extension  */
+    
+#define cl_khr_gl_sharing 1
+    
+typedef cl_uint     cl_gl_context_info;
+    
+/* Additional Error Codes  */
+#define CL_INVALID_GL_SHAREGROUP_REFERENCE_KHR  -1000
+    
+/* cl_gl_context_info  */
+#define CL_CURRENT_DEVICE_FOR_GL_CONTEXT_KHR    0x2006
+#define CL_DEVICES_FOR_GL_CONTEXT_KHR           0x2007
+    
+/* Additional cl_context_properties  */
+#define CL_GL_CONTEXT_KHR                       0x2008
+#define CL_EGL_DISPLAY_KHR                      0x2009
+#define CL_GLX_DISPLAY_KHR                      0x200A
+#define CL_WGL_HDC_KHR                          0x200B
+#define CL_CGL_SHAREGROUP_KHR                   0x200C
+    
+extern CL_API_ENTRY cl_int CL_API_CALL
+clGetGLContextInfoKHR(const cl_context_properties * /* properties */,
+                      cl_gl_context_info            /* param_name */,
+                      size_t                        /* param_value_size */,
+                      void *                        /* param_value */,
+                      size_t *                      /* param_value_size_ret */) CL_API_SUFFIX__VERSION_1_0;
+    
+typedef CL_API_ENTRY cl_int (CL_API_CALL *clGetGLContextInfoKHR_fn)(
+    const cl_context_properties * properties,
+    cl_gl_context_info            param_name,
+    size_t                        param_value_size,
+    void *                        param_value,
+    size_t *                      param_value_size_ret);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif  /* __OPENCL_CL_GL_H */

Godeps/_workspace/src/github.com/Gustav-Simonsson/go-opencl/cl/headers/1.2/cl_gl_ext.h

+/**********************************************************************************
+ * Copyright (c) 2008-2012 The Khronos Group Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and/or associated documentation files (the
+ * "Materials"), to deal in the Materials without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Materials, and to
+ * permit persons to whom the Materials are furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Materials.
+ *
+ * THE MATERIALS ARE PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY
+ * CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * MATERIALS OR THE USE OR OTHER DEALINGS IN THE MATERIALS.
+ **********************************************************************************/
+
+/* $Revision: 11708 $ on $Date: 2010-06-13 23:36:24 -0700 (Sun, 13 Jun 2010) $ */
+
+/* cl_gl_ext.h contains vendor (non-KHR) OpenCL extensions which have           */
+/* OpenGL dependencies.                                                         */
+
+#ifndef __OPENCL_CL_GL_EXT_H
+#define __OPENCL_CL_GL_EXT_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include <cl_gl.h>
+
+/*
+ * For each extension, follow this template
+ *  cl_VEN_extname extension  */
+/* #define cl_VEN_extname 1
+ * ... define new types, if any
+ * ... define new tokens, if any
+ * ... define new APIs, if any
+ *
+ *  If you need GLtypes here, mirror them with a cl_GLtype, rather than including a GL header
+ *  This allows us to avoid having to decide whether to include GL headers or GLES here.
+ */
+
+/* 
+ *  cl_khr_gl_event  extension
+ *  See section 9.9 in the OpenCL 1.1 spec for more information
+ */
+#define CL_COMMAND_GL_FENCE_SYNC_OBJECT_KHR     0x200D
+
+extern CL_API_ENTRY cl_event CL_API_CALL
+clCreateEventFromGLsyncKHR(cl_context           /* context */,
+                           cl_GLsync            /* cl_GLsync */,
+                           cl_int *             /* errcode_ret */) CL_EXT_SUFFIX__VERSION_1_1;
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif	/* __OPENCL_CL_GL_EXT_H  */

Godeps/_workspace/src/github.com/Gustav-Simonsson/go-opencl/cl/headers/1.2/opencl.h

+/*******************************************************************************
+ * Copyright (c) 2008-2012 The Khronos Group Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and/or associated documentation files (the
+ * "Materials"), to deal in the Materials without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Materials, and to
+ * permit persons to whom the Materials are furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Materials.
+ *
+ * THE MATERIALS ARE PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY
+ * CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * MATERIALS OR THE USE OR OTHER DEALINGS IN THE MATERIALS.
+ ******************************************************************************/
+
+/* $Revision: 11708 $ on $Date: 2010-06-13 23:36:24 -0700 (Sun, 13 Jun 2010) $ */
+
+#ifndef __OPENCL_H
+#define __OPENCL_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include <cl.h>
+#include <cl_gl.h>
+#include <cl_gl_ext.h>
+#include <cl_ext.h>
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif  /* __OPENCL_H   */
+

Godeps/_workspace/src/github.com/Gustav-Simonsson/go-opencl/cl/image.go

+// +build cl12
+
+package cl
+
+// #ifdef __APPLE__
+// #include "OpenCL/opencl.h"
+// #else
+// #include "cl.h"
+// #endif
+import "C"
+import (
+	"image"
+	"unsafe"
+)
+
+func (ctx *Context) CreateImage(flags MemFlag, imageFormat ImageFormat, imageDesc ImageDescription, data []byte) (*MemObject, error) {
+	format := imageFormat.toCl()
+	desc := imageDesc.toCl()
+	var dataPtr unsafe.Pointer
+	if data != nil {
+		dataPtr = unsafe.Pointer(&data[0])
+	}
+	var err C.cl_int
+	clBuffer := C.clCreateImage(ctx.clContext, C.cl_mem_flags(flags), &format, &desc, dataPtr, &err)
+	if err != C.CL_SUCCESS {
+		return nil, toError(err)
+	}
+	if clBuffer == nil {
+		return nil, ErrUnknown
+	}
+	return newMemObject(clBuffer, len(data)), nil
+}
+
+func (ctx *Context) CreateImageSimple(flags MemFlag, width, height int, channelOrder ChannelOrder, channelDataType ChannelDataType, data []byte) (*MemObject, error) {
+	format := ImageFormat{channelOrder, channelDataType}
+	desc := ImageDescription{
+		Type:   MemObjectTypeImage2D,
+		Width:  width,
+		Height: height,
+	}
+	return ctx.CreateImage(flags, format, desc, data)
+}
+
+func (ctx *Context) CreateImageFromImage(flags MemFlag, img image.Image) (*MemObject, error) {
+	switch m := img.(type) {
+	case *image.Gray:
+		format := ImageFormat{ChannelOrderIntensity, ChannelDataTypeUNormInt8}
+		desc := ImageDescription{
+			Type:     MemObjectTypeImage2D,
+			Width:    m.Bounds().Dx(),
+			Height:   m.Bounds().Dy(),
+			RowPitch: m.Stride,
+		}
+		return ctx.CreateImage(flags, format, desc, m.Pix)
+	case *image.RGBA:
+		format := ImageFormat{ChannelOrderRGBA, ChannelDataTypeUNormInt8}
+		desc := ImageDescription{
+			Type:     MemObjectTypeImage2D,
+			Width:    m.Bounds().Dx(),
+			Height:   m.Bounds().Dy(),
+			RowPitch: m.Stride,
+		}
+		return ctx.CreateImage(flags, format, desc, m.Pix)
+	}
+
+	b := img.Bounds()
+	w := b.Dx()
+	h := b.Dy()
+	data := make([]byte, w*h*4)
+	dataOffset := 0
+	for y := 0; y < h; y++ {
+		for x := 0; x < w; x++ {
+			c := img.At(x+b.Min.X, y+b.Min.Y)
+			r, g, b, a := c.RGBA()
+			data[dataOffset] = uint8(r >> 8)
+			data[dataOffset+1] = uint8(g >> 8)
+			data[dataOffset+2] = uint8(b >> 8)
+			data[dataOffset+3] = uint8(a >> 8)
+			dataOffset += 4
+		}
+	}
+	return ctx.CreateImageSimple(flags, w, h, ChannelOrderRGBA, ChannelDataTypeUNormInt8, data)
+}

Godeps/_workspace/src/github.com/Gustav-Simonsson/go-opencl/cl/kernel.go

+package cl
+
+// #ifdef __APPLE__
+// #include "OpenCL/opencl.h"
+// #else
+// #include "cl.h"
+// #endif
+import "C"
+
+import (
+	"fmt"
+	"unsafe"
+)
+
+type ErrUnsupportedArgumentType struct {
+	Index int
+	Value interface{}
+}
+
+func (e ErrUnsupportedArgumentType) Error() string {
+	return fmt.Sprintf("cl: unsupported argument type for index %d: %+v", e.Index, e.Value)
+}
+
+type Kernel struct {
+	clKernel C.cl_kernel
+	name     string
+}
+
+type LocalBuffer int
+
+func releaseKernel(k *Kernel) {
+	if k.clKernel != nil {
+		C.clReleaseKernel(k.clKernel)
+		k.clKernel = nil
+	}
+}
+
+func (k *Kernel) Release() {
+	releaseKernel(k)
+}
+
+func (k *Kernel) SetArgs(args ...interface{}) error {
+	for index, arg := range args {
+		if err := k.SetArg(index, arg); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+func (k *Kernel) SetArg(index int, arg interface{}) error {
+	switch val := arg.(type) {
+	case uint8:
+		return k.SetArgUint8(index, val)
+	case int8:
+		return k.SetArgInt8(index, val)
+	case uint32:
+		return k.SetArgUint32(index, val)
+	case uint64:
+		return k.SetArgUint64(index, val)
+	case int32:
+		return k.SetArgInt32(index, val)
+	case float32:
+		return k.SetArgFloat32(index, val)
+	case *MemObject:
+		return k.SetArgBuffer(index, val)
+	case LocalBuffer:
+		return k.SetArgLocal(index, int(val))
+	default:
+		return ErrUnsupportedArgumentType{Index: index, Value: arg}
+	}
+}
+
+func (k *Kernel) SetArgBuffer(index int, buffer *MemObject) error {
+	return k.SetArgUnsafe(index, int(unsafe.Sizeof(buffer.clMem)), unsafe.Pointer(&buffer.clMem))
+}
+
+func (k *Kernel) SetArgFloat32(index int, val float32) error {
+	return k.SetArgUnsafe(index, int(unsafe.Sizeof(val)), unsafe.Pointer(&val))
+}
+
+func (k *Kernel) SetArgInt8(index int, val int8) error {
+	return k.SetArgUnsafe(index, int(unsafe.Sizeof(val)), unsafe.Pointer(&val))
+}
+
+func (k *Kernel) SetArgUint8(index int, val uint8) error {
+	return k.SetArgUnsafe(index, int(unsafe.Sizeof(val)), unsafe.Pointer(&val))
+}
+
+func (k *Kernel) SetArgInt32(index int, val int32) error {
+	return k.SetArgUnsafe(index, int(unsafe.Sizeof(val)), unsafe.Pointer(&val))
+}
+
+func (k *Kernel) SetArgUint32(index int, val uint32) error {
+	return k.SetArgUnsafe(index, int(unsafe.Sizeof(val)), unsafe.Pointer(&val))
+}
+
+func (k *Kernel) SetArgUint64(index int, val uint64) error {
+	return k.SetArgUnsafe(index, int(unsafe.Sizeof(val)), unsafe.Pointer(&val))
+}
+
+func (k *Kernel) SetArgLocal(index int, size int) error {
+	return k.SetArgUnsafe(index, size, nil)
+}
+
+func (k *Kernel) SetArgUnsafe(index, argSize int, arg unsafe.Pointer) error {
+	//fmt.Println("FUNKY: ", index, argSize)
+	return toError(C.clSetKernelArg(k.clKernel, C.cl_uint(index), C.size_t(argSize), arg))
+}
+
+func (k *Kernel) PreferredWorkGroupSizeMultiple(device *Device) (int, error) {
+	var size C.size_t
+	err := C.clGetKernelWorkGroupInfo(k.clKernel, device.nullableId(), C.CL_KERNEL_PREFERRED_WORK_GROUP_SIZE_MULTIPLE, C.size_t(unsafe.Sizeof(size)), unsafe.Pointer(&size), nil)
+	return int(size), toError(err)
+}
+
+func (k *Kernel) WorkGroupSize(device *Device) (int, error) {
+	var size C.size_t
+	err := C.clGetKernelWorkGroupInfo(k.clKernel, device.nullableId(), C.CL_KERNEL_WORK_GROUP_SIZE, C.size_t(unsafe.Sizeof(size)), unsafe.Pointer(&size), nil)
+	return int(size), toError(err)
+}
+
+func (k *Kernel) NumArgs() (int, error) {
+	var num C.cl_uint
+	err := C.clGetKernelInfo(k.clKernel, C.CL_KERNEL_NUM_ARGS, C.size_t(unsafe.Sizeof(num)), unsafe.Pointer(&num), nil)
+	return int(num), toError(err)
+}

Godeps/_workspace/src/github.com/Gustav-Simonsson/go-opencl/cl/kernel10.go

+// +build !cl12
+
+package cl
+
+func (k *Kernel) ArgName(index int) (string, error) {
+	return "", ErrUnsupported
+}

Godeps/_workspace/src/github.com/Gustav-Simonsson/go-opencl/cl/kernel12.go

+// +build cl12
+
+package cl
+
+// #ifdef __APPLE__
+// #include "OpenCL/opencl.h"
+// #else
+// #include "cl.h"
+// #endif
+import "C"
+import "unsafe"
+
+func (k *Kernel) ArgName(index int) (string, error) {
+	var strC [1024]byte
+	var strN C.size_t
+	if err := C.clGetKernelArgInfo(k.clKernel, C.cl_uint(index), C.CL_KERNEL_ARG_NAME, 1024, unsafe.Pointer(&strC[0]), &strN); err != C.CL_SUCCESS {
+		return "", toError(err)
+	}
+	return string(strC[:strN]), nil
+}

Godeps/_workspace/src/github.com/Gustav-Simonsson/go-opencl/cl/platform.go

+package cl
+
+// #ifdef __APPLE__
+// #include "OpenCL/opencl.h"
+// #else
+// #include "cl.h"
+// #endif
+import "C"
+
+import "unsafe"
+
+const maxPlatforms = 32
+
+type Platform struct {
+	id C.cl_platform_id
+}
+
+// Obtain the list of platforms available.
+func GetPlatforms() ([]*Platform, error) {
+	var platformIds [maxPlatforms]C.cl_platform_id
+	var nPlatforms C.cl_uint
+	if err := C.clGetPlatformIDs(C.cl_uint(maxPlatforms), &platformIds[0], &nPlatforms); err != C.CL_SUCCESS {
+		return nil, toError(err)
+	}
+	platforms := make([]*Platform, nPlatforms)
+	for i := 0; i < int(nPlatforms); i++ {
+		platforms[i] = &Platform{id: platformIds[i]}
+	}
+	return platforms, nil
+}
+
+func (p *Platform) GetDevices(deviceType DeviceType) ([]*Device, error) {
+	return GetDevices(p, deviceType)
+}
+
+func (p *Platform) getInfoString(param C.cl_platform_info) (string, error) {
+	var strC [2048]byte
+	var strN C.size_t
+	if err := C.clGetPlatformInfo(p.id, param, 2048, unsafe.Pointer(&strC[0]), &strN); err != C.CL_SUCCESS {
+		return "", toError(err)
+	}
+	return string(strC[:(strN - 1)]), nil
+}
+
+func (p *Platform) Name() string {
+	if str, err := p.getInfoString(C.CL_PLATFORM_NAME); err != nil {
+		panic("Platform.Name() should never fail")
+	} else {
+		return str
+	}
+}
+
+func (p *Platform) Vendor() string {
+	if str, err := p.getInfoString(C.CL_PLATFORM_VENDOR); err != nil {
+		panic("Platform.Vendor() should never fail")
+	} else {
+		return str
+	}
+}
+
+func (p *Platform) Profile() string {
+	if str, err := p.getInfoString(C.CL_PLATFORM_PROFILE); err != nil {
+		panic("Platform.Profile() should never fail")
+	} else {
+		return str
+	}
+}
+
+func (p *Platform) Version() string {
+	if str, err := p.getInfoString(C.CL_PLATFORM_VERSION); err != nil {
+		panic("Platform.Version() should never fail")
+	} else {
+		return str
+	}
+}
+
+func (p *Platform) Extensions() string {
+	if str, err := p.getInfoString(C.CL_PLATFORM_EXTENSIONS); err != nil {
+		panic("Platform.Extensions() should never fail")
+	} else {
+		return str
+	}
+}

Godeps/_workspace/src/github.com/Gustav-Simonsson/go-opencl/cl/program.go

+package cl
+
+// #include <stdlib.h>
+// #ifdef __APPLE__
+// #include "OpenCL/opencl.h"
+// #else
+// #include "cl.h"
+// #endif
+import "C"
+
+import (
+	"fmt"
+	"runtime"
+	"unsafe"
+)
+
+type BuildError struct {
+	Message string
+	Device  *Device
+}
+
+func (e BuildError) Error() string {
+	if e.Device != nil {
+		return fmt.Sprintf("cl: build error on %q: %s", e.Device.Name(), e.Message)
+	} else {
+		return fmt.Sprintf("cl: build error: %s", e.Message)
+	}
+}
+
+type Program struct {
+	clProgram C.cl_program
+	devices   []*Device
+}
+
+func releaseProgram(p *Program) {
+	if p.clProgram != nil {
+		C.clReleaseProgram(p.clProgram)
+		p.clProgram = nil
+	}
+}
+
+func (p *Program) Release() {
+	releaseProgram(p)
+}
+
+func (p *Program) BuildProgram(devices []*Device, options string) error {
+	var cOptions *C.char
+	if options != "" {
+		cOptions = C.CString(options)
+		defer C.free(unsafe.Pointer(cOptions))
+	}
+	var deviceList []C.cl_device_id
+	var deviceListPtr *C.cl_device_id
+	numDevices := C.cl_uint(len(devices))
+	if devices != nil && len(devices) > 0 {
+		deviceList = buildDeviceIdList(devices)
+		deviceListPtr = &deviceList[0]
+	}
+	if err := C.clBuildProgram(p.clProgram, numDevices, deviceListPtr, cOptions, nil, nil); err != C.CL_SUCCESS {
+		buffer := make([]byte, 4096)
+		var bLen C.size_t
+		var err C.cl_int
+
+		for _, dev := range p.devices {
+			for i := 2; i >= 0; i-- {
+				err = C.clGetProgramBuildInfo(p.clProgram, dev.id, C.CL_PROGRAM_BUILD_LOG, C.size_t(len(buffer)), unsafe.Pointer(&buffer[0]), &bLen)
+				if err == C.CL_INVALID_VALUE && i > 0 && bLen < 1024*1024 {
+					// INVALID_VALUE probably means our buffer isn't large enough
+					buffer = make([]byte, bLen)
+				} else {
+					break
+				}
+			}
+			if err != C.CL_SUCCESS {
+				return toError(err)
+			}
+
+			if bLen > 1 {
+				return BuildError{
+					Device:  dev,
+					Message: string(buffer[:bLen-1]),
+				}
+			}
+		}
+
+		return BuildError{
+			Device:  nil,
+			Message: "build failed and produced no log entries",
+		}
+	}
+	return nil
+}
+
+func (p *Program) CreateKernel(name string) (*Kernel, error) {
+	cName := C.CString(name)
+	defer C.free(unsafe.Pointer(cName))
+	var err C.cl_int
+	clKernel := C.clCreateKernel(p.clProgram, cName, &err)
+	if err != C.CL_SUCCESS {
+		return nil, toError(err)
+	}
+	kernel := &Kernel{clKernel: clKernel, name: name}
+	runtime.SetFinalizer(kernel, releaseKernel)
+	return kernel, nil
+}

Godeps/_workspace/src/github.com/Gustav-Simonsson/go-opencl/cl/queue.go

+package cl
+
+// #ifdef __APPLE__
+// #include "OpenCL/opencl.h"
+// #else
+// #include "cl.h"
+// #endif
+import "C"
+
+import "unsafe"
+
+type CommandQueueProperty int
+
+const (
+	CommandQueueOutOfOrderExecModeEnable CommandQueueProperty = C.CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE
+	CommandQueueProfilingEnable          CommandQueueProperty = C.CL_QUEUE_PROFILING_ENABLE
+)
+
+type CommandQueue struct {
+	clQueue C.cl_command_queue
+	device  *Device
+}
+
+func releaseCommandQueue(q *CommandQueue) {
+	if q.clQueue != nil {
+		C.clReleaseCommandQueue(q.clQueue)
+		q.clQueue = nil
+	}
+}
+
+// Call clReleaseCommandQueue on the CommandQueue. Using the CommandQueue after Release will cause a panick.
+func (q *CommandQueue) Release() {
+	releaseCommandQueue(q)
+}
+
+// Blocks until all previously queued OpenCL commands in a command-queue are issued to the associated device and have completed.
+func (q *CommandQueue) Finish() error {
+	return toError(C.clFinish(q.clQueue))
+}
+
+// Issues all previously queued OpenCL commands in a command-queue to the device associated with the command-queue.
+func (q *CommandQueue) Flush() error {
+	return toError(C.clFlush(q.clQueue))
+}
+
+// Enqueues a command to map a region of the buffer object given by buffer into the host address space and returns a pointer to this mapped region.
+func (q *CommandQueue) EnqueueMapBuffer(buffer *MemObject, blocking bool, flags MapFlag, offset, size int, eventWaitList []*Event) (*MappedMemObject, *Event, error) {
+	var event C.cl_event
+	var err C.cl_int
+	ptr := C.clEnqueueMapBuffer(q.clQueue, buffer.clMem, clBool(blocking), flags.toCl(), C.size_t(offset), C.size_t(size), C.cl_uint(len(eventWaitList)), eventListPtr(eventWaitList), &event, &err)
+	if err != C.CL_SUCCESS {
+		return nil, nil, toError(err)
+	}
+	ev := newEvent(event)
+	if ptr == nil {
+		return nil, ev, ErrUnknown
+	}
+	return &MappedMemObject{ptr: ptr, size: size}, ev, nil
+}
+
+// Enqueues a command to map a region of an image object into the host address space and returns a pointer to this mapped region.
+func (q *CommandQueue) EnqueueMapImage(buffer *MemObject, blocking bool, flags MapFlag, origin, region [3]int, eventWaitList []*Event) (*MappedMemObject, *Event, error) {
+	cOrigin := sizeT3(origin)
+	cRegion := sizeT3(region)
+	var event C.cl_event
+	var err C.cl_int
+	var rowPitch, slicePitch C.size_t
+	ptr := C.clEnqueueMapImage(q.clQueue, buffer.clMem, clBool(blocking), flags.toCl(), &cOrigin[0], &cRegion[0], &rowPitch, &slicePitch, C.cl_uint(len(eventWaitList)), eventListPtr(eventWaitList), &event, &err)
+	if err != C.CL_SUCCESS {
+		return nil, nil, toError(err)
+	}
+	ev := newEvent(event)
+	if ptr == nil {
+		return nil, ev, ErrUnknown
+	}
+	size := 0 // TODO: could calculate this
+	return &MappedMemObject{ptr: ptr, size: size, rowPitch: int(rowPitch), slicePitch: int(slicePitch)}, ev, nil
+}
+
+// Enqueues a command to unmap a previously mapped region of a memory object.
+func (q *CommandQueue) EnqueueUnmapMemObject(buffer *MemObject, mappedObj *MappedMemObject, eventWaitList []*Event) (*Event, error) {
+	var event C.cl_event
+	if err := C.clEnqueueUnmapMemObject(q.clQueue, buffer.clMem, mappedObj.ptr, C.cl_uint(len(eventWaitList)), eventListPtr(eventWaitList), &event); err != C.CL_SUCCESS {
+		return nil, toError(err)
+	}
+	return newEvent(event), nil
+}
+
+// Enqueues a command to copy a buffer object to another buffer object.
+func (q *CommandQueue) EnqueueCopyBuffer(srcBuffer, dstBuffer *MemObject, srcOffset, dstOffset, byteCount int, eventWaitList []*Event) (*Event, error) {
+	var event C.cl_event
+	err := toError(C.clEnqueueCopyBuffer(q.clQueue, srcBuffer.clMem, dstBuffer.clMem, C.size_t(srcOffset), C.size_t(dstOffset), C.size_t(byteCount), C.cl_uint(len(eventWaitList)), eventListPtr(eventWaitList), &event))
+	return newEvent(event), err
+}
+
+// Enqueue commands to write to a buffer object from host memory.
+func (q *CommandQueue) EnqueueWriteBuffer(buffer *MemObject, blocking bool, offset, dataSize int, dataPtr unsafe.Pointer, eventWaitList []*Event) (*Event, error) {
+	var event C.cl_event
+	err := toError(C.clEnqueueWriteBuffer(q.clQueue, buffer.clMem, clBool(blocking), C.size_t(offset), C.size_t(dataSize), dataPtr, C.cl_uint(len(eventWaitList)), eventListPtr(eventWaitList), &event))
+	return newEvent(event), err
+}
+
+func (q *CommandQueue) EnqueueWriteBufferFloat32(buffer *MemObject, blocking bool, offset int, data []float32, eventWaitList []*Event) (*Event, error) {
+	dataPtr := unsafe.Pointer(&data[0])
+	dataSize := int(unsafe.Sizeof(data[0])) * len(data)
+	return q.EnqueueWriteBuffer(buffer, blocking, offset, dataSize, dataPtr, eventWaitList)
+}
+
+// Enqueue commands to read from a buffer object to host memory.
+func (q *CommandQueue) EnqueueReadBuffer(buffer *MemObject, blocking bool, offset, dataSize int, dataPtr unsafe.Pointer, eventWaitList []*Event) (*Event, error) {
+	var event C.cl_event
+	err := toError(C.clEnqueueReadBuffer(q.clQueue, buffer.clMem, clBool(blocking), C.size_t(offset), C.size_t(dataSize), dataPtr, C.cl_uint(len(eventWaitList)), eventListPtr(eventWaitList), &event))
+	return newEvent(event), err
+}
+
+func (q *CommandQueue) EnqueueReadBufferFloat32(buffer *MemObject, blocking bool, offset int, data []float32, eventWaitList []*Event) (*Event, error) {
+	dataPtr := unsafe.Pointer(&data[0])
+	dataSize := int(unsafe.Sizeof(data[0])) * len(data)
+	return q.EnqueueReadBuffer(buffer, blocking, offset, dataSize, dataPtr, eventWaitList)
+}
+
+// Enqueues a command to execute a kernel on a device.
+func (q *CommandQueue) EnqueueNDRangeKernel(kernel *Kernel, globalWorkOffset, globalWorkSize, localWorkSize []int, eventWaitList []*Event) (*Event, error) {
+	workDim := len(globalWorkSize)
+	var globalWorkOffsetList []C.size_t
+	var globalWorkOffsetPtr *C.size_t
+	if globalWorkOffset != nil {
+		globalWorkOffsetList = make([]C.size_t, len(globalWorkOffset))
+		for i, off := range globalWorkOffset {
+			globalWorkOffsetList[i] = C.size_t(off)
+		}
+		globalWorkOffsetPtr = &globalWorkOffsetList[0]
+	}
+	var globalWorkSizeList []C.size_t
+	var globalWorkSizePtr *C.size_t
+	if globalWorkSize != nil {
+		globalWorkSizeList = make([]C.size_t, len(globalWorkSize))
+		for i, off := range globalWorkSize {
+			globalWorkSizeList[i] = C.size_t(off)
+		}
+		globalWorkSizePtr = &globalWorkSizeList[0]
+	}
+	var localWorkSizeList []C.size_t
+	var localWorkSizePtr *C.size_t
+	if localWorkSize != nil {
+		localWorkSizeList = make([]C.size_t, len(localWorkSize))
+		for i, off := range localWorkSize {
+			localWorkSizeList[i] = C.size_t(off)
+		}
+		localWorkSizePtr = &localWorkSizeList[0]
+	}
+	var event C.cl_event
+	err := toError(C.clEnqueueNDRangeKernel(q.clQueue, kernel.clKernel, C.cl_uint(workDim), globalWorkOffsetPtr, globalWorkSizePtr, localWorkSizePtr, C.cl_uint(len(eventWaitList)), eventListPtr(eventWaitList), &event))
+	return newEvent(event), err
+}
+
+// Enqueues a command to read from a 2D or 3D image object to host memory.
+func (q *CommandQueue) EnqueueReadImage(image *MemObject, blocking bool, origin, region [3]int, rowPitch, slicePitch int, data []byte, eventWaitList []*Event) (*Event, error) {
+	cOrigin := sizeT3(origin)
+	cRegion := sizeT3(region)
+	var event C.cl_event
+	err := toError(C.clEnqueueReadImage(q.clQueue, image.clMem, clBool(blocking), &cOrigin[0], &cRegion[0], C.size_t(rowPitch), C.size_t(slicePitch), unsafe.Pointer(&data[0]), C.cl_uint(len(eventWaitList)), eventListPtr(eventWaitList), &event))
+	return newEvent(event), err
+}
+
+// Enqueues a command to write from a 2D or 3D image object to host memory.
+func (q *CommandQueue) EnqueueWriteImage(image *MemObject, blocking bool, origin, region [3]int, rowPitch, slicePitch int, data []byte, eventWaitList []*Event) (*Event, error) {
+	cOrigin := sizeT3(origin)
+	cRegion := sizeT3(region)
+	var event C.cl_event
+	err := toError(C.clEnqueueWriteImage(q.clQueue, image.clMem, clBool(blocking), &cOrigin[0], &cRegion[0], C.size_t(rowPitch), C.size_t(slicePitch), unsafe.Pointer(&data[0]), C.cl_uint(len(eventWaitList)), eventListPtr(eventWaitList), &event))
+	return newEvent(event), err
+}
+
+func (q *CommandQueue) EnqueueFillBuffer(buffer *MemObject, pattern unsafe.Pointer, patternSize, offset, size int, eventWaitList []*Event) (*Event, error) {
+	var event C.cl_event
+	err := toError(C.clEnqueueFillBuffer(q.clQueue, buffer.clMem, pattern, C.size_t(patternSize), C.size_t(offset), C.size_t(size), C.cl_uint(len(eventWaitList)), eventListPtr(eventWaitList), &event))
+	return newEvent(event), err
+}
+
+// A synchronization point that enqueues a barrier operation.
+func (q *CommandQueue) EnqueueBarrierWithWaitList(eventWaitList []*Event) (*Event, error) {
+	var event C.cl_event
+	err := toError(C.clEnqueueBarrierWithWaitList(q.clQueue, C.cl_uint(len(eventWaitList)), eventListPtr(eventWaitList), &event))
+	return newEvent(event), err
+}
+
+// Enqueues a marker command which waits for either a list of events to complete, or all previously enqueued commands to complete.
+func (q *CommandQueue) EnqueueMarkerWithWaitList(eventWaitList []*Event) (*Event, error) {
+	var event C.cl_event
+	err := toError(C.clEnqueueMarkerWithWaitList(q.clQueue, C.cl_uint(len(eventWaitList)), eventListPtr(eventWaitList), &event))
+	return newEvent(event), err
+}

Godeps/_workspace/src/github.com/Gustav-Simonsson/go-opencl/cl/types12.go

+// +build cl12
+
+package cl
+
+// #ifdef __APPLE__
+// #include "OpenCL/opencl.h"
+// #else
+// #include "cl.h"
+// #endif
+import "C"
+
+const (
+	ChannelDataTypeUNormInt24  ChannelDataType = C.CL_UNORM_INT24
+	ChannelOrderDepth          ChannelOrder    = C.CL_DEPTH
+	ChannelOrderDepthStencil   ChannelOrder    = C.CL_DEPTH_STENCIL
+	MemHostNoAccess            MemFlag         = C.CL_MEM_HOST_NO_ACCESS  // OpenCL 1.2
+	MemHostReadOnly            MemFlag         = C.CL_MEM_HOST_READ_ONLY  // OpenCL 1.2
+	MemHostWriteOnly           MemFlag         = C.CL_MEM_HOST_WRITE_ONLY // OpenCL 1.2
+	MemObjectTypeImage1D       MemObjectType   = C.CL_MEM_OBJECT_IMAGE1D
+	MemObjectTypeImage1DArray  MemObjectType   = C.CL_MEM_OBJECT_IMAGE1D_ARRAY
+	MemObjectTypeImage1DBuffer MemObjectType   = C.CL_MEM_OBJECT_IMAGE1D_BUFFER
+	MemObjectTypeImage2DArray  MemObjectType   = C.CL_MEM_OBJECT_IMAGE2D_ARRAY
+	// This flag specifies that the region being mapped in the memory object is being mapped for writing.
+	//
+	// The contents of the region being mapped are to be discarded. This is typically the case when the
+	// region being mapped is overwritten by the host. This flag allows the implementation to no longer
+	// guarantee that the pointer returned by clEnqueueMapBuffer or clEnqueueMapImage contains the
+	// latest bits in the region being mapped which can be a significant performance enhancement.
+	MapFlagWriteInvalidateRegion MapFlag = C.CL_MAP_WRITE_INVALIDATE_REGION
+)
+
+func init() {
+	errorMap[C.CL_COMPILE_PROGRAM_FAILURE] = ErrCompileProgramFailure
+	errorMap[C.CL_DEVICE_PARTITION_FAILED] = ErrDevicePartitionFailed
+	errorMap[C.CL_INVALID_COMPILER_OPTIONS] = ErrInvalidCompilerOptions
+	errorMap[C.CL_INVALID_DEVICE_PARTITION_COUNT] = ErrInvalidDevicePartitionCount
+	errorMap[C.CL_INVALID_IMAGE_DESCRIPTOR] = ErrInvalidImageDescriptor
+	errorMap[C.CL_INVALID_LINKER_OPTIONS] = ErrInvalidLinkerOptions
+	errorMap[C.CL_KERNEL_ARG_INFO_NOT_AVAILABLE] = ErrKernelArgInfoNotAvailable
+	errorMap[C.CL_LINK_PROGRAM_FAILURE] = ErrLinkProgramFailure
+	errorMap[C.CL_LINKER_NOT_AVAILABLE] = ErrLinkerNotAvailable
+	channelOrderNameMap[ChannelOrderDepth] = "Depth"
+	channelOrderNameMap[ChannelOrderDepthStencil] = "DepthStencil"
+	channelDataTypeNameMap[ChannelDataTypeUNormInt24] = "UNormInt24"
+}
+
+type ImageDescription struct {
+	Type                            MemObjectType
+	Width, Height, Depth            int
+	ArraySize, RowPitch, SlicePitch int
+	NumMipLevels, NumSamples        int
+	Buffer                          *MemObject
+}
+
+func (d ImageDescription) toCl() C.cl_image_desc {
+	var desc C.cl_image_desc
+	desc.image_type = C.cl_mem_object_type(d.Type)
+	desc.image_width = C.size_t(d.Width)
+	desc.image_height = C.size_t(d.Height)
+	desc.image_depth = C.size_t(d.Depth)
+	desc.image_array_size = C.size_t(d.ArraySize)
+	desc.image_row_pitch = C.size_t(d.RowPitch)
+	desc.image_slice_pitch = C.size_t(d.SlicePitch)
+	desc.num_mip_levels = C.cl_uint(d.NumMipLevels)
+	desc.num_samples = C.cl_uint(d.NumSamples)
+	desc.buffer = nil
+	if d.Buffer != nil {
+		desc.buffer = d.Buffer.clMem
+	}
+	return desc
+}

Godeps/_workspace/src/github.com/Gustav-Simonsson/go-opencl/cl/types_darwin.go

+package cl
+
+// #ifdef __APPLE__
+// #include "OpenCL/opencl.h"
+// #else
+// #include "cl.h"
+// #endif
+import "C"
+
+// Extension: cl_APPLE_fixed_alpha_channel_orders
+//
+// These selectors may be passed to clCreateImage2D() in the cl_image_format.image_channel_order field.
+// They are like CL_BGRA and CL_ARGB except that the alpha channel to be ignored.  On calls to read_imagef,
+// the alpha will be 0xff (1.0f) if the sample falls in the image and 0 if it does not fall in the image.
+// On calls to write_imagef, the alpha value is ignored and 0xff (1.0f) is written. These formats are
+// currently only available for the CL_UNORM_INT8 cl_channel_type. They are intended to support legacy
+// image formats.
+const (
+	ChannelOrder1RGBApple ChannelOrder = C.CL_1RGB_APPLE // Introduced in MacOS X.7.
+	ChannelOrderBGR1Apple ChannelOrder = C.CL_BGR1_APPLE // Introduced in MacOS X.7.
+)
+
+// Extension: cl_APPLE_biased_fixed_point_image_formats
+//
+// This selector may be passed to clCreateImage2D() in the cl_image_format.image_channel_data_type field.
+// It defines a biased signed 1.14 fixed point storage format, with range [-1, 3). The conversion from
+// float to this fixed point format is defined as follows:
+//
+//      ushort float_to_sfixed14( float x ){
+//          int i = convert_int_sat_rte( x * 0x1.0p14f );         // scale [-1, 3.0) to [-16384, 3*16384), round to nearest integer
+//          i = add_sat( i, 0x4000 );                             // apply bias, to convert to [0, 65535) range
+//          return convert_ushort_sat(i);                         // clamp to destination size
+//      }
+//
+// The inverse conversion is the reverse process. The formats are currently only available on the CPU with
+// the CL_RGBA channel layout.
+const (
+	ChannelDataTypeSFixed14Apple ChannelDataType = C.CL_SFIXED14_APPLE // Introduced in MacOS X.7.
+)
+
+func init() {
+	channelOrderNameMap[ChannelOrder1RGBApple] = "1RGBApple"
+	channelOrderNameMap[ChannelOrderBGR1Apple] = "RGB1Apple"
+	channelDataTypeNameMap[ChannelDataTypeSFixed14Apple] = "SFixed14Apple"
+}

Godeps/_workspace/src/github.com/ethereum/ethash/ethash.go

@@ -30,8 +30,8 @@ import (
 )
 
 var (
-	minDifficulty = new(big.Int).Exp(big.NewInt(2), big.NewInt(256), big.NewInt(0))
-	sharedLight   = new(Light)
+	maxUint256  = new(big.Int).Exp(big.NewInt(2), big.NewInt(256), big.NewInt(0))
+	sharedLight = new(Light)
 )
 
 const (
@@ -140,7 +140,7 @@ func (l *Light) Verify(block pow.Block) bool {
 	// the finalizer before the call completes.
 	_ = cache
 	// The actual check.
-	target := new(big.Int).Div(minDifficulty, difficulty)
+	target := new(big.Int).Div(maxUint256, difficulty)
 	return h256ToHash(ret.result).Big().Cmp(target) <= 0
 }
 
@@ -199,7 +199,7 @@ func (d *dag) generate() {
 		if d.dir == "" {
 			d.dir = DefaultDir
 		}
-		glog.V(logger.Info).Infof("Generating DAG for epoch %d (%x)", d.epoch, seedHash)
+		glog.V(logger.Info).Infof("Generating DAG for epoch %d (size %d) (%x)", d.epoch, dagSize, seedHash)
 		// Generate a temporary cache.
 		// TODO: this could share the cache with Light
 		cache := C.ethash_light_new_internal(cacheSize, (*C.ethash_h256_t)(unsafe.Pointer(&seedHash[0])))
@@ -220,14 +220,18 @@ func (d *dag) generate() {
 	})
 }
 
-func freeDAG(h *dag) {
-	C.ethash_full_delete(h.ptr)
-	h.ptr = nil
+func freeDAG(d *dag) {
+	C.ethash_full_delete(d.ptr)
+	d.ptr = nil
+}
+
+func (d *dag) Ptr() unsafe.Pointer {
+	return unsafe.Pointer(d.ptr.data)
 }
 
 //export ethashGoCallback
 func ethashGoCallback(percent C.unsigned) C.int {
-	glog.V(logger.Info).Infof("Still generating DAG: %d%%", percent)
+	glog.V(logger.Info).Infof("Generating DAG: %d%%", percent)
 	return 0
 }
 
@@ -273,7 +277,7 @@ func (pow *Full) getDAG(blockNum uint64) (d *dag) {
 	return d
 }
 
-func (pow *Full) Search(block pow.Block, stop <-chan struct{}) (nonce uint64, mixDigest []byte) {
+func (pow *Full) Search(block pow.Block, stop <-chan struct{}, index int) (nonce uint64, mixDigest []byte) {
 	dag := pow.getDAG(block.NumberU64())
 
 	r := rand.New(rand.NewSource(time.Now().UnixNano()))
@@ -286,7 +290,7 @@ func (pow *Full) Search(block pow.Block, stop <-chan struct{}) (nonce uint64, mi
 
 	nonce = uint64(r.Int63())
 	hash := hashToH256(block.HashNoNonce())
-	target := new(big.Int).Div(minDifficulty, diff)
+	target := new(big.Int).Div(maxUint256, diff)
 	for {
 		select {
 		case <-stop:

Godeps/_workspace/src/github.com/ethereum/ethash/ethash_test.go

@@ -92,7 +92,7 @@ func TestEthashConcurrentVerify(t *testing.T) {
 	defer os.RemoveAll(eth.Full.Dir)
 
 	block := &testBlock{difficulty: big.NewInt(10)}
-	nonce, md := eth.Search(block, nil)
+	nonce, md := eth.Search(block, nil, 0)
 	block.nonce = nonce
 	block.mixDigest = common.BytesToHash(md)
 
@@ -135,7 +135,7 @@ func TestEthashConcurrentSearch(t *testing.T) {
 	// launch n searches concurrently.
 	for i := 0; i < nsearch; i++ {
 		go func() {
-			nonce, md := eth.Search(block, stop)
+			nonce, md := eth.Search(block, stop, 0)
 			select {
 			case found <- searchRes{n: nonce, md: md}:
 			case <-stop:
@@ -167,7 +167,7 @@ func TestEthashSearchAcrossEpoch(t *testing.T) {
 	for i := epochLength - 40; i < epochLength+40; i++ {
 		block := &testBlock{number: i, difficulty: big.NewInt(90)}
 		rand.Read(block.hashNoNonce[:])
-		nonce, md := eth.Search(block, nil)
+		nonce, md := eth.Search(block, nil, 0)
 		block.nonce = nonce
 		block.mixDigest = common.BytesToHash(md)
 		if !eth.Verify(block) {

Makefile

@@ -6,7 +6,7 @@
 GOBIN = build/bin
 
 geth:
-	build/env.sh go install -v $(shell build/ldflags.sh) ./cmd/geth
+	build/env.sh go install -v $(shell build/flags.sh) ./cmd/geth
 	@echo "Done building."
 	@echo "Run \"$(GOBIN)/geth\" to launch geth."
 
@@ -39,12 +39,12 @@ evm:
 	@echo "Done building."
 	@echo "Run \"$(GOBIN)/evm to start the evm."
 mist:
-	build/env.sh go install -v $(shell build/ldflags.sh) ./cmd/mist
+	build/env.sh go install -v $(shell build/flags.sh) ./cmd/mist
 	@echo "Done building."
 	@echo "Run \"$(GOBIN)/mist --asset_path=cmd/mist/assets\" to launch mist."
 
 all:
-	build/env.sh go install -v $(shell build/ldflags.sh) ./...
+	build/env.sh go install -v $(shell build/flags.sh) ./...
 
 test: all
 	build/env.sh go test ./...

build/ldflags.sh → build/flags.sh

@@ -16,3 +16,7 @@ sep=$(go version | awk '{ if ($3 >= "go1.5" || index($3, "devel")) print "="; el
 if [ -f ".git/HEAD" ]; then
     echo "-ldflags '-X main.gitCommit$sep$(git rev-parse HEAD)'"
 fi
+
+if [ ! -z "$GO_OPENCL" ]; then
+   echo "-tags opencl"
+fi

cmd/geth/js_test.go

@@ -468,8 +468,7 @@ func processTxs(repl *testjethre, t *testing.T, expTxc int) bool {
 		t.Errorf("incorrect number of pending transactions, expected %v, got %v", expTxc, txc)
 		return false
 	}
-
-	err = repl.ethereum.StartMining(runtime.NumCPU())
+	err = repl.ethereum.StartMining(runtime.NumCPU(), "")
 	if err != nil {
 		t.Errorf("unexpected error mining: %v", err)
 		return false

cmd/utils/flags.go

@@ -155,6 +155,12 @@ var (
 	}
 
 	// miner settings
+	// TODO: refactor CPU vs GPU mining flags
+	MiningGPUFlag = cli.StringFlag{
+		Name:  "minegpu",
+		Usage: "Mine with given GPUs. '--minegpu 0,1' will mine with the first two GPUs found.",
+	}
+
 	MinerThreadsFlag = cli.IntFlag{
 		Name:  "minerthreads",
 		Usage: "Number of miner threads",

common/natspec/natspec_e2e_test.go

@@ -306,7 +306,7 @@ func processTxs(repl *testFrontend, t *testing.T, expTxc int) bool {
 		return false
 	}
 
-	err = repl.ethereum.StartMining(runtime.NumCPU())
+	err = repl.ethereum.StartMining(runtime.NumCPU(), "")
 	if err != nil {
 		t.Errorf("unexpected error mining: %v", err)
 		return false

core/chain_makers.go

@@ -32,7 +32,7 @@ import (
 // It returns true from Verify for any block.
 type FakePow struct{}
 
-func (f FakePow) Search(block pow.Block, stop <-chan struct{}) (uint64, []byte) {
+func (f FakePow) Search(block pow.Block, stop <-chan struct{}, index int) (uint64, []byte) {
 	return 0, nil
 }
 func (f FakePow) Verify(block pow.Block) bool { return true }

core/chain_pow_test.go

@@ -34,7 +34,7 @@ type failPow struct {
 	failing uint64
 }
 
-func (pow failPow) Search(pow.Block, <-chan struct{}) (uint64, []byte) {
+func (pow failPow) Search(pow.Block, <-chan struct{}, int) (uint64, []byte) {
 	return 0, nil
 }
 func (pow failPow) Verify(block pow.Block) bool { return block.NumberU64() != pow.failing }
@@ -47,7 +47,7 @@ type delayedPow struct {
 	delay time.Duration
 }
 
-func (pow delayedPow) Search(pow.Block, <-chan struct{}) (uint64, []byte) {
+func (pow delayedPow) Search(pow.Block, <-chan struct{}, int) (uint64, []byte) {
 	return 0, nil
 }
 func (pow delayedPow) Verify(block pow.Block) bool { time.Sleep(pow.delay); return true }

miner/agent.go

@@ -118,7 +118,7 @@ func (self *CpuAgent) mine(work *Work, stop <-chan struct{}) {
 	glog.V(logger.Debug).Infof("(re)started agent[%d]. mining...\n", self.index)
 
 	// Mine
-	nonce, mixDigest := self.pow.Search(work.Block, stop)
+	nonce, mixDigest := self.pow.Search(work.Block, stop, self.index)
 	if nonce != 0 {
 		block := work.Block.WithMiningResult(nonce, common.BytesToHash(mixDigest))
 		self.returnCh <- &Result{work, block}