Merge pull request #120 from torch/goodies

adds Sort, revamps TensorMath, adds masked* operations

FFI.lua

@@ -7,6 +7,7 @@ typedef struct THCState
 {
   struct THCRNGState* rngState;
   struct THCBlasState* blasState;
+  struct cudaDeviceProp* deviceProperties;
 } THCState;
 
 typedef struct THCudaStorage

Tensor.c

@@ -5,21 +5,6 @@
 
 /* everything is as the generic Storage.c, except few things (see below) */
 
-static void THCudaTensor_maskedFill(THCState *state, THCudaTensor *tensor, THByteTensor *mask, float value)
-{
-  THError("not yet implemented for CUDA");
-}
-
-static void THCudaTensor_maskedCopy(THCState *state, THCudaTensor *tensor, THByteTensor *mask, THCudaTensor* src)
-{
-  THError("not yet implemented for CUDA");
-}
-
-void THCudaTensor_maskedSelect(THCState *state, THCudaTensor *tensor, THCudaTensor* src, THByteTensor *mask)
-{
-  THError("not yet implemented for CUDA");
-}
-
 #define real float
 #define Real Cuda
 

Tensor.lua

@@ -30,14 +30,44 @@ local function Tensor__float(self,type)
    return self:type('torch.FloatTensor')
 end
 
+local function Tensor__byte(self,type)
+   return self:type('torch.ByteTensor')
+end
+
+local function Tensor__char(self,type)
+   return self:type('torch.CharTensor')
+end
+
+local function Tensor__int(self,type)
+   return self:type('torch.IntTensor')
+end
+
+local function Tensor__short(self,type)
+   return self:type('torch.ShortTensor')
+end
+
+local function Tensor__Long(self,type)
+   return self:type('torch.LongTensor')
+end
+
 rawset(torch.getmetatable('torch.DoubleTensor'), 'cuda', Tensor__cuda)
 rawset(torch.getmetatable('torch.FloatTensor'), 'cuda', Tensor__cuda)
+rawset(torch.getmetatable('torch.ByteTensor'), 'cuda', Tensor__cuda)
+rawset(torch.getmetatable('torch.CharTensor'), 'cuda', Tensor__cuda)
+rawset(torch.getmetatable('torch.IntTensor'), 'cuda', Tensor__cuda)
+rawset(torch.getmetatable('torch.ShortTensor'), 'cuda', Tensor__cuda)
+rawset(torch.getmetatable('torch.LongTensor'), 'cuda', Tensor__cuda)
 rawset(torch.getmetatable('torch.CudaTensor'), 'cuda', Tensor__cuda)
 
 rawset(torch.getmetatable('torch.CudaTensor'), 'type', Tensor__type)
 rawset(torch.getmetatable('torch.CudaTensor'), 'typeAs', Tensor__typeAs)
 rawset(torch.getmetatable('torch.CudaTensor'), 'double', Tensor__double)
 rawset(torch.getmetatable('torch.CudaTensor'), 'float', Tensor__float)
+rawset(torch.getmetatable('torch.CudaTensor'), 'byte', Tensor__byte)
+rawset(torch.getmetatable('torch.CudaTensor'), 'char', Tensor__char)
+rawset(torch.getmetatable('torch.CudaTensor'), 'int', Tensor__int)
+rawset(torch.getmetatable('torch.CudaTensor'), 'short', Tensor__short)
+rawset(torch.getmetatable('torch.CudaTensor'), 'long', Tensor__long)
 
 do
     local metatable = torch.getmetatable('torch.CudaTensor')

TensorMath.lua

@@ -169,7 +169,7 @@ end
 
 local function lastdim(argn)
    return function(arg)
-      return string.format("THCudaTensor_nDimension(%s)", arg.args[argn]:carg())
+      return string.format("THCudaTensor_nDimension(cutorch_getstate(L), %s)", arg.args[argn]:carg())
    end
 end
 
@@ -289,6 +289,24 @@ wrap("addcdiv",
         {name=Tensor},
         {name=Tensor}})
 
+wrap("maskedFill",
+     cname("maskedFill"),
+     {{name=Tensor, returned=true, method={default='nil'}},
+      {name=Tensor},
+      {name=real}})
+
+wrap("maskedCopy",
+     cname("maskedCopy"),
+     {{name=Tensor, returned=true, method={default='nil'}},
+	{name=Tensor},
+	{name=Tensor}})
+
+wrap("maskedSelect",
+     cname("maskedSelect"),
+     {{name=Tensor, returned=true, default=true},
+      {name=Tensor},
+      {name=Tensor}})
+
 do
    local Tensor = Tensor
    local real = real

init.c

@@ -40,6 +40,15 @@ static int cutorch_getDeviceCount(lua_State *L)
   return 1;
 }
 
+static int cutorch_getMemoryUsage(lua_State *L) {
+  size_t freeBytes = 0;
+  size_t totalBytes = 0;
+  THCudaCheck(cudaMemGetInfo(&freeBytes, &totalBytes));
+  lua_pushnumber(L, freeBytes);
+  lua_pushnumber(L, totalBytes);
+  return 2;
+}
+
 static int cutorch_setDevice(lua_State *L)
 {
   THCState *state = cutorch_getstate(L);
@@ -158,6 +167,7 @@ static const struct luaL_Reg cutorch_stuff__ [] = {
   {"deviceReset", cutorch_deviceReset},
   {"getDeviceCount", cutorch_getDeviceCount},
   {"getDeviceProperties", cutorch_getDeviceProperties},
+  {"getMemoryUsage", cutorch_getMemoryUsage},
   {"setDevice", cutorch_setDevice},
   {"seed", cutorch_seed},
   {"seedAll", cutorch_seedAll},

lib/THC/CMakeLists.txt

@@ -1,7 +1,28 @@
 SET(src
     THCGeneral.c THCStorage.c THCStorageCopy.c THCTensor.c THCTensorCopy.c)
 
-SET(src-cuda THC.cu)
+SET(src-cuda
+  THCReduceApplyUtils.cu
+  THCBlas.cu
+  THCStorage.cu
+  THCStorageCopy.cu
+  THCTensor.cu
+  THCTensorCopy.cu
+  THCTensorMath2.cu
+  THCTensorMathBlas.cu
+  THCTensorMathCompare.cu
+  THCTensorMathCompareT.cu
+  THCTensorMath.cu
+  THCTensorMathPairwise.cu
+  THCTensorMathPointwise.cu
+  THCTensorMathScan.cu
+  THCTensorMathTransformReduce.cu
+  THCTensorMasked.cu
+  THCTensorIndex.cu
+  THCTensorConv.cu
+  THCTensorRandom.cu
+  THCApply.cu
+  )
 
 CUDA_ADD_LIBRARY(THC SHARED ${src} ${src-cuda})
 CUDA_ADD_CUBLAS_TO_TARGET(THC)
@@ -23,4 +44,7 @@ INSTALL(FILES
           THCTensorRandom.h
           THCTensorMath.h
           THCTensorConv.h
+          THCApply.cuh
+          THCReduce.cuh
+          THCReduceApplyUtils.cuh
           DESTINATION "${Torch_INSTALL_INCLUDE_SUBDIR}/THC")

lib/THC/THCApply.cu

@@ -0,0 +1,10 @@
+#include "THCApply.cuh"
+
+// Implementation of copyIgnoringOverlaps, defined after pointwiseApply2.
+void THCudaTensor_copyIgnoringOverlaps(THCState* state,
+                                       THCudaTensor* dst,
+                                       THCudaTensor* src) {
+  THCudaTensor_pointwiseApply2(state, dst, src, CopyOp<float>(),
+                               ReadOnly, // ignore overwrites
+                               ReadOnly);
+}