Fix compile.

Author: abergeron

src/gpuarray_array_blas.c

@@ -102,11 +102,11 @@ int GpuArray_rgemv(cb_transpose transA, double alpha, GpuArray *A,
   GpuArray *Xp = X;
   GpuArray copyX;
   GpuArray *Yp = Y;
+  gpucontext *ctx = gpudata_context(Ap->data);
   size_t elsize;
   size_t m, n, lda;
   cb_order o;
   int err;
-  void *ctx = gpudata_context(Ap->data);
 
   if (A->typecode != GA_HALF &&
       A->typecode != GA_FLOAT &&
@@ -205,7 +205,7 @@ int GpuArray_rgemm(cb_transpose transA, cb_transpose transB, double alpha,
   GpuArray *Bp = B;
   GpuArray copyB;
   GpuArray *Cp = C;
-  void *ctx = gpudata_context(Ap->data);
+  gpucontext *ctx = gpudata_context(Ap->data);
   size_t elsize;
   size_t m, n, k, lda, ldb, ldc;
   cb_order o;
@@ -355,7 +355,7 @@ int GpuArray_rger(double alpha, GpuArray *X, GpuArray *Y, GpuArray *A,
   GpuArray *Yp = Y;
   GpuArray copyY;
   GpuArray *Ap = A;
-  void *ctx = gpudata_context(Xp->data);
+  gpucontext *ctx = gpudata_context(Xp->data);
   size_t elsize;
   size_t m, n, lda;
   cb_order o;
@@ -372,12 +372,12 @@ int GpuArray_rger(double alpha, GpuArray *X, GpuArray *Y, GpuArray *A,
 
   if (!(X->flags & GA_ALIGNED) || !(Y->flags & GA_ALIGNED) ||
       !(A->flags & GA_ALIGNED))
-    return error_set(ctx->err, GA_UNALIGNED_ERROR, "Unaligned inputs";
+    return error_set(ctx->err, GA_UNALIGNED_ERROR, "Unaligned inputs");
 
   m = X->dimensions[0];
   n = Y->dimensions[0];
   if (A->dimensions[0] != m || A->dimensions[1] != n)
-    return error_set(ctx->err, GA_VALUE_ERROR, "Incompatible shapes";
+    return error_set(ctx->err, GA_VALUE_ERROR, "Incompatible shapes");
 
   elsize = gpuarray_get_elsize(X->typecode);
 
@@ -468,7 +468,7 @@ int GpuArray_rgemmBatch_3d(cb_transpose transA, cb_transpose transB, double alph
   GpuArray *Bp = B;
   GpuArray copyB;
   GpuArray *Cp = C;
-  void *ctx = gpudata_context(A->data);
+  gpucontext *ctx = gpudata_context(A->data);
   size_t elsize;
   size_t batchCount, m, n, k, lda, ldb, ldc;
   cb_order o;

src/gpuarray_array_collectives.c

@@ -49,6 +49,7 @@ static inline int check_gpuarrays(int times_src, const GpuArray* src,
 
 int GpuArray_reduce_from(const GpuArray* src, int opcode, int root,
                          gpucomm* comm) {
+  gpucontext *ctx = gpudata_context(src->data);
   size_t total_elems;
   if (!GpuArray_ISALIGNED(src))
     return error_set(ctx->err, GA_UNALIGNED_ERROR, "Unaligned input");
@@ -90,9 +91,11 @@ int GpuArray_reduce_scatter(const GpuArray* src, GpuArray* dest, int opcode,
                                 comm);
 }
 
-int GpuArray_broadcast(GpuArray* array, int root, gpucomm* comm) {
-  int rank = 0;
+int GpuArray_broadcast(GpuArray *array, int root, gpucomm *comm) {
+  gpucontext *ctx = gpudata_context(array->data);
   size_t total_elems;
+  int rank = 0;
+
   GA_CHECK(gpucomm_get_rank(comm, &rank));
   if (rank == root) {
     if (!GpuArray_CHKFLAGS(array, GA_BEHAVED))

src/gpuarray_buffer.c

@@ -23,7 +23,7 @@ const gpuarray_buffer_ops *gpuarray_get_ops(const char *name) {
 int gpu_get_platform_count(const char* name, unsigned int* platcount) {
   const gpuarray_buffer_ops* ops = gpuarray_get_ops(name);
   if (ops == NULL) {
-    return error_set(&global_err, GA_INVALID_ERROR, "Invalid platform");
+    return error_set(global_err, GA_INVALID_ERROR, "Invalid platform");
   }
   return ops->get_platform_count(platcount);
 }
@@ -32,7 +32,7 @@ int gpu_get_device_count(const char* name, unsigned int platform,
                          unsigned int* devcount) {
   const gpuarray_buffer_ops* ops = gpuarray_get_ops(name);
   if (ops == NULL) {
-    return error_set(&global_err, GA_INVALID_ERROR, "Invalid platform");
+    return error_set(global_err, GA_INVALID_ERROR, "Invalid platform");
   }
   return ops->get_device_count(platform, devcount);
 }
@@ -121,8 +121,10 @@ int gpudata_transfer(gpudata *dst, size_t dstoff, gpudata *src, size_t srcoff,
 
   /* Fallback to host copy */
   tmp = malloc(sz);
-  if (tmp == NULL)
-    return error_sys(ctx->err, "malloc");
+  if (tmp == NULL) {
+    error_sys(src_ctx->err, "malloc");
+    return error_sys(dst_ctx->err, "malloc");
+  }
   res = src_ctx->ops->buffer_read(tmp, src, srcoff, sz);
   if (res != GA_NO_ERROR) {
     free(tmp);

src/gpuarray_elemwise.c

@@ -272,6 +272,7 @@ static int check_basic(GpuElemwise *ge, void **args, int flags,
                        size_t *_n, unsigned int *_nd, size_t **_dims,
                        ssize_t ***_strides, int *_call32) {
   size_t n;
+  gpucontext *ctx = GpuKernel_context(&ge->k_contig);
   GpuArray *a = NULL, *v;
   unsigned int i, j, p, num_arrays = 0, nd = 0, nnd;
   int call32 = 1;
@@ -443,7 +444,7 @@ static int gen_elemwise_contig_kernel(GpuKernel *k,
 
   ktypes = calloc(p, sizeof(int));
   if (ktypes == NULL) {
-    res = error_fmt(ctx->err, "calloc");
+    res = error_sys(ctx->err, "calloc");
     goto bail;
   }
 
@@ -511,7 +512,7 @@ static int gen_elemwise_contig_kernel(GpuKernel *k,
   strb_appends(&sb, "}\n}\n");
 
   if (strb_error(&sb)) {
-    error_set(ctx->err, GA_MISC_ERROR, "Formatting error creating kernel source");
+    res = error_set(ctx->err, GA_MISC_ERROR, "Formatting error creating kernel source");
     goto bail;
   }
 

src/gpuarray_kernel.c

@@ -2,6 +2,9 @@
 #include "gpuarray/error.h"
 #include "gpuarray/types.h"
 
+#include "util/error.h"
+#include "private.h"
+
 #include <stdlib.h>
 
 int GpuKernel_init(GpuKernel *k, gpucontext *ctx, unsigned int count,