[Beignet] [PATCH] fix scalarizing of llvm phi node

[Zhigang Gong]

LGTM, will push it latter. Thanks.

On Tue, Sep 24, 2013 at 11:11:21AM +0800, Homer Hsing wrote:
> llvm phi node can have odd number of args.
> this patch also contains a test case.
> 
> Signed-off-by: Homer Hsing <homer.xing at intel.com>
> ---
>  backend/src/llvm/llvm_scalarize.cpp |  1 -
>  kernels/compiler_vector_inc.cl      | 13 +++++++++++
>  utests/CMakeLists.txt               |  1 +
>  utests/compiler_vector_inc.cpp      | 46 +++++++++++++++++++++++++++++++++++++
>  4 files changed, 60 insertions(+), 1 deletion(-)
>  create mode 100644 kernels/compiler_vector_inc.cl
>  create mode 100644 utests/compiler_vector_inc.cpp
> 
> diff --git a/backend/src/llvm/llvm_scalarize.cpp b/backend/src/llvm/llvm_scalarize.cpp
> index 41674b6..7a40616 100644
> --- a/backend/src/llvm/llvm_scalarize.cpp
> +++ b/backend/src/llvm/llvm_scalarize.cpp
> @@ -383,7 +383,6 @@ namespace gbe {
>  
>      if (PHINode* phi = dyn_cast<PHINode>(inst)) {
>        PHINode* res = PHINode::Create(GetBasicType(inst), phi->getNumIncomingValues());
> -      assert(args.size() % 2 == 0 && "Odd number of arguments for a PHI");
>  
>        // Loop over pairs of operands: [Value*, BasicBlock*]
>        for (unsigned int i = 0; i < args.size(); i++) {
> diff --git a/kernels/compiler_vector_inc.cl b/kernels/compiler_vector_inc.cl
> new file mode 100644
> index 0000000..548dcb4
> --- /dev/null
> +++ b/kernels/compiler_vector_inc.cl
> @@ -0,0 +1,13 @@
> +kernel void compiler_vector_inc(global char *dst, global char *src) {
> +    size_t i = get_global_id(0);
> +    char2 dst2 = vload2(i, dst);
> +    if (src[i] == 0)
> +      dst2++;
> +    else if(src[i] == 1)
> +      ++dst2;
> +    else if(src[i] == 2)
> +      dst2--;
> +    else
> +      --dst2;
> +    vstore2(dst2, i, dst);
> +}
> diff --git a/utests/CMakeLists.txt b/utests/CMakeLists.txt
> index 24322b6..7e3c3ee 100644
> --- a/utests/CMakeLists.txt
> +++ b/utests/CMakeLists.txt
> @@ -102,6 +102,7 @@ set (utests_sources
>    compiler_get_image_info.cpp
>    compiler_vect_compare.cpp
>    compiler_vector_load_store.cpp
> +  compiler_vector_inc.cpp
>    compiler_cl_finish.cpp
>    get_cl_info.cpp
>    builtin_atan2.cpp
> diff --git a/utests/compiler_vector_inc.cpp b/utests/compiler_vector_inc.cpp
> new file mode 100644
> index 0000000..abc5408
> --- /dev/null
> +++ b/utests/compiler_vector_inc.cpp
> @@ -0,0 +1,46 @@
> +#include <cstdint>
> +#include <cstring>
> +#include <iostream>
> +#include "utest_helper.hpp"
> +
> +void compiler_vector_inc(void)
> +{
> +  const int n = 64;
> +  char dst[n];
> +  char src[n];
> +
> +  OCL_CREATE_KERNEL("compiler_vector_inc");
> +  OCL_CREATE_BUFFER(buf[0], 0, n, NULL);
> +  OCL_CREATE_BUFFER(buf[1], 0, n, NULL);
> +  OCL_SET_ARG(0, sizeof(cl_mem), &buf[0]);
> +  OCL_SET_ARG(1, sizeof(cl_mem), &buf[1]);
> +  globals[0] = n / 2;
> +  locals[0] = 16;
> +
> +  for (int i = 0; i < n; ++i) {
> +    dst[i] = i;
> +    src[i] = (i / 2) % 4;
> +  }
> +  OCL_MAP_BUFFER(0);
> +  OCL_MAP_BUFFER(1);
> +  memcpy(buf_data[0], dst, n);
> +  memcpy(buf_data[1], src, n);
> +  OCL_UNMAP_BUFFER(0);
> +  OCL_UNMAP_BUFFER(1);
> +
> +  OCL_NDRANGE(1);
> +
> +  OCL_MAP_BUFFER(0);
> +  char *dest = ((char *)buf_data[0]);
> +  for (int i=0; i<n; ++i) {
> +    char wish;
> +    if (src[i/2] < 2)
> +      wish = dst[i] + 1;
> +    else
> +      wish = dst[i] - 1;
> +    OCL_ASSERT(dest[i] == wish);
> +  }
> +  OCL_UNMAP_BUFFER(0);
> +}
> +
> +MAKE_UTEST_FROM_FUNCTION(compiler_vector_inc);
> -- 
> 1.8.1.2
> 
> _______________________________________________
> Beignet mailing list
> Beignet at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/beignet