resolve merge conflict

hipamd/include/hip/hcc_detail/functional_grid_launch.hpp

@@ -154,6 +154,20 @@ hipError_t hipOccupancyMaxPotentialBlockSize(uint32_t* gridSize, uint32_t* block
                                       dynSharedMemPerBlk, blockSizeLimit);
 }
 
+template <typename F>
+inline
+hipError_t hipOccupancyMaxActiveBlocksPerMultiprocessor(uint32_t* numBlocks, F kernel,
+    uint32_t blockSize, size_t dynSharedMemPerBlk) {
+
+    using namespace hip_impl;
+
+    hip_impl::hip_init();
+    auto f = get_program_state().kernel_descriptor(reinterpret_cast<std::uintptr_t>(kernel),
+                                                   target_agent(0));
+
+    return hipOccupancyMaxActiveBlocksPerMultiprocessor(numBlocks, f, blockSize, dynSharedMemPerBlk);
+}
+
 template <typename... Args, typename F = void (*)(Args...)>
 inline
 void hipLaunchKernelGGL(F kernel, const dim3& numBlocks, const dim3& dimBlocks,

hipamd/include/hip/hcc_detail/hip_runtime_api.h

@@ -2953,7 +2953,7 @@ hipError_t hipOccupancyMaxPotentialBlockSize(uint32_t* gridSize, uint32_t* block
  * @param [in]  dynSharedMemPerBlk dynamic shared memory usage (in bytes) intended for each block
  */
 hipError_t hipOccupancyMaxActiveBlocksPerMultiprocessor(
-   int* numBlocks, const void* f, int blockSize, size_t dynSharedMemPerBlk);
+   uint32_t* numBlocks, hipFunction_t f, uint32_t blockSize, size_t dynSharedMemPerBlk);
 
 /**
  * @brief Returns occupancy for a device function.
@@ -2964,8 +2964,7 @@ hipError_t hipOccupancyMaxActiveBlocksPerMultiprocessor(
  * @param [in]  dynSharedMemPerBlk dynamic shared memory usage (in bytes) intended for each block
  * @param [in]  flags            Extra flags for occupancy calculation (currently ignored)
  */
-hipError_t hipOccupancyMaxActiveBlocksPerMultiprocessorWithFlags(
-   int* numBlocks, const void* f, int blockSize, size_t dynSharedMemPerBlk, unsigned int flags __dparm(hipOccupancyDefault));
+   uint32_t* numBlocks, hipFunction_t f, uint32_t blockSize, size_t dynSharedMemPerBlk, unsigned int flags __dparm(hipOccupancyDefault));
 
 #if __HIP_VDI__ && !defined(__HCC__)
 /**
@@ -3235,6 +3234,21 @@ hipError_t hipLaunchKernel(const void* function_address,
 } /* extern "c" */
 #endif
 
+#if defined(__cplusplus) && !defined(__HCC__) && defined(__clang__) && defined(__HIP__)
+template <typename F>
+static hipError_t __host__ inline hipOccupancyMaxActiveBlocksPerMultiprocessor(
+    uint32_t* numBlocks, F func, uint32_t blockSize, size_t dynSharedMemPerBlk) {
+    return ::hipOccupancyMaxActiveBlocksPerMultiprocessor(numBlocks, (hipFunction_t)func, blockSize,
+                                                          dynSharedMemPerBlk);
+}
+template <typename F>
+static hipError_t __host__ inline hipOccupancyMaxActiveBlocksPerMultiprocessorWithFlags(
+    uint32_t* numBlocks, F func, uint32_t blockSize, size_t dynSharedMemPerBlk, unsigned int flags) {
+    return ::hipOccupancyMaxActiveBlocksPerMultiprocessorWithFlags(
+        numBlocks, (hipFunction_t)func, blockSize, dynSharedMemPerBlk, flags);
+}
+#endif  // defined(__cplusplus) && !defined(__HCC__) && defined(__clang__) && defined(__HIP__)
+
 #if USE_PROF_API
 #include <hip/hcc_detail/hip_prof_str.h>
 #endif
@@ -3375,20 +3389,6 @@ hipError_t hipBindTextureToMipmappedArray(const texture<T, dim, readMode>& tex,
     return hipSuccess;
 }
 
-template <class T>
-inline hipError_t hipOccupancyMaxActiveBlocksPerMultiprocessor(
-    int* numBlocks, T f, int blockSize, size_t dynSharedMemPerBlk) {
-    return hipOccupancyMaxActiveBlocksPerMultiprocessor(
-        numBlocks, reinterpret_cast<const void*>(f), blockSize, dynSharedMemPerBlk);
-}
-
-template <class T>
-inline hipError_t hipOccupancyMaxActiveBlocksPerMultiprocessorWithFlags(
-    int* numBlocks, T f, int blockSize, size_t dynSharedMemPerBlk, unsigned int flags) {
-    return hipOccupancyMaxActiveBlocksPerMultiprocessorWithFlags(
-        numBlocks, reinterpret_cast<const void*>(f), blockSize, dynSharedMemPerBlk, flags);
-}
-
 #if __HIP_VDI__ && !defined(__HCC__)
 template <class T>
 inline hipError_t hipLaunchCooperativeKernel(T f, dim3 gridDim, dim3 blockDim,

hipamd/include/hip/hcc_detail/hip_texture_types.h

@@ -57,6 +57,7 @@ struct __HIP_TEXTURE_ATTRIB texture : public textureReference {
     texture(int norm = 0, enum hipTextureFilterMode fMode = hipFilterModePoint,
             enum hipTextureAddressMode aMode = hipAddressModeClamp) {
         normalized = norm;
+        readMode = hipReadModeNormalizedFloat;
         filterMode = fMode;
         addressMode[0] = aMode;
         addressMode[1] = aMode;
@@ -68,6 +69,7 @@ struct __HIP_TEXTURE_ATTRIB texture : public textureReference {
     texture(int norm, enum hipTextureFilterMode fMode, enum hipTextureAddressMode aMode,
             struct hipChannelFormatDesc desc) {
         normalized = norm;
+        readMode = hipReadModeNormalizedFloat;
         filterMode = fMode;
         addressMode[0] = aMode;
         addressMode[1] = aMode;

hipamd/include/hip/hcc_detail/hip_vector_types.h

@@ -34,11 +34,11 @@ THE SOFTWARE.
 
 #include "hip/hcc_detail/host_defines.h"
 
-#if !defined(_MSC_VER)|| __clang__
+#if !defined(_MSC_VER) || __clang__
     #if __has_attribute(ext_vector_type)
-        #define __NATIVE_VECTOR__(n, ...) __attribute__((ext_vector_type(n)))
+        #define __NATIVE_VECTOR__(n, T) T __attribute__((ext_vector_type(n)))
     #else
-        #define __NATIVE_VECTOR__(n, ...) [n]
+        #define __NATIVE_VECTOR__(n, T) T[n]
     #endif
 
 #if defined(__cplusplus)
@@ -49,16 +49,16 @@ THE SOFTWARE.
     namespace hip_impl {
         template<typename, typename, unsigned int> struct Scalar_accessor;
     } // Namespace hip_impl.
-    
+
     namespace std {
         template<typename T, typename U, unsigned int n>
         struct is_integral<hip_impl::Scalar_accessor<T, U, n>>
             : is_integral<T> {};
         template<typename T, typename U, unsigned int n>
-        struct is_floating_point<hip_impl::Scalar_accessor<T, U, n>> 
+        struct is_floating_point<hip_impl::Scalar_accessor<T, U, n>>
             : is_floating_point<T> {};
     } // Namespace std.
-    
+
     namespace hip_impl {
         template<typename T, typename Vector, unsigned int idx>
         struct Scalar_accessor {
@@ -146,7 +146,7 @@ THE SOFTWARE.
 
             __host__ __device__
             Address operator&() const noexcept { return Address{this}; }
-            
+
             __host__ __device__
             Scalar_accessor& operator=(const Scalar_accessor& x) noexcept {
                 data[idx] = x.data[idx];
@@ -284,13 +284,20 @@ THE SOFTWARE.
                 return *this;
             }
         };
+
+        inline
+        constexpr
+        unsigned int next_pot(unsigned int x) {
+            // Precondition: x > 1.
+	        return 1u << (32u - __builtin_clz(x - 1u));
+        }
     } // Namespace hip_impl.
 
     template<typename T, unsigned int n> struct HIP_vector_base;
 
     template<typename T>
     struct HIP_vector_base<T, 1> {
-        using Native_vec_ = T __NATIVE_VECTOR__(1, T);
+        using Native_vec_ = __NATIVE_VECTOR__(1, T);
 
         union {
             Native_vec_ data;
@@ -304,7 +311,7 @@ THE SOFTWARE.
         };
 
         using value_type = T;
-        
+
         __host__ __device__
         HIP_vector_base& operator=(const HIP_vector_base& x) noexcept {
             #if __has_attribute(ext_vector_type)
@@ -319,9 +326,13 @@ THE SOFTWARE.
 
     template<typename T>
     struct HIP_vector_base<T, 2> {
-        using Native_vec_ = T __NATIVE_VECTOR__(2, T);
+        using Native_vec_ = __NATIVE_VECTOR__(2, T);
 
-        union {
+        union
+        #if !__has_attribute(ext_vector_type)
+            alignas(hip_impl::next_pot(2 * sizeof(T)))
+        #endif
+        {
             Native_vec_ data;
 #if __HIP_CLANG_ONLY__
             struct {
@@ -335,7 +346,7 @@ THE SOFTWARE.
         };
 
         using value_type = T;
-        
+
         __host__ __device__
         HIP_vector_base& operator=(const HIP_vector_base& x) noexcept {
             #if __has_attribute(ext_vector_type)
@@ -507,9 +518,13 @@ THE SOFTWARE.
 
     template<typename T>
     struct HIP_vector_base<T, 4> {
-        using Native_vec_ = T __NATIVE_VECTOR__(4, T);
+        using Native_vec_ = __NATIVE_VECTOR__(4, T);
 
-        union {
+        union
+        #if !__has_attribute(ext_vector_type)
+            alignas(hip_impl::next_pot(4 * sizeof(T)))
+        #endif
+        {
             Native_vec_ data;
 #if __HIP_CLANG_ONLY__
             struct {
@@ -527,7 +542,7 @@ THE SOFTWARE.
         };
 
         using value_type = T;
-        
+
         __host__ __device__
         HIP_vector_base& operator=(const HIP_vector_base& x) noexcept {
             #if __has_attribute(ext_vector_type)

hipamd/include/hip/hcc_detail/texture_functions.h

@@ -157,32 +157,16 @@ union TData {
 
 #define TEXTURE_RETURN_UINT_XYZW return make_uint4(texel.u.x, texel.u.y, texel.u.z, texel.u.w);
 
-#define TEXTURE_RETURN_FLOAT return (texFormatToSize[texRef.format] == 1)? texel.f.x : (float)texel.u.x/texFormatToSize[texRef.format];
+#define TEXTURE_RETURN_FLOAT return texel.f.x;
 
-#define TEXTURE_RETURN_FLOAT_X return (texFormatToSize[texRef.format] == 1)? make_float1(texel.f.x) : make_float1((float)texel.u.x/texFormatToSize[texRef.format]);
+#define TEXTURE_RETURN_FLOAT_X return make_float1(texel.f.x);
 
-#define TEXTURE_RETURN_FLOAT_XY return (texFormatToSize[texRef.format] == 1)? make_float2(texel.f.x, texel.f.y) : make_float2((float)texel.u.x/texFormatToSize[texRef.format], (float)texel.u.y/texFormatToSize[texRef.format]);
+#define TEXTURE_RETURN_FLOAT_XY return make_float2(texel.f.x, texel.f.y);
 
-#define TEXTURE_RETURN_FLOAT_XYZW return (texFormatToSize[texRef.format] == 1)? make_float4(texel.f.x, texel.f.y, texel.f.z, texel.f.w) : make_float4((float)texel.u.x/texFormatToSize[texRef.format], (float)texel.u.y/texFormatToSize[texRef.format], (float)texel.u.z/texFormatToSize[texRef.format], (float)texel.u.w/texFormatToSize[texRef.format]) ;
+#define TEXTURE_RETURN_FLOAT_XYZW return make_float4(texel.f.x, texel.f.y, texel.f.z, texel.f.w);
 
 extern "C" {
 
-// this is really a sparse array with only valid values being the ones indexed by the enum hipArray_Format(e.g. texFormatToSize[HIP_AD_FORMAT_UNSIGNED_INT8] = UCHAR_MAX)
-__device__ __constant__ static int texFormatToSize[] = {
-    1, /* HIP_AD_FORMAT_NOT_INITIALIZED */
-    UCHAR_MAX, /* HIP_AD_FORMAT_UNSIGNED_INT8 */
-    USHRT_MAX, /* HIP_AD_FORMAT_UNSIGNED_INT16 */
-    1, /* HIP_AD_FORMAT_UNSIGNED_INT32 */
-    1,1,1,1, /* Invalid values */
-    SCHAR_MAX, /* HIP_AD_FORMAT_SIGNED_INT8 */
-    SHRT_MAX, /* HIP_AD_FORMAT_SIGNED_INT16 */
-    1, /* HIP_AD_FORMAT_SIGNED_INT32 */
-    1,1,1,1,1, /* Invalid values */
-    1, /* HIP_AD_FORMAT_HALF */
-    1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* Invalid values */
-    1 /* HIP_AD_FORMAT_FLOAT */
-};
-
 __device__
 __hip_float4_vector_value_type __ockl_image_sample_1D(
     unsigned int ADDRESS_SPACE_CONSTANT* i, unsigned int ADDRESS_SPACE_CONSTANT* s,

hipamd/include/hip/hcc_detail/texture_types.h

@@ -73,6 +73,7 @@ enum hipTextureReadMode { hipReadModeElementType = 0, hipReadModeNormalizedFloat
  */
 typedef struct textureReference {
     int normalized;
+    enum hipTextureReadMode readMode;// used only for driver API's
     enum hipTextureFilterMode filterMode;
     enum hipTextureAddressMode addressMode[3];  // Texture address mode for up to 3 dimensions
     struct hipChannelFormatDesc channelDesc;