clean

driver/driver.hip.cpp

@@ -610,10 +610,10 @@ int main(int argc, char* argv[])
 #elif 1
     // 1x1 filter, 14x14 image, C = 512
     constexpr index_t N  = 128;
-    constexpr index_t C  = 128;
+    constexpr index_t C  = 512;
     constexpr index_t HI = 14;
     constexpr index_t WI = 14;
-    constexpr index_t K  = 128;
+    constexpr index_t K  = 512;
     constexpr index_t Y  = 1;
     constexpr index_t X  = 1;
 

src/include/amd_inline_asm.hip.hpp

@@ -13,7 +13,7 @@ extern "C" __attribute__((address_space(3))) void* __to_local(void* p)[[hc]];
 #define data4_t vector_type<float, 4>::MemoryType
 #define data_t float
 
-template<unsigned cnt>
+    template<unsigned cnt>
 inline __device__ void vmcnt()
 {
 #if !NO_VM_WAIT
@@ -111,8 +111,8 @@ inline __device__ void outerProduct1x4(const data_t& a,
             "v"(a.x),"v"(b.x),"v"(b.y),"v"(b.z),"v"(b.w)
             );
 #else
-//hijack here due to a compiler issue that cannot perform proper register
-//mapping for float4 c
+    //hijack here due to a compiler issue that cannot perform proper register
+    //mapping for float4 c
     data_t *c_p = (data_t *)&c;
     asm volatile("\n \
             v_mac_f32 %0, %4, %5 \n \
@@ -430,7 +430,7 @@ inline __device__ void global_load(data4_t& r,
 #endif
 }
 
-inline __device__ void
+    inline __device__ void
 ds_write_b128(const data4_t& r, void* lds, index_t offset = 0)
 {
 #if !NO_DS_WRITE