Improve NEON transpose performance

librapid/include/librapid/array/arrayContainer.hpp

@@ -943,6 +943,8 @@ namespace librapid {
 		}
 	} // namespace array
 
+	// template<typename ShapeType_, typename StorageType_>
+
 	namespace detail {
 		template<typename T>
 		struct IsArrayType {

librapid/include/librapid/array/linalg/transpose.hpp

@@ -168,8 +168,59 @@ namespace librapid {
 			_mm_storeu_pd(out + 1 * cols, _mm_mul_pd(tmp1Unpck, alphaVec));
 		}
 
-#	endif // LIBRAPID_MSVC
-#endif	   // LIBRAPID_NATIVE_ARCH
+#	elif defined(LIBRAPID_NEON)
+#		define LIBRAPID_F64_TRANSPOSE_KERNEL_SIZE 2
+#		define LIBRAPID_F32_TRANSPOSE_KERNEL_SIZE 4
+
+		template<typename Alpha>
+		LIBRAPID_ALWAYS_INLINE void transposeFloatKernel(float *__restrict out,
+														 float *__restrict in, Alpha alpha,
+														 int64_t cols) {
+			float32x4_t r0, r1, r2, r3;
+			float32x4_t t0, t1, t2, t3;
+
+			r0 = vld1q_f32(&in[0 * cols]);
+			r1 = vld1q_f32(&in[1 * cols]);
+			r2 = vld1q_f32(&in[2 * cols]);
+			r3 = vld1q_f32(&in[3 * cols]);
+
+			t0 = vzip1q_f32(r0, r1);
+			t1 = vzip2q_f32(r0, r1);
+			t2 = vzip1q_f32(r2, r3);
+			t3 = vzip2q_f32(r2, r3);
+
+			r0 = vcombine_f32(vget_low_f32(t0), vget_low_f32(t2));
+			r1 = vcombine_f32(vget_high_f32(t0), vget_high_f32(t2));
+			r2 = vcombine_f32(vget_low_f32(t1), vget_low_f32(t3));
+			r3 = vcombine_f32(vget_high_f32(t1), vget_high_f32(t3));
+
+			float32x4_t alphaVec = vdupq_n_f32(alpha);
+
+			vst1q_f32(&out[0 * cols], vmulq_f32(r0, alphaVec));
+			vst1q_f32(&out[1 * cols], vmulq_f32(r1, alphaVec));
+			vst1q_f32(&out[2 * cols], vmulq_f32(r2, alphaVec));
+			vst1q_f32(&out[3 * cols], vmulq_f32(r3, alphaVec));
+		}
+
+		template<typename Alpha>
+		LIBRAPID_ALWAYS_INLINE void transposeDoubleKernel(double *__restrict out,
+														  double *__restrict in, Alpha alpha,
+														  int64_t cols) {
+			float64x2_t r0, r1;
+
+			r0 = vld1q_f64(&in[0 * cols]);
+			r1 = vld1q_f64(&in[1 * cols]);
+
+			float64x2_t t0 = vzip1q_f64(r0, r1);
+			float64x2_t t1 = vzip2q_f64(r0, r1);
+
+			float64x2_t alphaVec = vdupq_n_f64(alpha);
+
+			vst1q_f64(&out[0 * cols], vmulq_f64(t0, alphaVec));
+			vst1q_f64(&out[1 * cols], vmulq_f64(t1, alphaVec));
+		}
+#	endif
+#endif // LIBRAPID_NATIVE_ARCH
 
 		// Ensure the kernel size is always defined, even if the above code doesn't define it
 #ifndef LIBRAPID_F32_TRANSPOSE_KERNEL_SIZE
@@ -310,7 +361,7 @@ namespace librapid {
 					}
 				}
 			}
-#endif	  // LIBRAPID_F64_TRANSPOSE_KERNEL_SIZE > 0
+#endif // LIBRAPID_F64_TRANSPOSE_KERNEL_SIZE > 0
 		} // namespace cpu
 
 #if defined(LIBRAPID_HAS_OPENCL)
@@ -431,8 +482,8 @@ namespace librapid {
 										   rows));
 			}
 		} // namespace cuda
-#endif	  // LIBRAPID_HAS_CUDA
-	}	  // namespace detail
+#endif // LIBRAPID_HAS_CUDA
+	} // namespace detail
 
 	namespace array {
 		template<typename TransposeType>

librapid/include/librapid/core/config.hpp

@@ -146,16 +146,17 @@
 #endif
 
 // Instruction sets
-#define ARCH_AVX512_2 10
-#define ARCH_AVX512	  9
-#define ARCH_AVX2	  8
-#define ARCH_AVX	  7
-#define ARCH_SSE4_2	  6
-#define ARCH_SSE4_1	  5
-#define ARCH_SSSE3	  4
-#define ARCH_SSE3	  3
-#define ARCH_SSE2	  2
-#define ARCH_SSE	  1
+#define ARCH_AVX512_2 11
+#define ARCH_AVX512	  10
+#define ARCH_AVX2	  9
+#define ARCH_AVX	  8
+#define ARCH_SSE4_2	  7
+#define ARCH_SSE4_1	  6
+#define ARCH_SSSE3	  5
+#define ARCH_SSE3	  4
+#define ARCH_SSE2	  3
+#define ARCH_SSE	  2
+#define ARCH_NEON	  1
 #define ARCH_NONE	  0
 
 // Instruction set detection
@@ -485,4 +486,4 @@ namespace librapid::backend {
 // Code to be run *before* main()
 #include "preMain.hpp"
 
-#endif // LIBRAPID_CORE_CONFIG_HPP
+#endif // LIBRAPID_CORE_CONFIG_HPP

librapid/include/librapid/core/log.hpp

@@ -20,7 +20,26 @@ namespace librapid::assert {
 												 (int)signature.length(),
 												 (int)strlen("ASSERTION FAILED"));
 
-				std::string formatted = fmt::format(
+				// std::string formatted = fmt::format(
+				//   "[{0:-^{6}}]\n[File {1:>{7}}]\n[Function "
+				//   "{2:>{8}}]\n[Line {3:>{9}}]\n[Condition "
+				//   "{4:>{10}}]\n{5}\n",
+				//   "ASSERTION FAILED",
+				//   filename,
+				//   signature,
+				//   line,
+				//   conditionString,
+				//   formattedMessage,
+				//   maxLen + 14,
+				//   maxLen + 9,
+				//   maxLen + 5,
+				//   maxLen + 9,
+				//   maxLen + 4);
+
+				// fmt::print(fmt::fg(fmt::color::red), formatted);
+				// fmt::vprint(fmt::fg(fmt::color::red), formatted);
+
+				fmt::print(fmt::fg(fmt::color::red),
 				  "[{0:-^{6}}]\n[File {1:>{7}}]\n[Function "
 				  "{2:>{8}}]\n[Line {3:>{9}}]\n[Condition "
 				  "{4:>{10}}]\n{5}\n",
@@ -35,8 +54,6 @@ namespace librapid::assert {
 				  maxLen + 5,
 				  maxLen + 9,
 				  maxLen + 4);
-
-				fmt::print(fmt::fg(fmt::color::red), formatted);
 			}
 
 			throw RaiseType(formattedMessage);

librapid/src/global.cpp

@@ -55,7 +55,10 @@ namespace librapid {
         // OpenBLAS threading
 #if defined(LIBRAPID_BLAS_OPENBLAS)
         openblas_set_num_threads((int)numThreads);
+
+    #if defined(_OPENMP)
         omp_set_num_threads((int)numThreads);
+    #endif // _OPENMP
         goto_set_num_threads((int)numThreads);
 
         setOpenBLASThreadsEnv((int)numThreads);

librapid/src/openclConfigure.cpp

@@ -29,15 +29,6 @@ __kernel void testAddition(__global const float *a, __global const float *b, __g
             cl::Program program(context, sources);
             err = program.build();
 
-            // if (err != CL_SUCCESS) {
-            // 	auto format = fmt::fg(fmt::color::red) | fmt::emphasis::bold;
-            // 	fmt::print(format,
-            // 			   "Error compiling test program: {}\n",
-            // 			   program.getBuildInfo<CL_PROGRAM_BUILD_LOG>(device));
-            // 	fmt::print(format, "Error Code [{}]: {}\n", err, opencl::getOpenCLErrorString(err));
-            // 	return false;
-            // }
-
             // Check the build status
             cl_build_status buildStatus = program.getBuildInfo<CL_PROGRAM_BUILD_STATUS>(device);