TRMM Optimization

This repository implements various parallel optimization techniques for TRMM. The starting point for all variants is baseline_op.c.

Optimization Techniques

• noifstatementvarXXX.c: Removes the if statement and reorders the loop
• blocked_JIP_IJ_X.c: Applies loop blocking for the I and J loops using different block sizes
• blocked_JIP_IP_X.c: Applies loop blocking for the I and P loops using different block sizes
• blocked_JIP_PJ_X.c: Applies loop blocking for the P and J loops using different block sizes
• blocked_JIP_JIP.c: Applies loop blocking for all loops
• mutex_critical_section.c: OpenMP 2 threads using omp critical
• mutex_lock.c: OpenMP 2 threads using omp_lock_t
• mutex_reduction.c: OpenMP 8 threads using reduction(+ : res)
• openMP_X.c: OpenMP using various thread sizes
• SIMD_X.c: Uses AVX2 to operate on eight elements at once if not along diagonal
• openMP_SIMD.c: Combines OpenMP and SIMD
• tuned_variantXX_op.cu: Applies the CUDA parallelization model to distribute work across GPU threads

File Descriptions

• baseline_op.c: The starting point for all variants.
• writeup.pdf: Contains a full description and results of the implemented optimization techniques.