Float Sum combine1: Maximum use of data abstraction:
11.43 cycles/element
Float Sum combine2: Take vec_length() out of loop:
9.55 cycles/element
Float Sum combine3: Array reference to vector data:
2.86 cycles/element
Float Sum combine4: Array reference, accumulate in temporary:
2.86 cycles/element
Float Sum combine4p: Pointer reference, accumulate in temporary:
2.86 cycles/element
Float Sum Array code, unrolled by 2:
2.86 cycles/element
Float Sum combine5p: Pointer code, unrolled by 3, for loop:
2.87 cycles/element
Float Sum Array code, unrolled by 3, while loop:
2.86 cycles/element
Float Sum Array code, unrolled by 4:
2.87 cycles/element
Float Sum Array code, unrolled by 8:
2.88 cycles/element
Float Sum Array code, unrolled by 16:
2.86 cycles/element
Float Sum Pointer code, unrolled by 2:
2.86 cycles/element
Float Sum Pointer code, unrolled by 3:
2.86 cycles/element
Float Sum Pointer code, unrolled by 4:
2.88 cycles/element
Float Sum Pointer code, unrolled by 8:
2.87 cycles/element
Float Sum Pointer code, unrolled by 16:
2.87 cycles/element
Float Sum combine6: Array code, unrolled by 2, Superscalar x2:
1.44 cycles/element
Float Sum Array code, unrolled by 4, Superscalar x2:
1.42 cycles/element
Float Sum Array code, unrolled by 8, Superscalar x2:
1.44 cycles/element
Float Sum Array code, unrolled by 3, Superscalar x3:
0.96 cycles/element
Float Sum Array code, unrolled by 4, Superscalar x4:
0.96 cycles/element
Float Sum Array code, unrolled by 8, Superscalar x4:
0.97 cycles/element
Float Sum Array code, unrolled by 6, Superscalar x6:
0.98 cycles/element
Float Sum Array code, unrolled by 8, Superscalar x8:
0.98 cycles/element
Float Sum Array code, unrolled by 10, Superscalar x10:
0.96 cycles/element
Float Sum Array code, unrolled by 12, Superscalar x6:
0.96 cycles/element
Float Sum Array code, unrolled by 12, Superscalar x12:
0.96 cycles/element
Float Sum Pointer code, unrolled by 8, Superscalar x2:
1.43 cycles/element
Float Sum Pointer code, unrolled by 8, Superscalar x4:
0.97 cycles/element
Float Sum Pointer code, unrolled by 8, Superscalar x8:
0.98 cycles/element
Float Sum Pointer code, unrolled by 9, Superscalar x3:
0.98 cycles/element
Float Sum Array code, Unroll x2, Superscalar x2, noninterleaved:
1.43 cycles/element
Float Sum Array code, unrolled by 2, different associativity:
1.43 cycles/element
Float Sum Array code, unrolled by 3, Different Associativity:
1.27 cycles/element
Float Sum Array code, unrolled by 4, Different Associativity:
0.95 cycles/element
Float Sum Array code, unrolled by 6, Different Associativity:
0.97 cycles/element
Float Sum Array code, unrolled by 8, Different Associativity:
0.98 cycles/element
Float Sum SSE code, 1*VSIZE-way parallelism:
0.72 cycles/element
Float Sum SSE code, 2*VSIZE-way parallelism:
0.36 cycles/element
Float Sum SSE code, 4*VSIZE-way parallelism:
0.23 cycles/element
Float Sum SSE code, 8*VSIZE-way parallelism:
0.24 cycles/element
Float Sum SSE code, 12*VSIZE-way parallelism:
0.23 cycles/element
Float Sum SSE code, 2*VSIZE-way parallelism, reassociate:
0.41 cycles/element
Float Sum SSE code, 4*VSIZE-way parallelism, reassociate:
0.23 cycles/element
Float Sum SSE code, 8*VSIZE-way parallelism, reassociate:
0.24 cycles/element