Other pages

• Step-by-step guide to optimization
  • Benchmark, detailed results
  • Benchmark listing
  • Software Pipelining Diagnostics
• Discussion of options
• Discussion of 64-bit issues

Software Pipelining Diagnostics

corona:mflop/mflop6> grep swps *.s

 #
 # Pipelined loop line 71 steady state
 #
 #     2 unrollings before pipelining
 #     6 cycles per 2 iterations
 #    10 flops        ( 41% of peak) (madds count as 2)
 #     6 flops        ( 50% of peak) (madds count as 1)
 #     4 madds        ( 33% of peak)
 #     8 mem refs     ( 66% of peak)
 #     6 integer ops  ( 50% of peak)
 #    20 instructions ( 83% of peak)
 #     3 short trip threshold
 #
 #
 # Pipelined loop line 78 steady state
 #
 #     2 unrollings before pipelining
 #     4 cycles per 2 iterations
 #     0 flops        (  0% of peak) (madds count as 2)
 #     0 flops        (  0% of peak) (madds count as 1)
 #     0 madds        (  0% of peak)
 #     8 mem refs     (100% of peak)
 #     2 integer ops  ( 25% of peak)
 #    10 instructions ( 62% of peak)
 #     0 short trip threshold
 #
 #
 # Pipelined loop line 78 steady state
 #
 #     2 unrollings before pipelining
 #     4 cycles per 2 iterations
 #     0 flops        (  0% of peak) (madds count as 2)
 #     0 flops        (  0% of peak) (madds count as 1)
 #     0 madds        (  0% of peak)
 #     8 mem refs     (100% of peak)
 #     2 integer ops  ( 25% of peak)
 #    10 instructions ( 62% of peak)
 #     2 short trip threshold
 #
 #
 # Pipelined loop line 89 steady state
 #
 #       Not unrolled before pipelining
 #     2 cycles per iteration
 #     0 flops        (  0% of peak) (madds count as 2)
 #     0 flops        (  0% of peak) (madds count as 1)
 #     0 madds        (  0% of peak)
 #     2 mem refs     ( 50% of peak)
 #     2 integer ops  ( 50% of peak)
 #     4 instructions ( 50% of peak)
 #     1 short trip threshold
 #
 #
 # Pipelined loop line 89 steady state
 #
 #     4 unrollings before pipelining
 #     4 cycles per 4 iterations
 #     0 flops        (  0% of peak) (madds count as 2)
 #     0 flops        (  0% of peak) (madds count as 1)
 #     0 madds        (  0% of peak)
 #     8 mem refs     (100% of peak)
 #     2 integer ops  ( 25% of peak)
 #    10 instructions ( 62% of peak)
 #     1 short trip threshold
 #
 #
 # Pipelined loop line 101 steady state
 #
 #       Not unrolled before pipelining
 #     2 cycles per iteration
 #     4 flops        ( 50% of peak) (madds count as 2)
 #     4 flops        (100% of peak) (madds count as 1)
 #     0 madds        (  0% of peak)
 #     2 mem refs     ( 50% of peak)
 #     2 integer ops  ( 50% of peak)
 #     8 instructions (100% of peak)
 #    10 short trip threshold
 #
 #
 # Pipelined loop line 101 steady state
 #
 #     4 unrollings before pipelining
 #     4 cycles per 4 iterations
 #     4 flops        ( 25% of peak) (madds count as 2)
 #     4 flops        ( 50% of peak) (madds count as 1)
 #     0 madds        (  0% of peak)
 #     8 mem refs     (100% of peak)
 #     2 integer ops  ( 25% of peak)
 #    14 instructions ( 87% of peak)
 #     1 short trip threshold
 #
 #
 # Pipelined loop line 112 steady state
 #
 #       Not unrolled before pipelining
 #     3 cycles per iteration
 #     4 flops        ( 33% of peak) (madds count as 2)
 #     4 flops        ( 66% of peak) (madds count as 1)
 #     0 madds        (  0% of peak)
 #     3 mem refs     ( 50% of peak)
 #     2 integer ops  ( 33% of peak)
 #     9 instructions ( 75% of peak)
 #     6 short trip threshold
 #
 #
 # Pipelined loop line 112 steady state
 #
 #     4 unrollings before pipelining
 #     6 cycles per 4 iterations
 #     4 flops        ( 16% of peak) (madds count as 2)
 #     4 flops        ( 33% of peak) (madds count as 1)
 #     0 madds        (  0% of peak)
 #    12 mem refs     (100% of peak)
 #     2 integer ops  ( 16% of peak)
 #    18 instructions ( 75% of peak)
 #     1 short trip threshold
 #
 #
 # Pipelined loop line 123 steady state
 #
 #       Not unrolled before pipelining
 #     2 cycles per iteration
 #     1 flop         ( 12% of peak) (madds count as 2)
 #     1 flop         ( 25% of peak) (madds count as 1)
 #     0 madds        (  0% of peak)
 #     3 mem refs     ( 75% of peak)
 #     2 integer ops  ( 50% of peak)
 #     6 instructions ( 75% of peak)
 #     6 short trip threshold
 #
 #
 # Pipelined loop line 123 steady state
 #
 #     2 unrollings before pipelining
 #     4 cycles per 2 iterations
 #     2 flops        ( 12% of peak) (madds count as 2)
 #     2 flops        ( 25% of peak) (madds count as 1)
 #     0 madds        (  0% of peak)
 #     6 mem refs     ( 75% of peak)
 #     2 integer ops  ( 25% of peak)
 #    10 instructions ( 62% of peak)
 #     3 short trip threshold
 #
 #   Software pipelining problems??!!
 #    3 cycles lower bound due to resources
 #     Pipeliner had problems finding a schedule:
 #       (Scheduled in 4 cycles vs lower bound of 3)
 #
 #
 # Pipelined loop line 134 steady state
 #
 #       Not unrolled before pipelining
 #     3 cycles per iteration
 #     8 flops        ( 66% of peak) (madds count as 2)
 #     4 flops        ( 66% of peak) (madds count as 1)
 #     4 madds        ( 66% of peak)
 #     3 mem refs     ( 50% of peak)
 #     2 integer ops  ( 33% of peak)
 #     9 instructions ( 75% of peak)
 #     6 short trip threshold
 #
 #
 # Pipelined loop line 134 steady state
 #
 #     2 unrollings before pipelining
 #     4 cycles per 2 iterations
 #     4 flops        ( 25% of peak) (madds count as 2)
 #     2 flops        ( 25% of peak) (madds count as 1)
 #     2 madds        ( 25% of peak)
 #     6 mem refs     ( 75% of peak)
 #     2 integer ops  ( 25% of peak)
 #    10 instructions ( 62% of peak)
 #     3 short trip threshold
 #
 #   Software pipelining problems??!!
 #    3 cycles lower bound due to resources
 #     Pipeliner had problems finding a schedule:
 #       (Scheduled in 4 cycles vs lower bound of 3)
 #
 #
 # Pipelined loop line 148 steady state
 #
 #       Not unrolled before pipelining
 #    12 cycles per iteration
 #    16 flops        ( 33% of peak) (madds count as 2)
 #     8 flops        ( 33% of peak) (madds count as 1)
 #     8 madds        ( 33% of peak)
 #    24 mem refs     (100% of peak)
 #     2 integer ops  (  8% of peak)
 #    34 instructions ( 70% of peak)
 #     6 short trip threshold
 #
 #
 # Pipelined loop line 148 steady state
 #
 #       Not unrolled before pipelining
 #     3 cycles per iteration
 #     4 flops        ( 33% of peak) (madds count as 2)
 #     2 flops        ( 33% of peak) (madds count as 1)
 #     2 madds        ( 33% of peak)
 #     6 mem refs     (100% of peak)
 #     2 integer ops  ( 33% of peak)
 #    10 instructions ( 83% of peak)
 #     7 short trip threshold
 #
 #
 # Pipelined loop line 159 steady state
 #
 #       Not unrolled before pipelining
 #     3 cycles per iteration
 #     8 flops        ( 66% of peak) (madds count as 2)
 #     4 flops        ( 66% of peak) (madds count as 1)
 #     4 madds        ( 66% of peak)
 #     4 mem refs     ( 66% of peak)
 #     2 integer ops  ( 33% of peak)
 #    10 instructions ( 83% of peak)
 #     6 short trip threshold
 #
 #
 # Pipelined loop line 159 steady state
 #
 #     2 unrollings before pipelining
 #     4 cycles per 2 iterations
 #     4 flops        ( 25% of peak) (madds count as 2)
 #     2 flops        ( 25% of peak) (madds count as 1)
 #     2 madds        ( 25% of peak)
 #     8 mem refs     (100% of peak)
 #     2 integer ops  ( 25% of peak)
 #    12 instructions ( 75% of peak)
 #     3 short trip threshold
 #
 #
 # Pipelined loop line 172 steady state
 #
 #       Not unrolled before pipelining
 #     8 cycles per iteration
 #    16 flops        ( 50% of peak) (madds count as 2)
 #     8 flops        ( 50% of peak) (madds count as 1)
 #     8 madds        ( 50% of peak)
 #    16 mem refs     (100% of peak)
 #     2 integer ops  ( 12% of peak)
 #    26 instructions ( 81% of peak)
 #     2 short trip threshold
 #
 #
 # Pipelined loop line 172 steady state
 #
 #     2 unrollings before pipelining
 #     4 cycles per 2 iterations
 #     8 flops        ( 50% of peak) (madds count as 2)
 #     4 flops        ( 50% of peak) (madds count as 1)
 #     4 madds        ( 50% of peak)
 #     8 mem refs     (100% of peak)
 #     2 integer ops  ( 25% of peak)
 #    14 instructions ( 87% of peak)
 #     1 short trip threshold
 #
 #
 # Pipelined loop line 183 steady state
 #
 #       Not unrolled before pipelining
 #     7 cycles per iteration
 #    13 flops        ( 46% of peak) (madds count as 2)
 #     9 flops        ( 64% of peak) (madds count as 1)
 #     4 madds        ( 28% of peak)
 #     5 mem refs     ( 35% of peak)
 #     2 integer ops  ( 14% of peak)
 #    16 instructions ( 57% of peak)
 #     8 short trip threshold
 #
 #
 # Pipelined loop line 183 steady state
 #
 #     2 unrollings before pipelining
 #     7 cycles per 2 iterations
 #     8 flops        ( 28% of peak) (madds count as 2)
 #     6 flops        ( 42% of peak) (madds count as 1)
 #     2 madds        ( 14% of peak)
 #    10 mem refs     ( 71% of peak)
 #     2 integer ops  ( 14% of peak)
 #    18 instructions ( 64% of peak)
 #     4 short trip threshold
 #
 #
 # Pipelined loop line 194 steady state
 #
 #       Not unrolled before pipelining
 #    16 cycles per iteration
 #    24 flops        ( 37% of peak) (madds count as 2)
 #    16 flops        ( 50% of peak) (madds count as 1)
 #     8 madds        ( 25% of peak)
 #     4 mem refs     ( 12% of peak)
 #     2 integer ops  (  6% of peak)
 #    22 instructions ( 34% of peak)
 #     7 short trip threshold
 #
 #
 # Pipelined loop line 194 steady state
 #
 #       Not unrolled before pipelining
 #     4 cycles per iteration
 #     6 flops        ( 37% of peak) (madds count as 2)
 #     4 flops        ( 50% of peak) (madds count as 1)
 #     2 madds        ( 25% of peak)
 #     4 mem refs     ( 50% of peak)
 #     2 integer ops  ( 25% of peak)
 #    10 instructions ( 62% of peak)
 #     6 short trip threshold
 #
 #
 # Pipelined loop line 207 steady state
 #
 #       Not unrolled before pipelining
 #     2 cycles per iteration
 #     5 flops        ( 62% of peak) (madds count as 2)
 #     3 flops        ( 75% of peak) (madds count as 1)
 #     2 madds        ( 50% of peak)
 #     2 mem refs     ( 50% of peak)
 #     2 integer ops  ( 50% of peak)
 #     7 instructions ( 87% of peak)
 #     7 short trip threshold
 #
 #
 # Pipelined loop line 207 steady state
 #
 #     2 unrollings before pipelining
 #     3 cycles per 2 iterations
 #    10 flops        ( 83% of peak) (madds count as 2)
 #     6 flops        (100% of peak) (madds count as 1)
 #     4 madds        ( 66% of peak)
 #     4 mem refs     ( 66% of peak)
 #     2 integer ops  ( 33% of peak)
 #    12 instructions (100% of peak)
 #     6 short trip threshold
 #
 #
 # Pipelined loop line 218 steady state
 #
 #       Not unrolled before pipelining
 #     2 cycles per iteration
 #     5 flops        ( 62% of peak) (madds count as 2)
 #     3 flops        ( 75% of peak) (madds count as 1)
 #     2 madds        ( 50% of peak)
 #     2 mem refs     ( 50% of peak)
 #     2 integer ops  ( 50% of peak)
 #     7 instructions ( 87% of peak)
 #     7 short trip threshold
 #
 #
 # Pipelined loop line 218 steady state
 #
 #     2 unrollings before pipelining
 #     3 cycles per 2 iterations
 #    10 flops        ( 83% of peak) (madds count as 2)
 #     6 flops        (100% of peak) (madds count as 1)
 #     4 madds        ( 66% of peak)
 #     4 mem refs     ( 66% of peak)
 #     2 integer ops  ( 33% of peak)
 #    12 instructions (100% of peak)
 #     5 short trip threshold
 #
 #
 # Pipelined loop line 229 steady state
 #
 #       Not unrolled before pipelining
 #     4 cycles per iteration
 #     5 flops        ( 31% of peak) (madds count as 2)
 #     4 flops        ( 50% of peak) (madds count as 1)
 #     1 madd         ( 12% of peak)
 #     2 mem refs     ( 25% of peak)
 #     3 integer ops  ( 37% of peak)
 #     9 instructions ( 56% of peak)
 #     4 short trip threshold
 #
 #
 # Pipelined loop line 229 steady state
 #
 #     4 unrollings before pipelining
 #    14 cycles per 4 iterations
 #    20 flops        ( 35% of peak) (madds count as 2)
 #    16 flops        ( 57% of peak) (madds count as 1)
 #     4 madds        ( 14% of peak)
 #     8 mem refs     ( 28% of peak)
 #     6 integer ops  ( 21% of peak)
 #    30 instructions ( 53% of peak)
 #     3 short trip threshold
 #
 #
 # Pipelined loop line 240 steady state
 #
 #       Not unrolled before pipelining
 #     7 cycles per iteration
 #     8 flops        ( 28% of peak) (madds count as 2)
 #     6 flops        ( 42% of peak) (madds count as 1)
 #     2 madds        ( 14% of peak)
 #     5 mem refs     ( 35% of peak)
 #    10 integer ops  ( 71% of peak)
 #    21 instructions ( 75% of peak)
 #     3 short trip threshold
 #
 #
 # Pipelined loop line 240 steady state
 #
 #     2 unrollings before pipelining
 #     7 cycles per 2 iterations
 #     8 flops        ( 28% of peak) (madds count as 2)
 #     6 flops        ( 42% of peak) (madds count as 1)
 #     2 madds        ( 14% of peak)
 #     6 mem refs     ( 42% of peak)
 #    10 integer ops  ( 71% of peak)
 #    22 instructions ( 78% of peak)
 #     3 short trip threshold
 #
 #
 # Pipelined loop line 253 steady state
 #
 #     4 unrollings before pipelining
 #    14 cycles per 4 iterations
 #    12 flops        ( 21% of peak) (madds count as 2)
 #    12 flops        ( 42% of peak) (madds count as 1)
 #     0 madds        (  0% of peak)
 #     8 mem refs     ( 28% of peak)
 #     2 integer ops  (  7% of peak)
 #    22 instructions ( 87% of peak)
 #     1 short trip threshold
 #