|
From: Felix S. <sch...@go...> - 2009-11-08 16:33:56
|
Dear Valgrind Developers, I'm using cachegrind at the moment. My valgrind version is Valgrind-3.6.0.SVN And I will receive strange results. Well I think calculating miss rates aren't right. Calculation the miss rates by hand will show a higher percentage, e.g. L2 miss rate would be 7.8% and not 0.0% Is there a missunderstanding by me? Could you help me ==3293== ==3293== I refs: 55,537,351 ==3293== I1 misses: 3,739 ==3293== L2i misses: 2,728 ==3293== I1 miss rate: 0.00% ==3293== L2i miss rate: 0.00% ==3293== ==3293== D refs: 31,150,436 (29,850,784 rd + 1,299,652 wr) ==3293== D1 misses: 172,610 ( 166,369 rd + 6,241 wr) ==3293== L2d misses: 11,092 ( 5,513 rd + 5,579 wr) ==3293== D1 miss rate: 0.5% ( 0.5% + 0.4% ) ==3293== L2d miss rate: 0.0% ( 0.0% + 0.4% ) ==3293== ==3293== L2 refs: 176,349 ( 170,108 rd + 6,241 wr) ==3293== L2 misses: 13,820 ( 8,241 rd + 5,579 wr) ==3293== L2 miss rate: 0.0% ( 0.0% + 0.4% ) |
|
From: Julian S. <js...@ac...> - 2009-11-08 17:20:33
|
On Sunday 08 November 2009, Felix Schmidt wrote: > Dear Valgrind Developers, > > I'm using cachegrind at the moment. My valgrind version is > Valgrind-3.6.0.SVN > > And I will receive strange results. Well I think calculating miss rates > aren't right. > > Calculation the miss rates by hand will show a higher percentage, e.g. L2 > miss rate would be 7.8% and not 0.0% > > Is there a missunderstanding by me? Yes. It is true that 7.8% of references presented to the L2 cache miss (7.8% == 13820 / 176349). The L2i miss rate tells you the proportion of instruction reads that missed L2, == 2728 / 55537351 == 0.0% as it says. Similarly for the L2d miss rate. J > Could you help me > > ==3293== > ==3293== I refs: 55,537,351 > ==3293== I1 misses: 3,739 > ==3293== L2i misses: 2,728 > ==3293== I1 miss rate: 0.00% > ==3293== L2i miss rate: 0.00% > ==3293== > ==3293== D refs: 31,150,436 (29,850,784 rd + 1,299,652 wr) > ==3293== D1 misses: 172,610 ( 166,369 rd + 6,241 wr) > ==3293== L2d misses: 11,092 ( 5,513 rd + 5,579 wr) > ==3293== D1 miss rate: 0.5% ( 0.5% + 0.4% ) > ==3293== L2d miss rate: 0.0% ( 0.0% + 0.4% ) > ==3293== > ==3293== L2 refs: 176,349 ( 170,108 rd + 6,241 wr) > ==3293== L2 misses: 13,820 ( 8,241 rd + 5,579 wr) > ==3293== L2 miss rate: 0.0% ( 0.0% + 0.4% ) |
|
From: Nicholas N. <n.n...@gm...> - 2009-11-09 01:49:30
|
On Mon, Nov 9, 2009 at 4:34 AM, Julian Seward <js...@ac...> wrote: >> >> Is there a missunderstanding by me? > > Yes. It is true that 7.8% of references presented to the L2 cache miss > (7.8% == 13820 / 176349). > > The L2i miss rate tells you the proportion of instruction > reads that missed L2, == 2728 / 55537351 == 0.0% as it says. > > Similarly for the L2d miss rate. And the docs (http://www.valgrind.org/docs/manual/cg-manual.html, section 5.2.1) make this quite clear: "Note that the L2 miss rate is computed relative to the total number of memory accesses, not the number of L1 misses. I.e. it is (I2mr + D2mr + D2mw) / (Ir + Dr + Dw) not (I2mr + D2mr + D2mw) / (I1mr + D1mr + D1mw)" Nick |