The new performance profiles give strange allocation figures:
Compiling Tachyon: 484.539 s, 0 MB allocated time=84% alloc=NaN% -- Machine code generation time=46% alloc=NaN% -- Linear scan time=20% alloc=NaN% -- Resolution time=10% alloc=NaN% -- IR to ASM translation time=2% alloc=NaN% -- Computing live intervals time=1% alloc=NaN% -- Assembly time=1% alloc=NaN% -- Operand assignment time=0% alloc=NaN% -- Number instructions time=0% alloc=NaN% -- Computing fixed intervals time=0% alloc=NaN% -- Order blocks time=13% alloc=NaN% -- IR lowering time=1% alloc=NaN% -- IR generation time=0% alloc=NaN% -- Machine code linking time=0% alloc=NaN% -- Parsing
There must be a bug in memAllocatedKB .
Moreover, shouldn't there be 2 versions of this procedure? One reporting the allocations on the V8 heap, and the other on the Tachyon heap?
Marc