Optimization Levels

-O optimization levels run from -1 to 6. Which options are enabled in these levels don’t seem to be documented in the User’s Guide or the mmc(1) man page. They are however documented in the compiler source code. -O2 is the default.

Note: Order matters! For example, --introduce-accumulators followed by -O6 will actually disable --introduce-accumulators, because it is disabled in -O6. For this reason you should put -O options first on the command line.

-O-1

Disables all optimizations which can be disabled.

• --no-common-data
• --no-llds-optimize
• --no-optimize-jumps
• --no-optimize-labels
• --no-optimize-peep
• --no-smart-indexing
• --no-static-ground-cells

-O0

Minimizes compilation time.

• --excess-assign
• --optimize-dead-procs
• --optimize-repeat 1
• --no-c-optimize
• --no-emit-c-loops
• --no-middle-rec
• --no-optimize-delay-slot
• --no-optimize-frames
• --no-optimize-tailcalls
• --no-use-local-vars

-O1

Cheap optimizations with good payoff.

• --no-common-struct
• --no-follow-code
• --no-inline-simple
• --no-inline-single-use
• --no-optimize-fulljumps
• --no-optimize-initializations
• --no-simple-neg

-O2

Optimizations with good payoff.

• --inline-compound-threshold 10
• --optimize-dups
• --optimize-repeat 1
• --user-guided-type-specialization

-O3

Slow optimizations with good payoff.

• --constraint-propagation
• --deforestation
• --local-constraint-propagation
• --optimize-higher-order
• --optimize-reassign
• --optimize-repeat 4
• --optimize-saved-vars
• --optimize-unused-args

-O4

Slow optimizations.

• --higher-order-size-limit 30
• --inline-compound-threshold 20
• --inline-simple-threshold 8

-O5

Very slow optimizations.

• --delay-constructs
• --eliminate-local-vars
• --higher-order-size-limit 40
• --inline-compound-threshold 100
• --loop-invariants
• --optimize-repeat 5

-O6

Unreasonably slow optimizations.

• --everything-in-one-c-function
• --inline-alloc
• --use-macro-for-redo-fail

--optimize-space

Optimize for code size rather than speed. This can be used in conjunction with one of the other levels, keeping in mind the aforementioned rules of ordering.

• --optimize-dead-procs
• --optimize-dups
• --optimize-proc-dups
• --optimize-reassign
• --unneeded-code-copy-limit 1
• --no-optimize-fulljumps

Additionally, --optimize-space turns off the following options if set:

• --inline-alloc
• --loop-invariants
• --use-macro-for-redo-fail
• --no-optimize-labels

Optimizations never enabled

These can make code run slower.

• --checked-nondet-tailcalls
• --constraint-propagation
• --introduce-accumulators (buggy with trailed updates)
• --optimize-constructor-last-call
• --optimize-duplicate-calls (buggy with array.init)
• --type-specialization
• --unneeded-code

Additionally, intermodule optimizations are never enabled by -O because they require special handling by mmc --make.