component-widening-thresholds: don't stumble on nans
View Changes

ir1opt: Converge types fed back into themselves in one place.
View Changes

constraint: Generalize threshold narrowing to other numeric types.
View Changes

ir1opt: Widen result types properly.
View Changes

Don't remove calls to sb-vm::%string-widetag-and-n-bits-shift
View Changes

Warn about (truncate n 0)
View Changes

Relax a restriction in fixup-code-object
View Changes

fold-call-derived-to-constant: handle multiple values
View Changes

ftruncate derive-type: ensure float results
View Changes

Use the same queue structure for immobile-space and fullcgc
View Changes

Derive that (/ n 0) is a float
View Changes

Better (/ n n) type
View Changes

Better (- n n) type
View Changes

CI: don't build x86 --without-sb-unicode
View Changes

Better types for (/ 0 x)
View Changes

Fix a test for -sb-unicode
View Changes

Better (* n 0) type derivation
View Changes

Better * type derivation around integers and 0
View Changes

Fix some issues in SB-UNICODE:SENTENCES
View Changes

Add a test case
View Changes

ir1opt: Forget the result type bound when the result becomes unused.
View Changes

perfecthash: Only compute the part of scramble[] that gets used.
View Changes

Remove workarounds for the perfect hash journal files.
View Changes

Port the perfect hash generator to Lisp and remove the journal files.
View Changes

ir1opt: Solve for the types of set variables optimistically as well.
View Changes

optimize-constant-loads: account for sc-element-size=2 for gc-barriers
View Changes

perfecthash: Only compute the part of scramble[] that gets used.
View Changes

Remove workarounds for the perfect hash journal files.
View Changes

Port the perfect hash generator to Lisp and remove the journal files.
View Changes

Fix optimize-constant-loads for s128
View Changes