T0b1/wasmtime
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Cranelift(Aarch64): Optimize lowering of icmps with immediates (#5252)

Browse Source 
We can encode more constants into 12-bit immediates if we do the following
rewrite for comparisons with odd constants:

        A >= B + 1
    ==> A - 1 >= B
    ==> A > B
This commit is contained in:
Nick Fitzgerald
2022-11-15 09:18:55 -08:00
committed by GitHub
parent 6dcdabf37e
commit 9967782726
7 changed files with 236 additions and 51 deletions
Download Patch File Download Diff File Expand all files Collapse all files

60
cranelift/codegen/src/isa/aarch64/lower.isle
View File

@@ -1695,14 +1695,22 @@
;;;; Rules for `select` ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
(rule (lower (has_type ty
(select _flags @ (icmp cc x @ (value_type in_ty) y) rn rm)))
(let ((cond Cond (cond_code cc)))
(lower_select
(lower_icmp_into_flags cc x y in_ty)
cond ty rn rm)))
(select (icmp cc
x @ (value_type in_ty)
y)
rn
rm)))
(let ((comparison FlagsAndCC (lower_icmp_into_flags cc x y in_ty)))
(lower_select (flags_and_cc_flags comparison)
(cond_code (flags_and_cc_cc comparison))
ty
rn
rm)))
(rule (lower (has_type ty
(select _flags @ (fcmp cc x @ (value_type in_ty) y) rn rm)))
(select (fcmp cc x @ (value_type in_ty) y)
rn
rm)))
(let ((cond Cond (fp_cond_code cc)))
(lower_select
(fpu_cmp (scalar_size in_ty) x y)
@@ -1729,12 +1737,16 @@
;;;; Rules for `select_spectre_guard` ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
(rule (lower (has_type ty
(select_spectre_guard
(icmp cc x @ (value_type in_ty) y) if_true if_false)))
(let ((cond Cond (cond_code cc))
(select_spectre_guard (icmp cc x @ (value_type in_ty) y)
if_true
if_false)))
(let ((comparison FlagsAndCC (lower_icmp_into_flags cc x y in_ty))
(dst ValueRegs (lower_select
(lower_icmp_into_flags cc x y in_ty)
cond ty if_true if_false))
(flags_and_cc_flags comparison)
(cond_code (flags_and_cc_cc comparison))
ty
if_true
if_false))
(_ InstOutput (side_effect (csdb))))
dst))
@@ -2381,23 +2393,27 @@
;; `brz` following `icmp`
(rule (lower_branch (brz (icmp cc x @ (value_type ty) y) _ _) targets)
(let ((cond Cond (cond_code cc))
(cond Cond (invert_cond cond)) ;; negate for `brz`
(let ((comparison FlagsAndCC (lower_icmp_into_flags cc x y ty))
;; Negate the condition for `brz`.
(cond Cond (invert_cond (cond_code (flags_and_cc_cc comparison))))
(taken BranchTarget (branch_target targets 0))
(not_taken BranchTarget (branch_target targets 1)))
(side_effect
(with_flags_side_effect (lower_icmp_into_flags cc x y ty)
(cond_br taken not_taken
(cond_br_cond cond))))))
(side_effect
(with_flags_side_effect (flags_and_cc_flags comparison)
(cond_br taken
not_taken
(cond_br_cond cond))))))
;; `brnz` following `icmp`
(rule (lower_branch (brnz (icmp cc x @ (value_type ty) y) _ _) targets)
(let ((cond Cond (cond_code cc))
(let ((comparison FlagsAndCC (lower_icmp_into_flags cc x y ty))
(cond Cond (cond_code (flags_and_cc_cc comparison)))
(taken BranchTarget (branch_target targets 0))
(not_taken BranchTarget (branch_target targets 1)))
(side_effect
(with_flags_side_effect (lower_icmp_into_flags cc x y ty)
(cond_br taken not_taken
(cond_br_cond cond))))))
(side_effect
(with_flags_side_effect (flags_and_cc_flags comparison)
(cond_br taken
not_taken
(cond_br_cond cond))))))
;; `brz` following `fcmp`
(rule (lower_branch (brz (fcmp cc x @ (value_type (ty_scalar_float ty)) y) _ _) targets)
(let ((cond Cond (fp_cond_code cc))