I found this patch from 2003 by Tamas Gergely that was never merged.
Though it seems to still be valid, since when applying it, the CSiBE benchmark
reduce total code size -170 bytes, making 35 out of 893 files smaller (from -4
to -12 bytes (1-3 instructions).
Bug
asd
asd
asd
updated patch:
diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index 20bdc2bec37..4c924499ad9 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,9 @@
+2019-06-01 Fredrik Hederstierna <fred...@hederstierna.com>
+ Tamas Gergely <ger...@rgai.hu>
+
+ * gcc/config/arm/arm.md: new peephole2 pattern added for extending
+ move and compare parallelization. Retake from 2003 patch adapted.
+
2019-06-01 Martin Sebor <mse...@redhat.com>
PR middle-end/90694
diff --git a/gcc/config/arm/arm.md b/gcc/config/arm/arm.md
index ae582172ab9..753d18fbfc5 100644
--- a/gcc/config/arm/arm.md
+++ b/gcc/config/arm/arm.md
@@ -10951,6 +10951,19 @@
""
)
+; The same as the previous, but with exchanged operands
+;
+(define_peephole2
+ [(set (match_operand:SI 0 "arm_general_register_operand" "")
+ (match_operand:SI 1 "arm_general_register_operand" ""))
+ (set (reg:CC CC_REGNUM)
+ (compare:CC (match_dup 0) (const_int 0)))]
+ "TARGET_ARM"
+ [(parallel [(set (reg:CC CC_REGNUM) (compare:CC (match_dup 1) (const_int 0)))
+ (set (match_dup 0) (match_dup 1))])]
+ ""
+)
+
(define_split
[(set (match_operand:SI 0 "s_register_operand" "")
(and:SI (ge:SI (match_operand:SI 1 "s_register_operand" "")
