https://gcc.gnu.org/g:59d1bcfe4713965941e6eec13b540c17c7ff1a14
commit 59d1bcfe4713965941e6eec13b540c17c7ff1a14
Author: Michael Meissner <meiss...@linux.ibm.com>
Date: Wed Jul 9 01:54:45 2025 -0400
PR target/117251: Improve vector or to vector nand fusion
See the following post for a complete explanation of what the patches
for PR target/117251:
* https://gcc.gnu.org/pipermail/gcc-patches/2025-June/686474.html
This is patch #41 of 45 to generate the 'XXEVAL' instruction on power10
and power11 instead of using the Altivec 'VOR' instruction feeding into
'VNAND'. The 'XXEVAL' instruction can use all 64 vector registers,
instead of the 32 registers that traditional Altivec vector
instructions use. By allowing all of the vector registers to be used,
it reduces the amount of spilling that a large benchmark generated.
Currently the following code:
vector int a, b, c, d;
a = ~ ((c | d) & b);
Generates:
vor t,c,d
vnand a,t,b
Now in addition with this patch, if the arguments or result is
allocated to a traditional FPR register, the GCC compiler will now
generate the following code instead of adding vector move instructions:
xxeval a,b,c,248
Since fusion using 2 Altivec instructions is slightly faster than using
the 'XXEVAL' instruction we prefer to generate the Altivec instructions
if we can. In addition, because 'XXEVAL' is a prefixed instruction, it
possibly might generate an extra NOP instruction to align the 'XXEVAL'
instruction.
I have tested these patches on both big endian and little endian
PowerPC servers, with no regressions. Can I check these patchs into
the trunk?
2025-07-09 Michael Meissner <meiss...@linux.ibm.com>
gcc/
PR target/117251
* config/rs6000/fusion.md: Regenerate.
* config/rs6000/genfusion.pl (gen_logical_addsubf): Add support
to generate vector or => nand fusion if XXEVAL is supported.
Diff:
---
gcc/config/rs6000/fusion.md | 15 +++++++++------
gcc/config/rs6000/genfusion.pl | 1 +
2 files changed, 10 insertions(+), 6 deletions(-)
diff --git a/gcc/config/rs6000/fusion.md b/gcc/config/rs6000/fusion.md
index 01b7fda17ecc..39b586918c17 100644
--- a/gcc/config/rs6000/fusion.md
+++ b/gcc/config/rs6000/fusion.md
@@ -2435,20 +2435,23 @@
;; logical-logical fusion pattern generated by gen_logical_addsubf
;; vector vor -> vnand
(define_insn "*fuse_vor_vnand"
- [(set (match_operand:VM 3 "altivec_register_operand" "=&0,&1,&v,v")
- (ior:VM (not:VM (ior:VM (match_operand:VM 0 "altivec_register_operand"
"v,v,v,v")
- (match_operand:VM 1 "altivec_register_operand"
"v,v,v,v")))
- (not:VM (match_operand:VM 2 "altivec_register_operand"
"v,v,v,v"))))
- (clobber (match_scratch:VM 4 "=X,X,X,&v"))]
+ [(set (match_operand:VM 3 "vector_fusion_operand" "=&0,&1,&v,wa,v")
+ (ior:VM (not:VM (ior:VM (match_operand:VM 0 "vector_fusion_operand"
"v,v,v,wa,v")
+ (match_operand:VM 1 "vector_fusion_operand"
"v,v,v,wa,v")))
+ (not:VM (match_operand:VM 2 "vector_fusion_operand"
"v,v,v,wa,v"))))
+ (clobber (match_scratch:VM 4 "=X,X,X,X,&v"))]
"(TARGET_P10_FUSION)"
"@
vor %3,%1,%0\;vnand %3,%3,%2
vor %3,%1,%0\;vnand %3,%3,%2
vor %3,%1,%0\;vnand %3,%3,%2
+ xxeval %x3,%x2,%x1,%x0,248
vor %4,%1,%0\;vnand %3,%4,%2"
[(set_attr "type" "fused_vector")
(set_attr "cost" "6")
- (set_attr "length" "8")])
+ (set_attr "length" "8")
+ (set_attr "prefixed" "*,*,*,yes,*")
+ (set_attr "isa" "*,*,*,xxeval,*")])
;; logical-logical fusion pattern generated by gen_logical_addsubf
;; vector vorc -> vnand
diff --git a/gcc/config/rs6000/genfusion.pl b/gcc/config/rs6000/genfusion.pl
index d4965b6df864..86bca81286ca 100755
--- a/gcc/config/rs6000/genfusion.pl
+++ b/gcc/config/rs6000/genfusion.pl
@@ -255,6 +255,7 @@ sub gen_logical_addsubf
"vorc_vnand" => 244,
"veqv_vnand" => 246,
"vnor_vnand" => 247,
+ "vor_vnand" => 248,
);
KIND: foreach $kind ('scalar','vector') {
