On 14/07/2025 10:35, Spencer Abson wrote:
On Mon, Jul 07, 2025 at 08:46:15AM +0000, Alfie Richards wrote:
Hello all,
This patch implements the couple of amin/amax instructions that are part of
SME2 + faminmax.
Regression testsed and bootstrapped for Aarch64.
Thanks,
Alfie
-- >8 --
Implements the sme2+faminmax svamin and svamax intrinsics.
gcc/ChangeLog:
* config/aarch64/aarch64-sme.md (@aarch64_sme_<faminmax_uns_op><mode>):
New patterns.
* config/aarch64/aarch64-sve-builtins-sme.def (svamin): New intrinsics.
(svamax): New intrinsics.
* config/aarch64/aarch64-sve-builtins-sve2.cc (class faminmaximpl): New
class.
(svamin): New function.
(svamax): New function.
gcc/testsuite/ChangeLog:
* gcc.target/aarch64/sme2/acle-asm/amax_f16_x2.c: New test.
* gcc.target/aarch64/sme2/acle-asm/amax_f16_x4.c: New test.
* gcc.target/aarch64/sme2/acle-asm/amax_f32_x2.c: New test.
* gcc.target/aarch64/sme2/acle-asm/amax_f32_x4.c: New test.
* gcc.target/aarch64/sme2/acle-asm/amax_f64_x2.c: New test.
* gcc.target/aarch64/sme2/acle-asm/amax_f64_x4.c: New test.
* gcc.target/aarch64/sme2/acle-asm/amin_f16_x2.c: New test.
* gcc.target/aarch64/sme2/acle-asm/amin_f16_x4.c: New test.
* gcc.target/aarch64/sme2/acle-asm/amin_f32_x2.c: New test.
* gcc.target/aarch64/sme2/acle-asm/amin_f32_x4.c: New test.
* gcc.target/aarch64/sme2/acle-asm/amin_f64_x2.c: New test.
* gcc.target/aarch64/sme2/acle-asm/amin_f64_x4.c: New test.
---
gcc/config/aarch64/aarch64-sme.md | 18 +++
.../aarch64/aarch64-sve-builtins-sme.def | 5 +
.../aarch64/aarch64-sve-builtins-sve2.cc | 44 +++++-
.../aarch64/sme2/acle-asm/amax_f16_x2.c | 97 +++++++++++++
.../aarch64/sme2/acle-asm/amax_f16_x4.c | 128 +++++++++++++++++
.../aarch64/sme2/acle-asm/amax_f32_x2.c | 96 +++++++++++++
.../aarch64/sme2/acle-asm/amax_f32_x4.c | 129 ++++++++++++++++++
.../aarch64/sme2/acle-asm/amax_f64_x2.c | 96 +++++++++++++
.../aarch64/sme2/acle-asm/amax_f64_x4.c | 128 +++++++++++++++++
.../aarch64/sme2/acle-asm/amin_f16_x2.c | 96 +++++++++++++
.../aarch64/sme2/acle-asm/amin_f16_x4.c | 128 +++++++++++++++++
.../aarch64/sme2/acle-asm/amin_f32_x2.c | 96 +++++++++++++
.../aarch64/sme2/acle-asm/amin_f32_x4.c | 128 +++++++++++++++++
.../aarch64/sme2/acle-asm/amin_f64_x2.c | 96 +++++++++++++
.../aarch64/sme2/acle-asm/amin_f64_x4.c | 128 +++++++++++++++++
15 files changed, 1409 insertions(+), 4 deletions(-)
create mode 100644
gcc/testsuite/gcc.target/aarch64/sme2/acle-asm/amax_f16_x2.c
create mode 100644
gcc/testsuite/gcc.target/aarch64/sme2/acle-asm/amax_f16_x4.c
create mode 100644
gcc/testsuite/gcc.target/aarch64/sme2/acle-asm/amax_f32_x2.c
create mode 100644
gcc/testsuite/gcc.target/aarch64/sme2/acle-asm/amax_f32_x4.c
create mode 100644
gcc/testsuite/gcc.target/aarch64/sme2/acle-asm/amax_f64_x2.c
create mode 100644
gcc/testsuite/gcc.target/aarch64/sme2/acle-asm/amax_f64_x4.c
create mode 100644
gcc/testsuite/gcc.target/aarch64/sme2/acle-asm/amin_f16_x2.c
create mode 100644
gcc/testsuite/gcc.target/aarch64/sme2/acle-asm/amin_f16_x4.c
create mode 100644
gcc/testsuite/gcc.target/aarch64/sme2/acle-asm/amin_f32_x2.c
create mode 100644
gcc/testsuite/gcc.target/aarch64/sme2/acle-asm/amin_f32_x4.c
create mode 100644
gcc/testsuite/gcc.target/aarch64/sme2/acle-asm/amin_f64_x2.c
create mode 100644
gcc/testsuite/gcc.target/aarch64/sme2/acle-asm/amin_f64_x4.c
diff --git a/gcc/config/aarch64/aarch64-sme.md
b/gcc/config/aarch64/aarch64-sme.md
index b8bb4cc14b6..bfe368e80b5 100644
--- a/gcc/config/aarch64/aarch64-sme.md
+++ b/gcc/config/aarch64/aarch64-sme.md
@@ -38,6 +38,7 @@
;; ---- Binary arithmetic on ZA tile
;; ---- Binary arithmetic on ZA slice
;; ---- Binary arithmetic, writing to ZA slice
+;; ---- Absolute minimum/maximum
;;
;; == Ternary arithmetic
;; ---- [INT] Dot product
@@ -1264,6 +1265,23 @@ (define_insn "*aarch64_sme_single_<optab><mode>_plus"
"<sme_int_op>\tza.<Vetype>[%w0, %1, vgx<vector_count>], %2, %3.<Vetype>"
)
Sorry for picking up on this after it was committed, but:
+;; -------------------------------------------------------------------------
+;; ---- Absolute minimum/maximum
+;; -------------------------------------------------------------------------
+;; Includes:
+;; - svamin (SME2+faminmax)
+;; - svamin (SME2+faminmax)
Even though these are currently exclusively used by the ACLE, I think we should
be listing the names of the instructions here. It looks like the convention
would be to capitalise "+faminax" too.
Sure, that makes sense.
+;; -------------------------------------------------------------------------
+
+(define_insn "@aarch64_sme_<faminmax_uns_op><mode>"
+ [(set (match_operand:SVE_Fx24 0 "register_operand" "=Uw<vector_count>")
+ (unspec:SVE_Fx24 [(match_operand:SVE_Fx24 1 "register_operand" "%0")
+ (match_operand:SVE_Fx24 2 "register_operand"
"Uw<vector_count>")]
+ FAMINMAX_UNS))]
+ "TARGET_SME2 && TARGET_FAMINMAX"
The frontend is rightly ensuring that the associated intrinsics can only be
used within streaming mode functions; IIUC, this predication should be:
"TARGET_STREAMING_SME2 && TARGET_FAMINMAX"
Agh, yeah you're right, I will clean this up with a new patch.
+ "<faminmax_uns_op>\t%0, %1, %2"
+)
+
Thanks,
Spencer
