https://gcc.gnu.org/g:ac4cdf5cb43c0b09e81760e2a1902ceebcf1a135
commit r15-3997-gac4cdf5cb43c0b09e81760e2a1902ceebcf1a135
Author: Saurabh Jha <saurabh....@arm.com>
Date: Mon Sep 30 10:37:16 2024 +0000
aarch64: Introduce new unspecs for smax/smin
Introduce two new unspecs, UNSPEC_COND_SMAX and UNSPEC_COND_SMIN,
corresponding to rtl operators smax and smin. UNSPEC_COND_SMAX is used
to generate fmaxnm instruction and UNSPEC_COND_SMIN is used to generate
fminnm instruction.
With these new unspecs, we can generate SVE2 max/min instructions using
existing generic unpredicated and predicated instruction patterns that
use optab attribute. Thus, we have removed specialised instruction
patterns for max/min instructions that were using
SVE_COND_FP_MAXMIN_PUBLIC iterator.
No new test cases as the existing test cases should be enough to test
this refactoring.
gcc/ChangeLog:
* config/aarch64/aarch64-sve.md
(<fmaxmin><mode>3): Remove this instruction pattern.
(cond_<fmaxmin><mode>): Remove this instruction pattern.
* config/aarch64/iterators.md: New unspecs and changes to
iterators and attrs to use the new unspecs
Diff:
---
gcc/config/aarch64/aarch64-sve.md | 33 ------------------
gcc/config/aarch64/iterators.md | 73 ++++++++++++++++++++++++---------------
2 files changed, 45 insertions(+), 61 deletions(-)
diff --git a/gcc/config/aarch64/aarch64-sve.md
b/gcc/config/aarch64/aarch64-sve.md
index f6c7c2f4cb31..ec1d059a2b1b 100644
--- a/gcc/config/aarch64/aarch64-sve.md
+++ b/gcc/config/aarch64/aarch64-sve.md
@@ -6600,39 +6600,6 @@
;; - FMINNM
;; -------------------------------------------------------------------------
-;; Unpredicated fmax/fmin (the libm functions). The optabs for the
-;; smax/smin rtx codes are handled in the generic section above.
-(define_expand "<fmaxmin><mode>3"
- [(set (match_operand:SVE_FULL_F 0 "register_operand")
- (unspec:SVE_FULL_F
- [(match_dup 3)
- (const_int SVE_RELAXED_GP)
- (match_operand:SVE_FULL_F 1 "register_operand")
- (match_operand:SVE_FULL_F 2 "aarch64_sve_float_maxmin_operand")]
- SVE_COND_FP_MAXMIN_PUBLIC))]
- "TARGET_SVE"
- {
- operands[3] = aarch64_ptrue_reg (<VPRED>mode);
- }
-)
-
-;; Predicated fmax/fmin (the libm functions). The optabs for the
-;; smax/smin rtx codes are handled in the generic section above.
-(define_expand "cond_<fmaxmin><mode>"
- [(set (match_operand:SVE_FULL_F 0 "register_operand")
- (unspec:SVE_FULL_F
- [(match_operand:<VPRED> 1 "register_operand")
- (unspec:SVE_FULL_F
- [(match_dup 1)
- (const_int SVE_RELAXED_GP)
- (match_operand:SVE_FULL_F 2 "register_operand")
- (match_operand:SVE_FULL_F 3 "aarch64_sve_float_maxmin_operand")]
- SVE_COND_FP_MAXMIN_PUBLIC)
- (match_operand:SVE_FULL_F 4 "aarch64_simd_reg_or_zero")]
- UNSPEC_SEL))]
- "TARGET_SVE"
-)
-
;; Predicated floating-point maximum/minimum.
(define_insn "@aarch64_pred_<optab><mode>"
[(set (match_operand:SVE_FULL_F 0 "register_operand")
diff --git a/gcc/config/aarch64/iterators.md b/gcc/config/aarch64/iterators.md
index c2fcd18306e4..0836dee61c9f 100644
--- a/gcc/config/aarch64/iterators.md
+++ b/gcc/config/aarch64/iterators.md
@@ -881,6 +881,8 @@
UNSPEC_COND_FSQRT ; Used in aarch64-sve.md.
UNSPEC_COND_FSUB ; Used in aarch64-sve.md.
UNSPEC_COND_SCVTF ; Used in aarch64-sve.md.
+ UNSPEC_COND_SMAX ; Used in aarch64-sve.md.
+ UNSPEC_COND_SMIN ; Used in aarch64-sve.md.
UNSPEC_COND_UCVTF ; Used in aarch64-sve.md.
UNSPEC_LASTA ; Used in aarch64-sve.md.
UNSPEC_LASTB ; Used in aarch64-sve.md.
@@ -3081,15 +3083,18 @@
(define_int_iterator SVE_COND_FCVTI [UNSPEC_COND_FCVTZS UNSPEC_COND_FCVTZU])
(define_int_iterator SVE_COND_ICVTF [UNSPEC_COND_SCVTF UNSPEC_COND_UCVTF])
-(define_int_iterator SVE_COND_FP_BINARY [UNSPEC_COND_FADD
- UNSPEC_COND_FDIV
- UNSPEC_COND_FMAX
- UNSPEC_COND_FMAXNM
- UNSPEC_COND_FMIN
- UNSPEC_COND_FMINNM
- UNSPEC_COND_FMUL
- UNSPEC_COND_FMULX
- UNSPEC_COND_FSUB])
+(define_int_iterator SVE_COND_FP_BINARY
+ [UNSPEC_COND_FADD
+ UNSPEC_COND_FDIV
+ UNSPEC_COND_FMAX
+ UNSPEC_COND_FMAXNM
+ UNSPEC_COND_FMIN
+ UNSPEC_COND_FMINNM
+ UNSPEC_COND_FMUL
+ UNSPEC_COND_FMULX
+ UNSPEC_COND_FSUB
+ UNSPEC_COND_SMAX
+ UNSPEC_COND_SMIN])
;; Same as SVE_COND_FP_BINARY, but without codes that have a dedicated
;; <optab><mode>3 expander.
@@ -3100,7 +3105,9 @@
UNSPEC_COND_FMINNM
UNSPEC_COND_FMUL
UNSPEC_COND_FMULX
- UNSPEC_COND_FSUB])
+ UNSPEC_COND_FSUB
+ UNSPEC_COND_SMAX
+ UNSPEC_COND_SMIN])
(define_int_iterator SVE_COND_FP_BINARY_INT [UNSPEC_COND_FSCALE])
@@ -3112,10 +3119,15 @@
UNSPEC_COND_FMAXNM
UNSPEC_COND_FMIN
UNSPEC_COND_FMINNM
- UNSPEC_COND_FMUL])
+ UNSPEC_COND_FMUL
+ UNSPEC_COND_SMAX
+ UNSPEC_COND_SMIN])
-(define_int_iterator SVE_COND_FP_BINARY_REG [UNSPEC_COND_FDIV
- UNSPEC_COND_FMULX])
+(define_int_iterator SVE_COND_FP_BINARY_REG
+ [UNSPEC_COND_FDIV
+ UNSPEC_COND_FMULX
+ UNSPEC_COND_SMAX
+ UNSPEC_COND_SMIN])
(define_int_iterator SVE_COND_FCADD [UNSPEC_COND_FCADD90
UNSPEC_COND_FCADD270])
@@ -3125,11 +3137,6 @@
UNSPEC_COND_FMIN
UNSPEC_COND_FMINNM])
-;; Floating-point max/min operations that correspond to optabs,
-;; as opposed to those that are internal to the port.
-(define_int_iterator SVE_COND_FP_MAXMIN_PUBLIC [UNSPEC_COND_FMAXNM
- UNSPEC_COND_FMINNM])
-
(define_int_iterator SVE_COND_FP_TERNARY [UNSPEC_COND_FMLA
UNSPEC_COND_FMLS
UNSPEC_COND_FNMLA
@@ -3705,9 +3712,9 @@
(UNSPEC_COND_FCVTZU "fixuns_trunc")
(UNSPEC_COND_FDIV "div")
(UNSPEC_COND_FMAX "fmax_nan")
- (UNSPEC_COND_FMAXNM "smax")
+ (UNSPEC_COND_FMAXNM "fmax")
(UNSPEC_COND_FMIN "fmin_nan")
- (UNSPEC_COND_FMINNM "smin")
+ (UNSPEC_COND_FMINNM "fmin")
(UNSPEC_COND_FMLA "fma")
(UNSPEC_COND_FMLS "fnma")
(UNSPEC_COND_FMUL "mul")
@@ -3727,6 +3734,8 @@
(UNSPEC_COND_FSQRT "sqrt")
(UNSPEC_COND_FSUB "sub")
(UNSPEC_COND_SCVTF "float")
+ (UNSPEC_COND_SMAX "smax")
+ (UNSPEC_COND_SMIN "smin")
(UNSPEC_COND_UCVTF "floatuns")])
(define_int_attr fmaxmin [(UNSPEC_FMAX "fmax_nan")
@@ -3734,9 +3743,7 @@
(UNSPEC_FMAXNMV "fmax")
(UNSPEC_FMIN "fmin_nan")
(UNSPEC_FMINNM "fmin")
- (UNSPEC_FMINNMV "fmin")
- (UNSPEC_COND_FMAXNM "fmax")
- (UNSPEC_COND_FMINNM "fmin")])
+ (UNSPEC_FMINNMV "fmin")])
(define_int_attr maxmin_uns_op [(UNSPEC_UMAXV "umax")
(UNSPEC_UMINV "umin")
@@ -4251,7 +4258,9 @@
(UNSPEC_COND_FRINTZ "frintz")
(UNSPEC_COND_FSCALE "fscale")
(UNSPEC_COND_FSQRT "fsqrt")
- (UNSPEC_COND_FSUB "fsub")])
+ (UNSPEC_COND_FSUB "fsub")
+ (UNSPEC_COND_SMAX "fmaxnm")
+ (UNSPEC_COND_SMIN "fminnm")])
(define_int_attr sve_fp_op_rev [(UNSPEC_COND_FADD "fadd")
(UNSPEC_COND_FDIV "fdivr")
@@ -4261,7 +4270,9 @@
(UNSPEC_COND_FMINNM "fminnm")
(UNSPEC_COND_FMUL "fmul")
(UNSPEC_COND_FMULX "fmulx")
- (UNSPEC_COND_FSUB "fsubr")])
+ (UNSPEC_COND_FSUB "fsubr")
+ (UNSPEC_COND_SMAX "fmaxnm")
+ (UNSPEC_COND_SMIN "fminnm")])
(define_int_attr sme_int_op [(UNSPEC_SME_ADD_WRITE "add")
(UNSPEC_SME_SUB_WRITE "sub")])
@@ -4397,7 +4408,9 @@
(UNSPEC_COND_FMINNM "register_operand")
(UNSPEC_COND_FMUL "register_operand")
(UNSPEC_COND_FMULX "register_operand")
- (UNSPEC_COND_FSUB "aarch64_sve_float_arith_operand")])
+ (UNSPEC_COND_FSUB "aarch64_sve_float_arith_operand")
+ (UNSPEC_COND_SMAX "register_operand")
+ (UNSPEC_COND_SMIN "register_operand")])
;; The predicate to use for the second input operand in a floating-point
;; <optab><mode>3 pattern.
@@ -4410,7 +4423,9 @@
(UNSPEC_COND_FMINNM "aarch64_sve_float_maxmin_operand")
(UNSPEC_COND_FMUL "aarch64_sve_float_mul_operand")
(UNSPEC_COND_FMULX "register_operand")
- (UNSPEC_COND_FSUB "register_operand")])
+ (UNSPEC_COND_FSUB "register_operand")
+ (UNSPEC_COND_SMAX "aarch64_sve_float_maxmin_operand")
+ (UNSPEC_COND_SMIN "aarch64_sve_float_maxmin_operand")])
;; Likewise for immediates only.
(define_int_attr sve_pred_fp_rhs2_immediate
@@ -4418,7 +4433,9 @@
(UNSPEC_COND_FMAXNM "aarch64_sve_float_maxmin_immediate")
(UNSPEC_COND_FMIN "aarch64_sve_float_maxmin_immediate")
(UNSPEC_COND_FMINNM "aarch64_sve_float_maxmin_immediate")
- (UNSPEC_COND_FMUL "aarch64_sve_float_mul_immediate")])
+ (UNSPEC_COND_FMUL "aarch64_sve_float_mul_immediate")
+ (UNSPEC_COND_SMAX "aarch64_sve_float_maxmin_immediate")
+ (UNSPEC_COND_SMIN "aarch64_sve_float_maxmin_immediate")])
;; The maximum number of element bits that an instruction can handle.
(define_int_attr max_elem_bits [(UNSPEC_UADDV "64") (UNSPEC_SADDV "32")
