This patch makes us use the same define_insn for both the smax/smin
and fmax/fmin optabs. It also continues the process started by
the earlier FP unary patch of moving predicated FP patterns from
rtx codes to unspecs.
There's no need to handle the FMAX and FMIN instructions until
the ACLE patch, since we only use FMAXNM and FMINNM at present.
Tested on aarch64-linux-gnu (with and without SVE) and aarch64_be-elf.
Applied as r274188.
Richard
2019-08-07 Richard Sandiford <richard.sandif...@arm.com>
gcc/
* config/aarch64/iterators.md (SVE_COND_FP_MAXMIN_PUBLIC): New
int iterator.
(maxmin_uns_op): Handle UNSPEC_COND_FMAXNM and UNSPEC_COND_FMINNM.
* config/aarch64/aarch64-sve.md
(<FMAXMIN:su><FMAXMIN:maxmin><SVE_F:mode>3): Rename to...
(<SVE_COND_FP_MAXMIN_PUBLIC:optab><SVE_F:mode>3): ...this and
use a single unspec for the rhs.
(*<su><maxmin><mode>3): Delete.
(<maxmin_uns><SVE_F:mode>3): Use a single unspec for the rhs.
Index: gcc/config/aarch64/iterators.md
===================================================================
--- gcc/config/aarch64/iterators.md 2019-08-07 19:51:33.208173107 +0100
+++ gcc/config/aarch64/iterators.md 2019-08-07 19:53:53.831126190 +0100
@@ -1630,6 +1630,11 @@ (define_int_iterator SVE_COND_FP_BINARY
UNSPEC_COND_FMUL
UNSPEC_COND_FSUB])
+;; 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
@@ -1709,7 +1714,9 @@ (define_int_attr maxmin_uns [(UNSPEC_UM
(UNSPEC_FMINNMV "smin")
(UNSPEC_FMINV "smin_nan")
(UNSPEC_FMAXNM "fmax")
- (UNSPEC_FMINNM "fmin")])
+ (UNSPEC_FMINNM "fmin")
+ (UNSPEC_COND_FMAXNM "fmax")
+ (UNSPEC_COND_FMINNM "fmin")])
(define_int_attr maxmin_uns_op [(UNSPEC_UMAXV "umax")
(UNSPEC_UMINV "umin")
Index: gcc/config/aarch64/aarch64-sve.md
===================================================================
--- gcc/config/aarch64/aarch64-sve.md 2019-08-07 19:51:33.204173137 +0100
+++ gcc/config/aarch64/aarch64-sve.md 2019-08-07 19:53:53.831126190 +0100
@@ -2186,69 +2186,51 @@ (define_expand "xorsign<mode>3"
;; ---- [FP] Maximum and minimum
;; -------------------------------------------------------------------------
;; Includes:
-;; - FMAX
;; - FMAXNM
-;; - FMIN
;; - FMINNM
;; -------------------------------------------------------------------------
-;; Unpredicated floating-point MAX/MIN.
-(define_expand "<su><maxmin><mode>3"
+;; Unpredicated floating-point MAX/MIN (the rtx codes). These are more
+;; relaxed than fmax/fmin, but we implement them in the same way.
+(define_expand "<optab><mode>3"
[(set (match_operand:SVE_F 0 "register_operand")
(unspec:SVE_F
[(match_dup 3)
- (FMAXMIN:SVE_F (match_operand:SVE_F 1 "register_operand")
- (match_operand:SVE_F 2 "register_operand"))]
- UNSPEC_MERGE_PTRUE))]
+ (match_operand:SVE_F 1 "register_operand")
+ (match_operand:SVE_F 2 "register_operand")]
+ SVE_COND_FP_MAXMIN_PUBLIC))]
"TARGET_SVE"
{
operands[3] = aarch64_ptrue_reg (<VPRED>mode);
}
)
-;; Floating-point MAX/MIN predicated with a PTRUE.
-(define_insn "*<su><maxmin><mode>3"
- [(set (match_operand:SVE_F 0 "register_operand" "=w, ?&w")
- (unspec:SVE_F
- [(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
- (FMAXMIN:SVE_F (match_operand:SVE_F 2 "register_operand" "%0, w")
- (match_operand:SVE_F 3 "register_operand" "w, w"))]
- UNSPEC_MERGE_PTRUE))]
- "TARGET_SVE"
- "@
- f<maxmin>nm\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
- movprfx\t%0, %2\;f<maxmin>nm\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>"
- [(set_attr "movprfx" "*,yes")]
-)
-
-;; Unpredicated fmax/fmin.
+;; Unpredicated fmax/fmin (the libm functions).
(define_expand "<maxmin_uns><mode>3"
[(set (match_operand:SVE_F 0 "register_operand")
(unspec:SVE_F
[(match_dup 3)
- (unspec:SVE_F [(match_operand:SVE_F 1 "register_operand")
- (match_operand:SVE_F 2 "register_operand")]
- FMAXMIN_UNS)]
- UNSPEC_MERGE_PTRUE))]
+ (match_operand:SVE_F 1 "register_operand")
+ (match_operand:SVE_F 2 "register_operand")]
+ SVE_COND_FP_MAXMIN_PUBLIC))]
"TARGET_SVE"
{
operands[3] = aarch64_ptrue_reg (<VPRED>mode);
}
)
-;; fmax/fmin predicated with a PTRUE.
-(define_insn "*<maxmin_uns><mode>3"
+;; Predicated floating-point maximum/minimum.
+(define_insn "*<optab><mode>3"
[(set (match_operand:SVE_F 0 "register_operand" "=w, ?&w")
(unspec:SVE_F
[(match_operand:<VPRED> 1 "register_operand" "Upl, Upl")
- (unspec:SVE_F [(match_operand:SVE_F 2 "register_operand" "%0, w")
- (match_operand:SVE_F 3 "register_operand" "w, w")]
- FMAXMIN_UNS)]
- UNSPEC_MERGE_PTRUE))]
+ (match_operand:SVE_F 2 "register_operand" "%0, w")
+ (match_operand:SVE_F 3 "register_operand" "w, w")]
+ SVE_COND_FP_MAXMIN_PUBLIC))]
"TARGET_SVE"
"@
- <maxmin_uns_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
- movprfx\t%0, %2\;<maxmin_uns_op>\t%0.<Vetype>, %1/m, %0.<Vetype>,
%3.<Vetype>"
+ <sve_fp_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>
+ movprfx\t%0, %2\;<sve_fp_op>\t%0.<Vetype>, %1/m, %0.<Vetype>, %3.<Vetype>"
[(set_attr "movprfx" "*,yes")]
)
