* [PATCH][2/5] aarch64: [US]Q(R)SHR(U)N scalar forms refactoring
@ 2023-06-16 13:07 Kyrylo Tkachov
0 siblings, 0 replies; only message in thread
From: Kyrylo Tkachov @ 2023-06-16 13:07 UTC (permalink / raw)
To: gcc-patches
[-- Attachment #1: Type: text/plain, Size: 1281 bytes --]
Some instructions from the previous patch have scalar forms:
SQSHRN,SQRSHRN,UQSHRN,UQRSHRN,SQSHRUN,SQRSHRUN.
This patch converts the patterns for these to use standard RTL codes.
Their MD patterns deviate slightly from the vector forms mostly due to
things like operands being scalar rather than vectors.
One nuance is in the SQSHRUN,SQRSHRUN patterns. These end in a truncate
to the scalar narrow mode e.g. SI -> QI. This gets simplified by the
RTL passes to a subreg rather than keeping it as a truncate.
So we end up representing these without the truncate and in the expander
read the narrow subreg in order to comply with the expected width of the
intrinsic.
Bootstrapped and tested on aarch64-none-linux-gnu and
aarch64_be-none-elf.
gcc/ChangeLog:
* config/aarch64/aarch64-simd.md (aarch64_<sur>q<r>shr<u>n_n<mode>):
Rename to...
(aarch64_<shrn_op>shrn_n<mode>): ... This. Reimplement with RTL codes.
(*aarch64_<shrn_op>rshrn_n<mode>_insn): New define_insn.
(aarch64_sqrshrun_n<mode>_insn): Likewise.
(aarch64_sqshrun_n<mode>_insn): Likewise.
(aarch64_<shrn_op>rshrn_n<mode>): New define_expand.
(aarch64_sqshrun_n<mode>): Likewise.
(aarch64_sqrshrun_n<mode>): Likewise.
* config/aarch64/iterators.md (V2XWIDE): Add HI and SI modes.
[-- Attachment #2: s2.patch --]
[-- Type: application/octet-stream, Size: 7795 bytes --]
commit d20b2ad845876eec0ee80a3933ad49f9f6c4ee30
Author: Kyrylo Tkachov <kyrylo.tkachov@arm.com>
Date: Tue Jun 6 23:35:52 2023 +0100
aarch64: [US]Q(R)SHR(U)N scalar forms refactoring
Some instructions from the previous patch have scalar forms:
SQSHRN,SQRSHRN,UQSHRN,UQRSHRN,SQSHRUN,SQRSHRUN.
This patch converts the patterns for these to use standard RTL codes.
Their MD patterns deviate slightly from the vector forms mostly due to
things like operands being scalar rather than vectors.
One nuance is in the SQSHRUN,SQRSHRUN patterns. These end in a truncate
to the scalar narrow mode e.g. SI -> QI. This gets simplified by the
RTL passes to a subreg rather than keeping it as a truncate.
So we end up representing these without the truncate and in the expander
read the narrow subreg in order to comply with the expected width of the
intrinsic.
Bootstrapped and tested on aarch64-none-linux-gnu and
aarch64_be-none-elf.
gcc/ChangeLog:
* config/aarch64/aarch64-simd.md (aarch64_<sur>q<r>shr<u>n_n<mode>):
Rename to...
(aarch64_<shrn_op>shrn_n<mode>): ... This. Reimplement with RTL codes.
(*aarch64_<shrn_op>rshrn_n<mode>_insn): New define_insn.
(aarch64_sqrshrun_n<mode>_insn): Likewise.
(aarch64_sqshrun_n<mode>_insn): Likewise.
(aarch64_<shrn_op>rshrn_n<mode>): New define_expand.
(aarch64_sqshrun_n<mode>): Likewise.
(aarch64_sqrshrun_n<mode>): Likewise.
* config/aarch64/iterators.md (V2XWIDE): Add HI and SI modes.
diff --git a/gcc/config/aarch64/aarch64-simd.md b/gcc/config/aarch64/aarch64-simd.md
index 8b92981bebb..bbb54344eb7 100644
--- a/gcc/config/aarch64/aarch64-simd.md
+++ b/gcc/config/aarch64/aarch64-simd.md
@@ -6654,15 +6654,15 @@ (define_insn "aarch64_<sur>qshl<u>_n<mode>"
;; vq(r)shr(u)n_n
-(define_insn "aarch64_<sur>q<r>shr<u>n_n<mode>"
+(define_insn "aarch64_<shrn_op>shrn_n<mode>"
[(set (match_operand:<VNARROWQ> 0 "register_operand" "=w")
- (unspec:<VNARROWQ> [(match_operand:SD_HSDI 1 "register_operand" "w")
- (match_operand:SI 2
- "aarch64_simd_shift_imm_offset_<ve_mode>" "i")]
- VQSHRN_N))]
+ (SAT_TRUNC:<VNARROWQ>
+ (<TRUNC_SHIFT>:SD_HSDI
+ (match_operand:SD_HSDI 1 "register_operand" "w")
+ (match_operand:SI 2 "aarch64_simd_shift_imm_offset_<ve_mode>"))))]
"TARGET_SIMD"
- "<sur>q<r>shr<u>n\\t%<vn2>0<Vmntype>, %<v>1<Vmtype>, %2"
- [(set_attr "type" "neon_sat_shift_imm_narrow_q")]
+ "<shrn_op>shrn\t%<vn2>0<Vmntype>, %<v>1<Vmtype>, %2"
+ [(set_attr "type" "neon_shift_imm_narrow_q")]
)
(define_insn "*aarch64_<shrn_op>shrn_n<mode>_insn<vczle><vczbe>"
@@ -6704,6 +6704,41 @@ (define_insn "*aarch64_<shrn_op>rshrn_n<mode>_insn<vczle><vczbe>"
[(set_attr "type" "neon_shift_imm_narrow_q")]
)
+(define_insn "*aarch64_<shrn_op>rshrn_n<mode>_insn"
+ [(set (match_operand:<VNARROWQ> 0 "register_operand" "=w")
+ (SAT_TRUNC:<VNARROWQ>
+ (<TRUNC_SHIFT>:<DWI>
+ (plus:<DWI>
+ (<TRUNCEXTEND>:<DWI>
+ (match_operand:SD_HSDI 1 "register_operand" "w"))
+ (match_operand:<DWI> 3 "aarch64_simd_rsra_rnd_imm_vec"))
+ (match_operand:SI 2 "aarch64_simd_shift_imm_offset_<ve_mode>"))))]
+ "TARGET_SIMD
+ && aarch64_const_vec_rnd_cst_p (operands[3], operands[2])"
+ "<shrn_op>rshrn\t%<vn2>0<Vmntype>, %<v>1<Vmtype>, %2"
+ [(set_attr "type" "neon_shift_imm_narrow_q")]
+)
+
+(define_expand "aarch64_<shrn_op>rshrn_n<mode>"
+ [(set (match_operand:<VNARROWQ> 0 "register_operand")
+ (SAT_TRUNC:<VNARROWQ>
+ (<TRUNC_SHIFT>:<V2XWIDE>
+ (plus:<V2XWIDE>
+ (<TRUNCEXTEND>:<V2XWIDE>
+ (match_operand:SD_HSDI 1 "register_operand"))
+ (match_dup 3))
+ (match_operand:SI 2 "aarch64_simd_shift_imm_offset_<ve_mode>"))))]
+ "TARGET_SIMD"
+ {
+ /* Use this expander to create the rounding constant vector, which is
+ 1 << (shift - 1). Use wide_int here to ensure that the right TImode
+ RTL is generated when handling the DImode expanders. */
+ int prec = GET_MODE_UNIT_PRECISION (<V2XWIDE>mode);
+ wide_int rnd_wi = wi::set_bit_in_zero (INTVAL (operands[2]) - 1, prec);
+ operands[3] = immed_wide_int_const (rnd_wi, GET_MODE_INNER (<V2XWIDE>mode));
+ }
+)
+
(define_expand "aarch64_<shrn_op>rshrn_n<mode>"
[(set (match_operand:<VNARROWQ> 0 "register_operand")
(ALL_TRUNC:<VNARROWQ>
@@ -6748,6 +6783,34 @@ (define_insn "*aarch64_sqshrun_n<mode>_insn<vczle><vczbe>"
[(set_attr "type" "neon_shift_imm_narrow_q")]
)
+(define_insn "aarch64_sqshrun_n<mode>_insn"
+ [(set (match_operand:SD_HSDI 0 "register_operand" "=w")
+ (smin:SD_HSDI
+ (smax:SD_HSDI
+ (ashiftrt:SD_HSDI
+ (match_operand:SD_HSDI 1 "register_operand" "w")
+ (match_operand:SI 2 "aarch64_simd_shift_imm_offset_<ve_mode>"))
+ (const_int 0))
+ (const_int <half_mask>)))]
+ "TARGET_SIMD"
+ "sqshrun\t%<vn2>0<Vmntype>, %<v>1<Vmtype>, %2"
+ [(set_attr "type" "neon_shift_imm_narrow_q")]
+)
+
+(define_expand "aarch64_sqshrun_n<mode>"
+ [(match_operand:<VNARROWQ> 0 "register_operand")
+ (match_operand:SD_HSDI 1 "register_operand")
+ (match_operand:SI 2 "aarch64_simd_shift_imm_offset_<ve_mode>")]
+ "TARGET_SIMD"
+ {
+ rtx dst = gen_reg_rtx (<MODE>mode);
+ emit_insn (gen_aarch64_sqshrun_n<mode>_insn (dst, operands[1],
+ operands[2]));
+ emit_move_insn (operands[0], gen_lowpart (<VNARROWQ>mode, dst));
+ DONE;
+ }
+)
+
(define_expand "aarch64_sqshrun_n<mode>"
[(set (match_operand:<VNARROWQ> 0 "register_operand")
(truncate:<VNARROWQ>
@@ -6788,6 +6851,40 @@ (define_insn "*aarch64_sqrshrun_n<mode>_insn<vczle><vczbe>"
[(set_attr "type" "neon_shift_imm_narrow_q")]
)
+(define_insn "aarch64_sqrshrun_n<mode>_insn"
+ [(set (match_operand:<V2XWIDE> 0 "register_operand" "=w")
+ (smin:<V2XWIDE>
+ (smax:<V2XWIDE>
+ (ashiftrt:<V2XWIDE>
+ (plus:<V2XWIDE>
+ (sign_extend:<V2XWIDE>
+ (match_operand:SD_HSDI 1 "register_operand" "w"))
+ (match_operand:<V2XWIDE> 3 "aarch64_simd_rsra_rnd_imm_vec"))
+ (match_operand:SI 2 "aarch64_simd_shift_imm_offset_<ve_mode>"))
+ (const_int 0))
+ (const_int <half_mask>)))]
+ "TARGET_SIMD
+ && aarch64_const_vec_rnd_cst_p (operands[3], operands[2])"
+ "sqrshrun\t%<vn2>0<Vmntype>, %<v>1<Vmtype>, %2"
+ [(set_attr "type" "neon_shift_imm_narrow_q")]
+)
+
+(define_expand "aarch64_sqrshrun_n<mode>"
+ [(match_operand:<VNARROWQ> 0 "register_operand")
+ (match_operand:SD_HSDI 1 "register_operand")
+ (match_operand:SI 2 "aarch64_simd_shift_imm_offset_<ve_mode>")]
+ "TARGET_SIMD"
+ {
+ int prec = GET_MODE_UNIT_PRECISION (<V2XWIDE>mode);
+ wide_int rnd_wi = wi::set_bit_in_zero (INTVAL (operands[2]) - 1, prec);
+ rtx rnd = immed_wide_int_const (rnd_wi, <V2XWIDE>mode);
+ rtx dst = gen_reg_rtx (<V2XWIDE>mode);
+ emit_insn (gen_aarch64_sqrshrun_n<mode>_insn (dst, operands[1], operands[2], rnd));
+ emit_move_insn (operands[0], gen_lowpart (<VNARROWQ>mode, dst));
+ DONE;
+ }
+)
+
(define_expand "aarch64_sqrshrun_n<mode>"
[(set (match_operand:<VNARROWQ> 0 "register_operand")
(truncate:<VNARROWQ>
diff --git a/gcc/config/aarch64/iterators.md b/gcc/config/aarch64/iterators.md
index e8c62c88b14..acc7a3ec46e 100644
--- a/gcc/config/aarch64/iterators.md
+++ b/gcc/config/aarch64/iterators.md
@@ -1532,7 +1532,8 @@ (define_mode_attr VWIDE [(V8QI "V8HI") (V4HI "V4SI")
(define_mode_attr V2XWIDE [(V8QI "V8HI") (V4HI "V4SI")
(V16QI "V16HI") (V8HI "V8SI")
(V2SI "V2DI") (V4SI "V4DI")
- (V2DI "V2TI") (DI "TI")])
+ (V2DI "V2TI") (DI "TI")
+ (HI "SI") (SI "DI")])
;; Predicate mode associated with VWIDE.
(define_mode_attr VWIDE_PRED [(VNx8HF "VNx4BI") (VNx4SF "VNx2BI")])
^ permalink raw reply [flat|nested] only message in thread
only message in thread, other threads:[~2023-06-16 13:07 UTC | newest]
Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2023-06-16 13:07 [PATCH][2/5] aarch64: [US]Q(R)SHR(U)N scalar forms refactoring Kyrylo Tkachov
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).