* [PATCH][committed] aarch64: Reimplement RSHRN intrinsic patterns with standard RTL codes
@ 2023-04-26 14:09 Kyrylo Tkachov
0 siblings, 0 replies; only message in thread
From: Kyrylo Tkachov @ 2023-04-26 14:09 UTC (permalink / raw)
To: gcc-patches
[-- Attachment #1: Type: text/plain, Size: 804 bytes --]
Hi all,
This patch reimplements the backend patterns for the rshrn intrinsics using standard RTL codes rather than UNSPECS.
We already represent shrn as truncate of a shift. rshrn can be represented as truncate (src + (1 << (shft - 1)) >> shft),
similar to how LLVM treats it.
I have a follow-up patch to do the same for the rshrn2 pattern, which will allow us to remove the UNSPEC_RSHRN entirely.
Bootstrapped and tested on aarch64-none-linux-gnu.
Pushing to trunk.
Thanks,
Kyrill
gcc/ChangeLog:
* config/aarch64/aarch64-simd.md (aarch64_rshrn<mode>_insn_le): Reimplement
with standard RTL codes instead of an UNSPEC.
(aarch64_rshrn<mode>_insn_be): Likewise.
(aarch64_rshrn<mode>): Adjust for the above.
* config/aarch64/predicates.md (aarch64_simd_rshrn_imm_vec): Define.
[-- Attachment #2: rshrn.patch --]
[-- Type: application/octet-stream, Size: 3905 bytes --]
diff --git a/gcc/config/aarch64/aarch64-simd.md b/gcc/config/aarch64/aarch64-simd.md
index 9f4785140c0b7511a7c8d4803bd00017b8687bf5..47a52482149f9ead50bbeba9b886ce26aff18adb 100644
--- a/gcc/config/aarch64/aarch64-simd.md
+++ b/gcc/config/aarch64/aarch64-simd.md
@@ -1918,11 +1918,15 @@ (define_expand "aarch64_shrn<mode>"
(define_insn "aarch64_rshrn<mode>_insn_le"
[(set (match_operand:<VNARROWQ2> 0 "register_operand" "=w")
(vec_concat:<VNARROWQ2>
- (unspec:<VNARROWQ> [(match_operand:VQN 1 "register_operand" "w")
- (match_operand:VQN 2
- "aarch64_simd_shift_imm_vec_<vn_mode>")] UNSPEC_RSHRN)
- (match_operand:<VNARROWQ> 3 "aarch64_simd_or_scalar_imm_zero")))]
- "TARGET_SIMD && !BYTES_BIG_ENDIAN"
+ (truncate:<VNARROWQ>
+ (lshiftrt:VQN
+ (plus:VQN (match_operand:VQN 1 "register_operand" "w")
+ (match_operand:VQN 3 "aarch64_simd_rshrn_imm_vec"))
+ (match_operand:VQN 2 "aarch64_simd_shift_imm_vec_<vn_mode>")))
+ (match_operand:<VNARROWQ> 4 "aarch64_simd_or_scalar_imm_zero")))]
+ "TARGET_SIMD && !BYTES_BIG_ENDIAN
+ && INTVAL (CONST_VECTOR_ELT (operands[3], 0))
+ == (HOST_WIDE_INT_1 << (INTVAL (CONST_VECTOR_ELT (operands[2], 0)) - 1))"
"rshrn\\t%0.<Vntype>, %1.<Vtype>, %2"
[(set_attr "type" "neon_shift_imm_narrow_q")]
)
@@ -1930,11 +1934,15 @@ (define_insn "aarch64_rshrn<mode>_insn_le"
(define_insn "aarch64_rshrn<mode>_insn_be"
[(set (match_operand:<VNARROWQ2> 0 "register_operand" "=w")
(vec_concat:<VNARROWQ2>
- (match_operand:<VNARROWQ> 3 "aarch64_simd_or_scalar_imm_zero")
- (unspec:<VNARROWQ> [(match_operand:VQN 1 "register_operand" "w")
- (match_operand:VQN 2 "aarch64_simd_shift_imm_vec_<vn_mode>")]
- UNSPEC_RSHRN)))]
- "TARGET_SIMD && BYTES_BIG_ENDIAN"
+ (match_operand:<VNARROWQ> 4 "aarch64_simd_or_scalar_imm_zero")
+ (truncate:<VNARROWQ>
+ (lshiftrt:VQN
+ (plus:VQN (match_operand:VQN 1 "register_operand" "w")
+ (match_operand:VQN 3 "aarch64_simd_rshrn_imm_vec"))
+ (match_operand:VQN 2 "aarch64_simd_shift_imm_vec_<vn_mode>")))))]
+ "TARGET_SIMD && BYTES_BIG_ENDIAN
+ && INTVAL (CONST_VECTOR_ELT (operands[3], 0))
+ == (HOST_WIDE_INT_1 << (INTVAL (CONST_VECTOR_ELT (operands[2], 0)) - 1))"
"rshrn\\t%0.<Vntype>, %1.<Vtype>, %2"
[(set_attr "type" "neon_shift_imm_narrow_q")]
)
@@ -1952,18 +1960,22 @@ (define_expand "aarch64_rshrn<mode>"
}
else
{
+ rtx shft
+ = aarch64_simd_gen_const_vector_dup (<MODE>mode,
+ HOST_WIDE_INT_1U
+ << (INTVAL (operands[2]) - 1));
rtx tmp = gen_reg_rtx (<VNARROWQ2>mode);
operands[2] = aarch64_simd_gen_const_vector_dup (<MODE>mode,
INTVAL (operands[2]));
if (BYTES_BIG_ENDIAN)
emit_insn (
gen_aarch64_rshrn<mode>_insn_be (tmp, operands[1],
- operands[2],
+ operands[2], shft,
CONST0_RTX (<VNARROWQ>mode)));
else
emit_insn (
gen_aarch64_rshrn<mode>_insn_le (tmp, operands[1],
- operands[2],
+ operands[2], shft,
CONST0_RTX (<VNARROWQ>mode)));
/* The intrinsic expects a narrow result, so emit a subreg that will
diff --git a/gcc/config/aarch64/predicates.md b/gcc/config/aarch64/predicates.md
index 3f5f4df8c468aae4db8f5b6af0b8400a1a96bcdb..242f10aea1f483cc2e54435701d62df36301ad39 100644
--- a/gcc/config/aarch64/predicates.md
+++ b/gcc/config/aarch64/predicates.md
@@ -587,6 +587,12 @@ (define_predicate "aarch64_simd_shift_imm_vec_di"
(and (match_code "const_vector")
(match_test "aarch64_const_vec_all_same_in_range_p (op, 1, 64)")))
+(define_predicate "aarch64_simd_rshrn_imm_vec"
+ (and (match_code "const_vector")
+ (match_test "aarch64_const_vec_all_same_in_range_p (op, 1,
+ HOST_WIDE_INT_1U
+ << (GET_MODE_UNIT_BITSIZE (mode) - 1))")))
+
(define_predicate "aarch64_simd_shift_imm_bitsize_qi"
(and (match_code "const_int")
(match_test "IN_RANGE (INTVAL (op), 0, 8)")))
^ permalink raw reply [flat|nested] only message in thread
only message in thread, other threads:[~2023-04-26 14:09 UTC | newest]
Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2023-04-26 14:09 [PATCH][committed] aarch64: Reimplement RSHRN intrinsic patterns with standard RTL codes 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).