[gcc r14-1565] aarch64: Reimplement URSHR, SRSHR patterns with standard RTL codes

[gcc r14-1565] aarch64: Reimplement URSHR, SRSHR patterns with standard RTL codes

[Kyrylo Tkachov]

https://gcc.gnu.org/g:9371640999eedb8bac3fb9d1429db8a1a905b853

commit r14-1565-g9371640999eedb8bac3fb9d1429db8a1a905b853
Author: Kyrylo Tkachov <kyrylo.tkachov@arm.com>
Date:   Tue Jun 6 10:51:34 2023 +0100

    aarch64: Reimplement URSHR,SRSHR patterns with standard RTL codes
    
    Having converted the patterns for the URSRA,SRSRA instructions to standard RTL codes we can also
    easily convert the non-accumulating forms URSHR,SRSHR.
    This patch does that, reusing the various helpers and predicates from that patch in a straightforward way.
    This allows GCC to perform the optimisations in the testcase, matching what Clang does.
    
    Bootstrapped and tested on aarch64-none-linux-gnu and aarch64_be-none-elf.
    
    gcc/ChangeLog:
    
            * config/aarch64/aarch64-simd.md (aarch64_<sur>shr_n<mode>): Delete.
            (aarch64_<sra_op>rshr_n<mode><vczle><vczbe>_insn): New define_insn.
            (aarch64_<sra_op>rshr_n<mode>): New define_expand.
    
    gcc/testsuite/ChangeLog:
    
            * gcc.target/aarch64/simd/vrshr_1.c: New test.

Diff:
---
 gcc/config/aarch64/aarch64-simd.md              | 44 +++++++++++++++----
 gcc/testsuite/gcc.target/aarch64/simd/vrshr_1.c | 56 +++++++++++++++++++++++++
 2 files changed, 93 insertions(+), 7 deletions(-)

diff --git a/gcc/config/aarch64/aarch64-simd.md b/gcc/config/aarch64/aarch64-simd.md
index f7cf39f930c..dd1b084f856 100644
--- a/gcc/config/aarch64/aarch64-simd.md
+++ b/gcc/config/aarch64/aarch64-simd.md
@@ -6586,17 +6586,47 @@
 
 ;; vrshr_n
 
-(define_insn "aarch64_<sur>shr_n<mode>"
+(define_insn "aarch64_<sra_op>rshr_n<mode><vczle><vczbe>_insn"
   [(set (match_operand:VSDQ_I_DI 0 "register_operand" "=w")
-        (unspec:VSDQ_I_DI [(match_operand:VSDQ_I_DI 1 "register_operand" "w")
-			   (match_operand:SI 2
-			     "aarch64_simd_shift_imm_offset_<ve_mode>" "i")]
-			  VRSHR_N))]
-  "TARGET_SIMD"
-  "<sur>shr\\t%<v>0<Vmtype>, %<v>1<Vmtype>, %2"
+	(truncate:VSDQ_I_DI
+	  (SHIFTRT:<V2XWIDE>
+	    (plus:<V2XWIDE>
+	      (<SHIFTEXTEND>:<V2XWIDE>
+		(match_operand:VSDQ_I_DI 1 "register_operand" "w"))
+	      (match_operand:<V2XWIDE> 3 "aarch64_simd_rsra_rnd_imm_vec"))
+	    (match_operand:VSDQ_I_DI 2 "aarch64_simd_shift_imm_<vec_or_offset>_<Vel>"))))]
+  "TARGET_SIMD
+   && aarch64_const_vec_rnd_cst_p (operands[3], operands[2])"
+  "<sra_op>rshr\t%<v>0<Vmtype>, %<v>1<Vmtype>, %2"
   [(set_attr "type" "neon_sat_shift_imm<q>")]
 )
 
+(define_expand "aarch64_<sra_op>rshr_n<mode>"
+  [(match_operand:VSDQ_I_DI 0 "register_operand")
+   (SHIFTRT:VSDQ_I_DI
+     (match_operand:VSDQ_I_DI 1 "register_operand")
+     (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);
+    rtx shft = gen_int_mode (INTVAL (operands[2]), DImode);
+    rtx rnd = immed_wide_int_const (rnd_wi, GET_MODE_INNER (<V2XWIDE>mode));
+    if (VECTOR_MODE_P (<MODE>mode))
+      {
+	shft = gen_const_vec_duplicate (<MODE>mode, shft);
+	rnd = gen_const_vec_duplicate (<V2XWIDE>mode, rnd);
+      }
+
+    emit_insn (gen_aarch64_<sra_op>rshr_n<mode>_insn (operands[0], operands[1],
+						      shft, rnd));
+    DONE;
+  }
+)
+
 ;; v(r)sra_n
 
 (define_insn "aarch64_<sur>sra_ndi"
diff --git a/gcc/testsuite/gcc.target/aarch64/simd/vrshr_1.c b/gcc/testsuite/gcc.target/aarch64/simd/vrshr_1.c
new file mode 100644
index 00000000000..a5e10ff99e9
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/simd/vrshr_1.c
@@ -0,0 +1,56 @@
+/* { dg-do compile } */
+/* { dg-additional-options "-O" } */
+/* { dg-final { check-function-bodies "**" "" "" } } */
+
+#include <arm_neon.h>
+
+/*
+** foo1:
+**	srsra	v0\.16b, v1\.16b, 3
+**	ret
+*/
+
+int8x16_t
+foo1 (int8x16_t acc, int8x16_t a)
+{
+  return vaddq_s8 (acc, vrshrq_n_s8 (a, 3));
+}
+
+/*
+** foo2:
+**	srshr	v0\.16b, v1\.16b, 3
+**	ret
+*/
+
+int8x16_t
+foo2 (int8x16_t acc, int8x16_t a)
+{
+  int8x16_t z = vdupq_n_s8 (0);
+  return vrsraq_n_s8 (z, a, 3);
+}
+
+/*
+** foo3:
+**	ursra	v0\.16b, v1\.16b, 3
+**	ret
+*/
+
+uint8x16_t
+foo3 (uint8x16_t acc, uint8x16_t a)
+{
+  return vaddq_u8 (acc, vrshrq_n_u8 (a, 3));
+}
+
+/*
+** foo4:
+**	urshr	v0\.16b, v1\.16b, 3
+**	ret
+*/
+
+uint8x16_t
+foo4 (uint8x16_t acc, uint8x16_t a)
+{
+  uint8x16_t z = vdupq_n_u8 (0);
+  return vrsraq_n_u8 (z, a, 3);
+}
+