AArch64: Optimize right shift rounding narrowing

This optimizes right shift rounding narrow instructions to
rounding add narrow high where one vector is 0 when the shift amount is half
that of the original input type.

i.e.

uint32x4_t foo (uint64x2_t a, uint64x2_t b)
{
  return vrshrn_high_n_u64 (vrshrn_n_u64 (a, 32), b, 32);
}

now generates:

foo:
        movi    v3.4s, 0
        raddhn  v0.2s, v2.2d, v3.2d
        raddhn2 v0.4s, v2.2d, v3.2d

instead of:

foo:
        rshrn   v0.2s, v0.2d, 32
        rshrn2  v0.4s, v1.2d, 32
        ret

On Arm cores this is an improvement in both latency and throughput.
Because a vector zero is needed I created a new method
aarch64_gen_shareable_zero that creates zeros using V4SI and then takes a subreg
of the zero to the desired type.  This allows CSE to share all the zero
constants.

gcc/ChangeLog:

	* config/aarch64/aarch64-protos.h (aarch64_gen_shareable_zero): New.
	* config/aarch64/aarch64-simd.md (aarch64_rshrn<mode>,
	aarch64_rshrn2<mode>): Generate rounding half-ing add when appropriate.
	* config/aarch64/aarch64.c (aarch64_gen_shareable_zero): New.

gcc/testsuite/ChangeLog:

	* gcc.target/aarch64/advsimd-intrinsics/shrn-1.c: New test.
	* gcc.target/aarch64/advsimd-intrinsics/shrn-2.c: New test.
	* gcc.target/aarch64/advsimd-intrinsics/shrn-3.c: New test.
	* gcc.target/aarch64/advsimd-intrinsics/shrn-4.c: New test.

7 files changed, 106 insertions, 20 deletions

diff --git a/gcc/config/aarch64/aarch64-protos.h b/gcc/config/aarch64/aarch64-protos.h
index f7887d0..f7f5cae 100644
--- a/gcc/config/aarch64/aarch64-protos.h
+++ b/gcc/config/aarch64/aarch64-protos.h
@@ -846,6 +846,7 @@ const char *aarch64_output_move_struct (rtx *operands);
 rtx aarch64_return_addr_rtx (void);
 rtx aarch64_return_addr (int, rtx);
 rtx aarch64_simd_gen_const_vector_dup (machine_mode, HOST_WIDE_INT);
+rtx aarch64_gen_shareable_zero (machine_mode);
 bool aarch64_simd_mem_operand_p (rtx);
 bool aarch64_sve_ld1r_operand_p (rtx);
 bool aarch64_sve_ld1rq_operand_p (rtx);
diff --git a/gcc/config/aarch64/aarch64-simd.md b/gcc/config/aarch64/aarch64-simd.md
index 8e61dd9..175a9f0 100644
--- a/gcc/config/aarch64/aarch64-simd.md
+++ b/gcc/config/aarch64/aarch64-simd.md
@@ -1956,20 +1956,32 @@
    (match_operand:SI 2 "aarch64_simd_shift_imm_offset_<vn_mode>")]
   "TARGET_SIMD"
   {
-    operands[2] = aarch64_simd_gen_const_vector_dup (<MODE>mode,
-						 INTVAL (operands[2]));
-    rtx tmp = gen_reg_rtx (<VNARROWQ2>mode);
-    if (BYTES_BIG_ENDIAN)
-      emit_insn (gen_aarch64_rshrn<mode>_insn_be (tmp, operands[1],
-				operands[2], CONST0_RTX (<VNARROWQ>mode)));
+    if (INTVAL (operands[2]) == GET_MODE_UNIT_BITSIZE (<VNARROWQ>mode))
+      {
+	rtx tmp0 = aarch64_gen_shareable_zero (<MODE>mode);
+	emit_insn (gen_aarch64_raddhn<mode> (operands[0], operands[1], tmp0));
+      }
     else
-      emit_insn (gen_aarch64_rshrn<mode>_insn_le (tmp, operands[1],
-				operands[2], CONST0_RTX (<VNARROWQ>mode)));
-
-    /* The intrinsic expects a narrow result, so emit a subreg that will get
-       optimized away as appropriate.  */
-    emit_move_insn (operands[0], lowpart_subreg (<VNARROWQ>mode, tmp,
-						 <VNARROWQ2>mode));
+      {
+	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],
+						 CONST0_RTX (<VNARROWQ>mode)));
+	else
+	  emit_insn (
+		gen_aarch64_rshrn<mode>_insn_le (tmp, operands[1],
+						 operands[2],
+						 CONST0_RTX (<VNARROWQ>mode)));
+
+	/* The intrinsic expects a narrow result, so emit a subreg that will
+	   get optimized away as appropriate.  */
+	emit_move_insn (operands[0], lowpart_subreg (<VNARROWQ>mode, tmp,
+						     <VNARROWQ2>mode));
+      }
     DONE;
   }
 )
@@ -2049,14 +2061,27 @@
    (match_operand:SI 3 "aarch64_simd_shift_imm_offset_<vn_mode>")]
   "TARGET_SIMD"
   {
-    operands[3] = aarch64_simd_gen_const_vector_dup (<MODE>mode,
-						 INTVAL (operands[3]));
-    if (BYTES_BIG_ENDIAN)
-      emit_insn (gen_aarch64_rshrn2<mode>_insn_be (operands[0], operands[1],
-						  operands[2], operands[3]));
+    if (INTVAL (operands[3]) == GET_MODE_UNIT_BITSIZE (<VNARROWQ2>mode))
+      {
+	rtx tmp = aarch64_gen_shareable_zero (<MODE>mode);
+	emit_insn (gen_aarch64_raddhn2<mode> (operands[0], operands[1],
+					      operands[2], tmp));
+      }
     else
-      emit_insn (gen_aarch64_rshrn2<mode>_insn_le (operands[0], operands[1],
-						  operands[2], operands[3]));
+      {
+	operands[3] = aarch64_simd_gen_const_vector_dup (<MODE>mode,
+							 INTVAL (operands[3]));
+	if (BYTES_BIG_ENDIAN)
+	  emit_insn (gen_aarch64_rshrn2<mode>_insn_be (operands[0],
+						       operands[1],
+						       operands[2],
+						       operands[3]));
+	else
+	  emit_insn (gen_aarch64_rshrn2<mode>_insn_le (operands[0],
+						       operands[1],
+						       operands[2],
+						       operands[3]));
+      }
     DONE;
   }
 )
diff --git a/gcc/config/aarch64/aarch64.c b/gcc/config/aarch64/aarch64.c
index 7389b59..be24b73 100644
--- a/gcc/config/aarch64/aarch64.c
+++ b/gcc/config/aarch64/aarch64.c
@@ -20414,6 +20414,18 @@ aarch64_mov_operand_p (rtx x, machine_mode mode)
     == SYMBOL_TINY_ABSOLUTE;
 }
 
+/* Create a 0 constant that is based on V4SI to allow CSE to optimally share
+   the constant creation.  */
+
+rtx
+aarch64_gen_shareable_zero (machine_mode mode)
+{
+  machine_mode zmode = V4SImode;
+  rtx tmp = gen_reg_rtx (zmode);
+  emit_move_insn (tmp, CONST0_RTX (zmode));
+  return lowpart_subreg (mode, tmp, zmode);
+}
+
 /* Return a const_int vector of VAL.  */
 rtx
 aarch64_simd_gen_const_vector_dup (machine_mode mode, HOST_WIDE_INT val)
diff --git a/gcc/testsuite/gcc.target/aarch64/advsimd-intrinsics/shrn-1.c b/gcc/testsuite/gcc.target/aarch64/advsimd-intrinsics/shrn-1.c
new file mode 100644
index 0000000..4bc3aa9
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/advsimd-intrinsics/shrn-1.c
@@ -0,0 +1,15 @@
+/* { dg-do compile { target { aarch64*-*-* } } } */
+/* { dg-skip-if "" { *-*-* } { "-O0" } { "" } } */
+
+#include <arm_neon.h>
+
+uint8x16_t foo (uint32x4_t a, uint32x4_t b)
+{
+  uint16x4_t a1 = vrshrn_n_u32 (a, 16);
+  uint16x8_t b1 = vrshrn_high_n_u32 (a1, b, 16);
+  return vrshrn_high_n_u16 (vrshrn_n_u16 (b1, 8), b1, 8);
+}
+
+/* { dg-final { scan-assembler-times {\tmovi\t} 1 } } */
+/* { dg-final { scan-assembler-times {\traddhn\t} 2 } } */
+/* { dg-final { scan-assembler-times {\traddhn2\t} 2 } } */
diff --git a/gcc/testsuite/gcc.target/aarch64/advsimd-intrinsics/shrn-2.c b/gcc/testsuite/gcc.target/aarch64/advsimd-intrinsics/shrn-2.c
new file mode 100644
index 0000000..09d913e
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/advsimd-intrinsics/shrn-2.c
@@ -0,0 +1,11 @@
+/* { dg-do compile { target { aarch64*-*-* } } } */
+
+#include <arm_neon.h>
+
+uint32x4_t foo (uint64x2_t a, uint64x2_t b)
+{
+  return vrshrn_high_n_u64 (vrshrn_n_u64 (a, 32), b, 32);
+}
+
+/* { dg-final { scan-assembler-times {\traddhn\t} 1 } } */
+/* { dg-final { scan-assembler-times {\traddhn2\t} 1 } } */
diff --git a/gcc/testsuite/gcc.target/aarch64/advsimd-intrinsics/shrn-3.c b/gcc/testsuite/gcc.target/aarch64/advsimd-intrinsics/shrn-3.c
new file mode 100644
index 0000000..bdccbb3
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/advsimd-intrinsics/shrn-3.c
@@ -0,0 +1,11 @@
+/* { dg-do compile { target { aarch64*-*-* } } } */
+
+#include <arm_neon.h>
+
+uint16x8_t foo (uint32x4_t a, uint32x4_t b)
+{
+  return vrshrn_high_n_u32 (vrshrn_n_u32 (a, 16), b, 16);
+}
+
+/* { dg-final { scan-assembler-times {\traddhn\t} 1 } } */
+/* { dg-final { scan-assembler-times {\traddhn2\t} 1 } } */
diff --git a/gcc/testsuite/gcc.target/aarch64/advsimd-intrinsics/shrn-4.c b/gcc/testsuite/gcc.target/aarch64/advsimd-intrinsics/shrn-4.c
new file mode 100644
index 0000000..4b23edd
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/advsimd-intrinsics/shrn-4.c
@@ -0,0 +1,11 @@
+/* { dg-do compile { target { aarch64*-*-* } } } */
+
+#include <arm_neon.h>
+
+uint8x16_t foo (uint16x8_t a, uint16x8_t b)
+{
+  return vrshrn_high_n_u16 (vrshrn_n_u16 (a, 8), b, 8);
+}
+
+/* { dg-final { scan-assembler-times {\traddhn\t} 1 } } */
+/* { dg-final { scan-assembler-times {\traddhn2\t} 1 } } */