aarch64: Add vector/vector vec_extract patterns [PR92822]

Part of the problem in this PR is that we don't provide patterns
to extract a 64-bit vector from one half of a 128-bit vector.
Adding them fixes:

FAIL: gcc.target/aarch64/fmul_intrinsic_1.c scan-assembler-times fmul\\td[0-9]+, d[0-9]+, d[0-9]+ 1
FAIL: gcc.target/aarch64/fmul_intrinsic_1.c scan-assembler-times fmul\\tv[0-9]+.2d, v[0-9]+.2d, v[0-9]+.d\\[[0-9]+\\] 3

The 2s failures need target-independent changes, after which they rely
on these patterns too.

2020-01-27  Richard Sandiford  <richard.sandiford@arm.com>

gcc/
	PR target/92822
	* config/aarch64/aarch64-simd.md (aarch64_get_half<mode>): New
	expander.
	(@aarch64_split_simd_mov<mode>): Use it.
	(aarch64_simd_mov_from_<mode>low): Add a GPR alternative.
	Leave the vec_extract patterns to handle 2-element vectors.
	(aarch64_simd_mov_from_<mode>high): Likewise.
	(vec_extract<VQMOV_NO2E:mode><Vhalf>): New expander.
	(vec_extractv2dfv1df): Likewise.

2 files changed, 77 insertions, 22 deletions

diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index 5bbe0e3..c1ef611 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,5 +1,17 @@
 2020-01-27  Richard Sandiford  <richard.sandiford@arm.com>
 
+	PR target/92822
+	* config/aarch64/aarch64-simd.md (aarch64_get_half<mode>): New
+	expander.
+	(@aarch64_split_simd_mov<mode>): Use it.
+	(aarch64_simd_mov_from_<mode>low): Add a GPR alternative.
+	Leave the vec_extract patterns to handle 2-element vectors.
+	(aarch64_simd_mov_from_<mode>high): Likewise.
+	(vec_extract<VQMOV_NO2E:mode><Vhalf>): New expander.
+	(vec_extractv2dfv1df): Likewise.
+
+2020-01-27  Richard Sandiford  <richard.sandiford@arm.com>
+
 	* config/aarch64/aarch64.c (aarch64_if_then_else_costs): Match
 	jump conditions for *compare_condjump<GPI:mode>.
 
diff --git a/gcc/config/aarch64/aarch64-simd.md b/gcc/config/aarch64/aarch64-simd.md
index 97f46f9..5a58051 100644
--- a/gcc/config/aarch64/aarch64-simd.md
+++ b/gcc/config/aarch64/aarch64-simd.md
@@ -282,37 +282,51 @@
         rtx dst_high_part = gen_highpart (<VHALF>mode, dst);
 	rtx lo = aarch64_simd_vect_par_cnst_half (<MODE>mode, <nunits>, false);
 	rtx hi = aarch64_simd_vect_par_cnst_half (<MODE>mode, <nunits>, true);
-
-        emit_insn
-          (gen_aarch64_simd_mov_from_<mode>low (dst_low_part, src, lo));
-        emit_insn
-          (gen_aarch64_simd_mov_from_<mode>high (dst_high_part, src, hi));
+        emit_insn (gen_aarch64_get_half<mode> (dst_low_part, src, lo));
+        emit_insn (gen_aarch64_get_half<mode> (dst_high_part, src, hi));
       }
     DONE;
   }
 )
 
-(define_insn "aarch64_simd_mov_from_<mode>low"
-  [(set (match_operand:<VHALF> 0 "register_operand" "=r")
+(define_expand "aarch64_get_half<mode>"
+  [(set (match_operand:<VHALF> 0 "register_operand")
         (vec_select:<VHALF>
-          (match_operand:VQMOV 1 "register_operand" "w")
-          (match_operand:VQMOV 2 "vect_par_cnst_lo_half" "")))]
-  "TARGET_SIMD && reload_completed"
-  "umov\t%0, %1.d[0]"
-  [(set_attr "type" "neon_to_gp<q>")
-   (set_attr "length" "4")
-  ])
+          (match_operand:VQMOV 1 "register_operand")
+          (match_operand 2 "ascending_int_parallel")))]
+  "TARGET_SIMD"
+)
+
+(define_insn_and_split "aarch64_simd_mov_from_<mode>low"
+  [(set (match_operand:<VHALF> 0 "register_operand" "=w,?r")
+        (vec_select:<VHALF>
+          (match_operand:VQMOV_NO2E 1 "register_operand" "w,w")
+          (match_operand:VQMOV_NO2E 2 "vect_par_cnst_lo_half" "")))]
+  "TARGET_SIMD"
+  "@
+   #
+   umov\t%0, %1.d[0]"
+  "&& reload_completed && aarch64_simd_register (operands[0], <VHALF>mode)"
+  [(set (match_dup 0) (match_dup 1))]
+  {
+    operands[1] = aarch64_replace_reg_mode (operands[1], <VHALF>mode);
+  }
+  [(set_attr "type" "mov_reg,neon_to_gp<q>")
+   (set_attr "length" "4")]
+)
 
 (define_insn "aarch64_simd_mov_from_<mode>high"
-  [(set (match_operand:<VHALF> 0 "register_operand" "=r")
+  [(set (match_operand:<VHALF> 0 "register_operand" "=w,?r")
         (vec_select:<VHALF>
-          (match_operand:VQMOV 1 "register_operand" "w")
-          (match_operand:VQMOV 2 "vect_par_cnst_hi_half" "")))]
-  "TARGET_SIMD && reload_completed"
-  "umov\t%0, %1.d[1]"
-  [(set_attr "type" "neon_to_gp<q>")
-   (set_attr "length" "4")
-  ])
+          (match_operand:VQMOV_NO2E 1 "register_operand" "w,w")
+          (match_operand:VQMOV_NO2E 2 "vect_par_cnst_hi_half" "")))]
+  "TARGET_SIMD"
+  "@
+   dup\\t%d0, %1.d[1]
+   umov\t%0, %1.d[1]"
+  [(set_attr "type" "neon_dup<q>,neon_to_gp<q>")
+   (set_attr "length" "4")]
+)
 
 (define_insn "orn<mode>3"
  [(set (match_operand:VDQ_I 0 "register_operand" "=w")
@@ -6140,6 +6154,35 @@
     DONE;
 })
 
+;; Extract a 64-bit vector from one half of a 128-bit vector.
+(define_expand "vec_extract<mode><Vhalf>"
+  [(match_operand:<VHALF> 0 "register_operand")
+   (match_operand:VQMOV_NO2E 1 "register_operand")
+   (match_operand 2 "immediate_operand")]
+  "TARGET_SIMD"
+{
+  int start = INTVAL (operands[2]);
+  if (start != 0 && start != <nunits> / 2)
+    FAIL;
+  rtx sel = aarch64_gen_stepped_int_parallel (<nunits> / 2, start, 1);
+  emit_insn (gen_aarch64_get_half<mode> (operands[0], operands[1], sel));
+  DONE;
+})
+
+;; Extract a single-element 64-bit vector from one half of a 128-bit vector.
+(define_expand "vec_extractv2dfv1df"
+  [(match_operand:V1DF 0 "register_operand")
+   (match_operand:V2DF 1 "register_operand")
+   (match_operand 2 "immediate_operand")]
+  "TARGET_SIMD"
+{
+  /* V1DF is rarely used by other patterns, so it should be better to hide
+     it in a subreg destination of a normal DF op.  */
+  rtx scalar0 = gen_lowpart (DFmode, operands[0]);
+  emit_insn (gen_vec_extractv2dfdf (scalar0, operands[1], operands[2]));
+  DONE;
+})
+
 ;; aes
 
 (define_insn "aarch64_crypto_aes<aes_op>v16qi"