arm: Auto-vectorization for MVE: vld2/vst2

This patch enables MVE vld2/vst2 instructions for auto-vectorization.
We move the existing expanders from neon.md and enable them for MVE,
calling the respective emitter.

2021-03-12  Christophe Lyon  <christophe.lyon@linaro.org>

	gcc/
	* config/arm/neon.md (vec_load_lanesoi<mode>)
	(vec_store_lanesoi<mode>): Move ...
	* config/arm/vec-common.md: here.

	gcc/testsuite/
	* gcc.target/arm/simd/mve-vld2.c: New test, derived from
	slp-perm-2.c

diff --git a/gcc/config/arm/neon.md b/gcc/config/arm/neon.md
index cc82d06..25d4252 100644 (file)
--- a/gcc/config/arm/neon.md
+++ b/gcc/config/arm/neon.md
@@ -5066,13 +5066,6 @@ if (BYTES_BIG_ENDIAN)
                     (const_string "neon_load2_2reg<q>")))]
 )
 
-(define_expand "vec_load_lanesoi<mode>"
-  [(set (match_operand:OI 0 "s_register_operand")
-        (unspec:OI [(match_operand:OI 1 "neon_struct_operand")
-                    (unspec:VQ2 [(const_int 0)] UNSPEC_VSTRUCTDUMMY)]
-                  UNSPEC_VLD2))]
-  "TARGET_NEON")
-
 (define_insn "neon_vld2<mode>"
   [(set (match_operand:OI 0 "s_register_operand" "=w")
         (unspec:OI [(match_operand:OI 1 "neon_struct_operand" "Um")
@@ -5200,13 +5193,6 @@ if (BYTES_BIG_ENDIAN)
                     (const_string "neon_store2_one_lane<q>")))]
 )
 
-(define_expand "vec_store_lanesoi<mode>"
-  [(set (match_operand:OI 0 "neon_struct_operand")
-       (unspec:OI [(match_operand:OI 1 "s_register_operand")
-                    (unspec:VQ2 [(const_int 0)] UNSPEC_VSTRUCTDUMMY)]
-                   UNSPEC_VST2))]
-  "TARGET_NEON")
-
 (define_insn "neon_vst2<mode>"
   [(set (match_operand:OI 0 "neon_struct_operand" "=Um")
        (unspec:OI [(match_operand:OI 1 "s_register_operand" "w")

diff --git a/gcc/config/arm/vec-common.md b/gcc/config/arm/vec-common.md
index 265fa40..0b79e68 100644 (file)
--- a/gcc/config/arm/vec-common.md
+++ b/gcc/config/arm/vec-common.md
@@ -483,6 +483,33 @@
     }
   else
     gcc_unreachable ();
+  DONE;
+})
 
+(define_expand "vec_load_lanesoi<mode>"
+  [(set (match_operand:OI 0 "s_register_operand")
+        (unspec:OI [(match_operand:OI 1 "neon_struct_operand")
+                    (unspec:VQ2 [(const_int 0)] UNSPEC_VSTRUCTDUMMY)]
+                  UNSPEC_VLD2))]
+  "TARGET_NEON || TARGET_HAVE_MVE"
+{
+  if (TARGET_NEON)
+    emit_insn (gen_neon_vld2<mode> (operands[0], operands[1]));
+  else
+    emit_insn (gen_mve_vld2q<mode> (operands[0], operands[1]));
+  DONE;
+})
+
+(define_expand "vec_store_lanesoi<mode>"
+  [(set (match_operand:OI 0 "neon_struct_operand")
+       (unspec:OI [(match_operand:OI 1 "s_register_operand")
+                    (unspec:VQ2 [(const_int 0)] UNSPEC_VSTRUCTDUMMY)]
+                   UNSPEC_VST2))]
+  "TARGET_NEON || TARGET_HAVE_MVE"
+{
+  if (TARGET_NEON)
+    emit_insn (gen_neon_vst2<mode> (operands[0], operands[1]));
+  else
+    emit_insn (gen_mve_vst2q<mode> (operands[0], operands[1]));
   DONE;
 })

diff --git a/gcc/testsuite/gcc.target/arm/simd/mve-vld2.c b/gcc/testsuite/gcc.target/arm/simd/mve-vld2.c
new file mode 100644 (file)
index 0000000..9c7c3f5
--- /dev/null
+++ b/gcc/testsuite/gcc.target/arm/simd/mve-vld2.c
@@ -0,0 +1,96 @@
+/* { dg-do assemble } */
+/* { dg-require-effective-target arm_v8_1m_mve_fp_ok } */
+/* { dg-add-options arm_v8_1m_mve_fp } */
+/* { dg-additional-options "-O3" } */
+
+#include <stdint.h>
+
+#define M00 100
+#define M10 216
+#define M01 1322
+#define M11 13
+
+#define N 128
+
+
+/* Integer tests.  */
+#define FUNC(SIGN, TYPE, BITS)                                         \
+  void foo_##SIGN##BITS##x (TYPE##BITS##_t *__restrict__ pInput,       \
+                           TYPE##BITS##_t *__restrict__ pOutput)       \
+  {                                                                    \
+    unsigned int i;                                                    \
+    TYPE##BITS##_t  a, b;                                              \
+                                                                       \
+    for (i = 0; i < N / BITS; i++)                                     \
+      {                                                                        \
+       a = *pInput++;                                                  \
+       b = *pInput++;                                                  \
+                                                                       \
+       *pOutput++ = M00 * a + M01 * b;                                 \
+       *pOutput++ = M10 * a + M11 * b;                                 \
+      }                                                                        \
+  }
+
+FUNC(s, int, 8)
+FUNC(u, uint, 8)
+FUNC(s, int, 16)
+FUNC(u, uint, 16)
+FUNC(s, int, 32)
+FUNC(u, uint, 32)
+
+/* float test, keep the macro because it's similar to the above, but does not
+   need the ##BITS##_t.  */
+#define FUNC_FLOAT(SIGN, TYPE, BITS)                                   \
+  void foo_##SIGN##BITS##x (TYPE *__restrict__ pInput,                 \
+                           TYPE *__restrict__ pOutput)                 \
+  {                                                                    \
+    unsigned int i;                                                    \
+    TYPE a, b;                                                         \
+                                                                       \
+    for (i = 0; i < N / BITS; i++)                                     \
+      {                                                                        \
+       a = *pInput++;                                                  \
+       b = *pInput++;                                                  \
+                                                                       \
+       *pOutput++ = M00 * a + M01 * b;                                 \
+       *pOutput++ = M10 * a + M11 * b;                                 \
+      }                                                                        \
+  }
+
+FUNC_FLOAT(f, float, 32)
+
+/* __fp16 test, needs explicit casts to avoid conversions to floating-point and
+   failure to vectorize.  */
+__fp16 M00_fp16 = 100.0f16;
+__fp16 M10_fp16 = 216.0f16;
+__fp16 M01_fp16 = 1322.0f16;
+__fp16 M11_fp16 = 13.0f16;
+
+#define FUNC_FLOAT_FP16(SIGN, TYPE, BITS)                              \
+  void foo_##SIGN##BITS##x (TYPE *__restrict__ pInput,                 \
+                           TYPE *__restrict__ pOutput)                 \
+  {                                                                    \
+    unsigned int i;                                                    \
+    TYPE a, b;                                                         \
+                                                                       \
+    for (i = 0; i < N / BITS; i++)                                     \
+      {                                                                        \
+       a = *pInput++;                                                  \
+       b = *pInput++;                                                  \
+                                                                       \
+       *pOutput++ = (__fp16)(M00_fp16 * a) + (__fp16)(M01_fp16 * b);   \
+       *pOutput++ = (__fp16)(M10_fp16 * a) + (__fp16)(M11_fp16 * b);   \
+      }                                                                        \
+  }
+
+FUNC_FLOAT_FP16(f, __fp16, 16)
+
+/* vld2X.8 is used for signed and unsigned chars: 2 pairs.  */
+/* vld2X.16 is used for signed and unsigned shorts and __fp16: 3 pairs.  */
+/* vld2X.32 is used for signed and unsigned ints and float: 3 pairs.  */
+/* { dg-final { scan-assembler-times {vld2[01].8\t.q[0-9]+, q[0-9]+., } 4 } } */
+/* { dg-final { scan-assembler-times {vld2[01].16\t.q[0-9]+, q[0-9]+., } 6 } } */
+/* { dg-final { scan-assembler-times {vld2[01].32\t.q[0-9]+, q[0-9]+., } 6 } } */
+/* { dg-final { scan-assembler-times {vst2[01].8\t.q[0-9]+, q[0-9]+., } 4 } } */
+/* { dg-final { scan-assembler-times {vst2[01].16\t.q[0-9]+, q[0-9]+., } 6 } } */
+/* { dg-final { scan-assembler-times {vst2[01].32\t.q[0-9]+, q[0-9]+., } 6 } } */