Re: [AArch64, 1/4] Add the missing support of vfms_n_f32, vfmsq_n_f32, vfmsq_n_f64

Wed, 18 May 2016 01:33:01 -0700 


On 18/05/16 09:25, Christophe Lyon wrote:

On 17 May 2016 at 14:42, Kyrill Tkachov <kyrylo.tkac...@foss.arm.com> wrote:

On 17/05/16 13:40, Kyrill Tkachov wrote:


On 17/05/16 13:20, James Greenhalgh wrote:

On Mon, May 16, 2016 at 10:09:26AM +0100, Jiong Wang wrote:

The support of vfma_n_f64, vfms_n_f32, vfmsq_n_f32, vfmsq_n_f64 are
missing in current gcc arm_neon.h.

Meanwhile, besides "(fma (vec_dup (vec_select)))", fma by element can
also comes from "(fma (vec_dup(scalar" where the scalar value is already
sitting in vector register then duplicated to other lanes, and there is
no lane size change.

This patch implement this and can generate better code under some
context. For example:

cat test.c
===
typedef __Float32x2_t float32x2_t;
typedef float float32_t;

float32x2_t
vfma_n_f32 (float32x2_t __a, float32x2_t __b, float32_t __c)
{
        return __builtin_aarch64_fmav2sf (__b,  (float32x2_t) {__c,
__c}, __a);
}

before (-O2)
===
vfma_n_f32:
          dup     v2.2s, v2.s[0]
          fmla    v0.2s, v1.2s, v2.2s
          ret
after
===
vfma_n_f32:
          fmla    v0.2s, v1.2s, v2.s[0]
          ret

OK for trunk?

2016-05-16  Jiong Wang <jiong.w...@arm.com>

This ChangeLog entry is not correctly formatted. There should be two
spaces between your name and your email, and each line should start with
a tab.


gcc/
    * config/aarch64/aarch64-simd.md (*aarch64_fma4_elt_to_128df): Rename
    to *aarch64_fma4_elt_from_dup<mode>.
    (*aarch64_fnma4_elt_to_128df): Rename to
*aarch64_fnma4_elt_from_dup<mode>.
    * config/aarch64/arm_neon.h (vfma_n_f64): New.
    (vfms_n_f32): Likewise.
    (vfms_n_f64): Likewise.
    (vfmsq_n_f32): Likewise.
    (vfmsq_n_f64): Likewise.

gcc/testsuite/
    * gcc/testsuite/gcc.target/aarch64/fmla_intrinsic_1.c: Use
standard syntax.
    * gcc/testsuite/gcc.target/aarch64/fmls_intrinsic_1.c: Likewise.

The paths of these two entries are incorrect. Remove the gcc/testsuite
from the front. I don't understand what you mean by "Use standard
syntax.",
please fix this to describe what you are actually changing.


    * gcc.target/aarch64/advsimd-intrinsics/arm-neon-ref.h: New entry
for float64x1.
    * gcc.target/aarch64/advsimd-intrinsics/vfms_vfma_n.c: New.

These two changes need approval from an ARM maintainer as they are in
common files.

  From an AArch64 perspective, this patch is OK with a fixed ChangeLog.
Please
wait for an ARM OK for the test changes.


Considering that the tests' functionality is guarded on #if
defined(__aarch64__)
it's a noop on arm and so is ok from that perspective (we have precedence
for
tests guarded in such a way in advsimd-intrinsics.exp)


Of course I meant precedent rather than precedence :/


Unfortunately, the guard is not correct :(

The float64_t type is not available on arm, so the new
declarations/definitions in arm-neon-ref.h
need a guard.


Indeed.
A patch to guard the float64_t definition in arm-neon-ref.h on __aarch64__
is preapproved (or I think would count as obvious in the interest of unbreaking
a large number of tests)

Sorry for not catching it earlier.

Kyrill


Since this patch was checked-in, all the advsimd intrinsics tests fail
to compile
on arm:
In file included from
/aci-gcc-fsf/sources/gcc-fsf/gccsrc/gcc/testsuite/gcc.target/aarch64/advsimd-intrinsics/vaba.c:2:0:
/aci-gcc-fsf/sources/gcc-fsf/gccsrc/gcc/testsuite/gcc.target/aarch64/advsimd-intrinsics/arm-neon-ref.h:139:22:
error: unknown type name 'float64_t'
/aci-gcc-fsf/sources/gcc-fsf/gccsrc/gcc/testsuite/gcc.target/aarch64/advsimd-intrinsics/arm-neon-ref.h:51:35:
note: in definition of macro 'VECT_VAR_DECL'
/aci-gcc-fsf/sources/gcc-fsf/gccsrc/gcc/testsuite/gcc.target/aarch64/advsimd-intrinsics/arm-neon-ref.h:139:8:
note: in expansion of macro 'ARRAY'


Christophe



Kyrill



The arm-neon-ref.h additions are ok too.

Thanks,
Kyrill



Thanks,
James


diff --git a/gcc/config/aarch64/aarch64-simd.md
b/gcc/config/aarch64/aarch64-simd.md
index
bd73bce64414e8bc01732d14311d742cf28f4586..90eaca176b4706e6cc42f16ce2c956f1c8ad17b1
100644
--- a/gcc/config/aarch64/aarch64-simd.md
+++ b/gcc/config/aarch64/aarch64-simd.md
@@ -1579,16 +1579,16 @@
     [(set_attr "type" "neon_fp_mla_<Vetype>_scalar<q>")]
   )
   -(define_insn "*aarch64_fma4_elt_to_128df"
-  [(set (match_operand:V2DF 0 "register_operand" "=w")
-    (fma:V2DF
-      (vec_duplicate:V2DF
-      (match_operand:DF 1 "register_operand" "w"))
-      (match_operand:V2DF 2 "register_operand" "w")
-      (match_operand:V2DF 3 "register_operand" "0")))]
+(define_insn "*aarch64_fma4_elt_from_dup<mode>"
+  [(set (match_operand:VMUL 0 "register_operand" "=w")
+    (fma:VMUL
+      (vec_duplicate:VMUL
+      (match_operand:<VEL> 1 "register_operand" "w"))
+      (match_operand:VMUL 2 "register_operand" "w")
+      (match_operand:VMUL 3 "register_operand" "0")))]
     "TARGET_SIMD"
-  "fmla\\t%0.2d, %2.2d, %1.2d[0]"
-  [(set_attr "type" "neon_fp_mla_d_scalar_q")]
+  "fmla\t%0.<Vtype>, %2.<Vtype>, %1.<Vetype>[0]"
+  [(set_attr "type" "neon<fp>_mla_<Vetype>_scalar<q>")]
   )
     (define_insn "*aarch64_fma4_elt_to_64v2df"
@@ -1656,17 +1656,17 @@
     [(set_attr "type" "neon_fp_mla_<Vetype>_scalar<q>")]
   )
   -(define_insn "*aarch64_fnma4_elt_to_128df"
-  [(set (match_operand:V2DF 0 "register_operand" "=w")
-    (fma:V2DF
-      (neg:V2DF
-        (match_operand:V2DF 2 "register_operand" "w"))
-      (vec_duplicate:V2DF
-    (match_operand:DF 1 "register_operand" "w"))
-      (match_operand:V2DF 3 "register_operand" "0")))]
-  "TARGET_SIMD"
-  "fmls\\t%0.2d, %2.2d, %1.2d[0]"
-  [(set_attr "type" "neon_fp_mla_d_scalar_q")]
+(define_insn "*aarch64_fnma4_elt_from_dup<mode>"
+  [(set (match_operand:VMUL 0 "register_operand" "=w")
+    (fma:VMUL
+      (neg:VMUL
+        (match_operand:VMUL 2 "register_operand" "w"))
+      (vec_duplicate:VMUL
+    (match_operand:<VEL> 1 "register_operand" "w"))
+      (match_operand:VMUL 3 "register_operand" "0")))]
+  "TARGET_SIMD"
+  "fmls\t%0.<Vtype>, %2.<Vtype>, %1.<Vetype>[0]"
+  [(set_attr "type" "neon<fp>_mla_<Vetype>_scalar<q>")]
   )
     (define_insn "*aarch64_fnma4_elt_to_64v2df"
diff --git a/gcc/config/aarch64/arm_neon.h
b/gcc/config/aarch64/arm_neon.h
index
2612a325718918cf7cd808f28c09c9c4c7b11c07..ca7ace5aa656163826569d046fcbf02f9f7d4d6c
100644
--- a/gcc/config/aarch64/arm_neon.h
+++ b/gcc/config/aarch64/arm_neon.h
@@ -14456,6 +14456,12 @@ vfma_n_f32 (float32x2_t __a, float32x2_t __b,
float32_t __c)
     return __builtin_aarch64_fmav2sf (__b, vdup_n_f32 (__c), __a);
   }
   +__extension__ static __inline float64x1_t __attribute__
((__always_inline__))
+vfma_n_f64 (float64x1_t __a, float64x1_t __b, float64_t __c)
+{
+  return (float64x1_t) {__b[0] * __c + __a[0]};
+}
+
   __extension__ static __inline float32x4_t __attribute__
((__always_inline__))
   vfmaq_n_f32 (float32x4_t __a, float32x4_t __b, float32_t __c)
   {
@@ -14597,6 +14603,29 @@ vfmsq_f64 (float64x2_t __a, float64x2_t __b,
float64x2_t __c)
     return __builtin_aarch64_fmav2df (-__b, __c, __a);
   }
   +__extension__ static __inline float32x2_t __attribute__
((__always_inline__))
+vfms_n_f32 (float32x2_t __a, float32x2_t __b, float32_t __c)
+{
+  return __builtin_aarch64_fmav2sf (-__b, vdup_n_f32 (__c), __a);
+}
+
+__extension__ static __inline float64x1_t __attribute__
((__always_inline__))
+vfms_n_f64 (float64x1_t __a, float64x1_t __b, float64_t __c)
+{
+  return (float64x1_t) {-__b[0] * __c + __a[0]};
+}
+
+__extension__ static __inline float32x4_t __attribute__
((__always_inline__))
+vfmsq_n_f32 (float32x4_t __a, float32x4_t __b, float32_t __c)
+{
+  return __builtin_aarch64_fmav4sf (-__b, vdupq_n_f32 (__c), __a);
+}
+
+__extension__ static __inline float64x2_t __attribute__
((__always_inline__))
+vfmsq_n_f64 (float64x2_t __a, float64x2_t __b, float64_t __c)
+{
+  return __builtin_aarch64_fmav2df (-__b, vdupq_n_f64 (__c), __a);
+}
     /* vfms_lane  */
   diff --git
a/gcc/testsuite/gcc.target/aarch64/advsimd-intrinsics/arm-neon-ref.h
b/gcc/testsuite/gcc.target/aarch64/advsimd-intrinsics/arm-neon-ref.h
index
49fbd843e507ede8aa81d02c175a82a1221750a4..cf90825f87391b72aca9a29980210d21f4321c04
100644
--- a/gcc/testsuite/gcc.target/aarch64/advsimd-intrinsics/arm-neon-ref.h
+++ b/gcc/testsuite/gcc.target/aarch64/advsimd-intrinsics/arm-neon-ref.h
@@ -136,6 +136,7 @@ static ARRAY(result, poly, 16, 4);
   #if defined (__ARM_FP16_FORMAT_IEEE) || defined
(__ARM_FP16_FORMAT_ALTERNATIVE)
   static ARRAY(result, float, 16, 4);
   #endif
+static ARRAY(result, float, 64, 1);
   static ARRAY(result, float, 32, 2);
   static ARRAY(result, int, 8, 16);
   static ARRAY(result, int, 16, 8);
@@ -169,6 +170,7 @@ extern ARRAY(expected, poly, 8, 8);
   extern ARRAY(expected, poly, 16, 4);
   extern ARRAY(expected, hfloat, 16, 4);
   extern ARRAY(expected, hfloat, 32, 2);
+extern ARRAY(expected, hfloat, 64, 1);
   extern ARRAY(expected, int, 8, 16);
   extern ARRAY(expected, int, 16, 8);
   extern ARRAY(expected, int, 32, 4);
diff --git
a/gcc/testsuite/gcc.target/aarch64/advsimd-intrinsics/vfms_vfma_n.c
b/gcc/testsuite/gcc.target/aarch64/advsimd-intrinsics/vfms_vfma_n.c
new file mode 100644
index
0000000000000000000000000000000000000000..26223763c59c849607b5320f6ec37098a556ce2e
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/advsimd-intrinsics/vfms_vfma_n.c
@@ -0,0 +1,490 @@
+#include <arm_neon.h>
+#include "arm-neon-ref.h"
+#include "compute-ref-data.h"
+
+#define A0 123.4f
+#define A1 -3.8f
+#define A2 -29.4f
+#define A3 (__builtin_inff ())
+#define A4 0.0f
+#define A5 24.0f
+#define A6 124.0f
+#define A7 1024.0f
+
+#define B0 -5.8f
+#define B1 -0.0f
+#define B2 -10.8f
+#define B3 10.0f
+#define B4 23.4f
+#define B5 -1234.8f
+#define B6 8.9f
+#define B7 4.0f
+
+#define E0 9.8f
+#define E1 -1024.0f
+#define E2 (-__builtin_inff ())
+#define E3 479.0f
+float32_t elem0 = E0;
+float32_t elem1 = E1;
+float32_t elem2 = E2;
+float32_t elem3 = E3;
+
+#define DA0 1231234.4
+#define DA1 -3.8
+#define DA2 -2980.4
+#define DA3 -5.8
+#define DA4 0.01123
+#define DA5 24.0
+#define DA6 124.12345
+#define DA7 1024.0
+
+#define DB0 -5.8
+#define DB1 (__builtin_inf ())
+#define DB2 -105.8
+#define DB3 10.0
+#define DB4 (-__builtin_inf ())
+#define DB5 -1234.8
+#define DB6 848.9
+#define DB7 44444.0
+
+#define DE0 9.8
+#define DE1 -1024.0
+#define DE2 105.8
+#define DE3 479.0
+float64_t delem0 = DE0;
+float64_t delem1 = DE1;
+float64_t delem2 = DE2;
+float64_t delem3 = DE3;
+
+#if defined(__aarch64__) && defined(__ARM_FEATURE_FMA)
+
+/* Expected results for vfms_n.  */
+
+VECT_VAR_DECL(expectedfms0, float, 32, 2) [] = {A0 + -B0 * E0, A1 + -B1
* E0};
+VECT_VAR_DECL(expectedfms1, float, 32, 2) [] = {A2 + -B2 * E1, A3 + -B3
* E1};
+VECT_VAR_DECL(expectedfms2, float, 32, 2) [] = {A4 + -B4 * E2, A5 + -B5
* E2};
+VECT_VAR_DECL(expectedfms3, float, 32, 2) [] = {A6 + -B6 * E3, A7 + -B7
* E3};
+VECT_VAR_DECL(expectedfma0, float, 32, 2) [] = {A0 + B0 * E0, A1 + B1 *
E0};
+VECT_VAR_DECL(expectedfma1, float, 32, 2) [] = {A2 + B2 * E1, A3 + B3 *
E1};
+VECT_VAR_DECL(expectedfma2, float, 32, 2) [] = {A4 + B4 * E2, A5 + B5 *
E2};
+VECT_VAR_DECL(expectedfma3, float, 32, 2) [] = {A6 + B6 * E3, A7 + B7 *
E3};
+
+hfloat32_t * VECT_VAR (expectedfms0_static, hfloat, 32, 2) =
+  (hfloat32_t *) VECT_VAR (expectedfms0, float, 32, 2);
+hfloat32_t * VECT_VAR (expectedfms1_static, hfloat, 32, 2) =
+  (hfloat32_t *) VECT_VAR (expectedfms1, float, 32, 2);
+hfloat32_t * VECT_VAR (expectedfms2_static, hfloat, 32, 2) =
+  (hfloat32_t *) VECT_VAR (expectedfms2, float, 32, 2);
+hfloat32_t * VECT_VAR (expectedfms3_static, hfloat, 32, 2) =
+  (hfloat32_t *) VECT_VAR (expectedfms3, float, 32, 2);
+hfloat32_t * VECT_VAR (expectedfma0_static, hfloat, 32, 2) =
+  (hfloat32_t *) VECT_VAR (expectedfma0, float, 32, 2);
+hfloat32_t * VECT_VAR (expectedfma1_static, hfloat, 32, 2) =
+  (hfloat32_t *) VECT_VAR (expectedfma1, float, 32, 2);
+hfloat32_t * VECT_VAR (expectedfma2_static, hfloat, 32, 2) =
+  (hfloat32_t *) VECT_VAR (expectedfma2, float, 32, 2);
+hfloat32_t * VECT_VAR (expectedfma3_static, hfloat, 32, 2) =
+  (hfloat32_t *) VECT_VAR (expectedfma3, float, 32, 2);
+
+
+VECT_VAR_DECL(expectedfms0, float, 32, 4) [] = {A0 + -B0 * E0, A1 + -B1
* E0,
+                        A2 + -B2 * E0, A3 + -B3 * E0};
+VECT_VAR_DECL(expectedfms1, float, 32, 4) [] = {A4 + -B4 * E1, A5 + -B5
* E1,
+                        A6 + -B6 * E1, A7 + -B7 * E1};
+VECT_VAR_DECL(expectedfms2, float, 32, 4) [] = {A0 + -B0 * E2, A2 + -B2
* E2,
+                        A4 + -B4 * E2, A6 + -B6 * E2};
+VECT_VAR_DECL(expectedfms3, float, 32, 4) [] = {A1 + -B1 * E3, A3 + -B3
* E3,
+                        A5 + -B5 * E3, A7 + -B7 * E3};
+VECT_VAR_DECL(expectedfma0, float, 32, 4) [] = {A0 + B0 * E0, A1 + B1 *
E0,
+                        A2 + B2 * E0, A3 + B3 * E0};
+VECT_VAR_DECL(expectedfma1, float, 32, 4) [] = {A4 + B4 * E1, A5 + B5 *
E1,
+                        A6 + B6 * E1, A7 + B7 * E1};
+VECT_VAR_DECL(expectedfma2, float, 32, 4) [] = {A0 + B0 * E2, A2 + B2 *
E2,
+                        A4 + B4 * E2, A6 + B6 * E2};
+VECT_VAR_DECL(expectedfma3, float, 32, 4) [] = {A1 + B1 * E3, A3 + B3 *
E3,
+                        A5 + B5 * E3, A7 + B7 * E3};
+
+hfloat32_t * VECT_VAR (expectedfms0_static, hfloat, 32, 4) =
+  (hfloat32_t *) VECT_VAR (expectedfms0, float, 32, 4);
+hfloat32_t * VECT_VAR (expectedfms1_static, hfloat, 32, 4) =
+  (hfloat32_t *) VECT_VAR (expectedfms1, float, 32, 4);
+hfloat32_t * VECT_VAR (expectedfms2_static, hfloat, 32, 4) =
+  (hfloat32_t *) VECT_VAR (expectedfms2, float, 32, 4);
+hfloat32_t * VECT_VAR (expectedfms3_static, hfloat, 32, 4) =
+  (hfloat32_t *) VECT_VAR (expectedfms3, float, 32, 4);
+hfloat32_t * VECT_VAR (expectedfma0_static, hfloat, 32, 4) =
+  (hfloat32_t *) VECT_VAR (expectedfma0, float, 32, 4);
+hfloat32_t * VECT_VAR (expectedfma1_static, hfloat, 32, 4) =
+  (hfloat32_t *) VECT_VAR (expectedfma1, float, 32, 4);
+hfloat32_t * VECT_VAR (expectedfma2_static, hfloat, 32, 4) =
+  (hfloat32_t *) VECT_VAR (expectedfma2, float, 32, 4);
+hfloat32_t * VECT_VAR (expectedfma3_static, hfloat, 32, 4) =
+  (hfloat32_t *) VECT_VAR (expectedfma3, float, 32, 4);
+
+VECT_VAR_DECL(expectedfms0, float, 64, 2) [] = {DA0 + -DB0 * DE0,
+                        DA1 + -DB1 * DE0};
+VECT_VAR_DECL(expectedfms1, float, 64, 2) [] = {DA2 + -DB2 * DE1,
+                        DA3 + -DB3 * DE1};
+VECT_VAR_DECL(expectedfms2, float, 64, 2) [] = {DA4 + -DB4 * DE2,
+                        DA5 + -DB5 * DE2};
+VECT_VAR_DECL(expectedfms3, float, 64, 2) [] = {DA6 + -DB6 * DE3,
+                        DA7 + -DB7 * DE3};
+VECT_VAR_DECL(expectedfma0, float, 64, 2) [] = {DA0 + DB0 * DE0,
+                        DA1 + DB1 * DE0};
+VECT_VAR_DECL(expectedfma1, float, 64, 2) [] = {DA2 + DB2 * DE1,
+                        DA3 + DB3 * DE1};
+VECT_VAR_DECL(expectedfma2, float, 64, 2) [] = {DA4 + DB4 * DE2,
+                        DA5 + DB5 * DE2};
+VECT_VAR_DECL(expectedfma3, float, 64, 2) [] = {DA6 + DB6 * DE3,
+                        DA7 + DB7 * DE3};
+hfloat64_t * VECT_VAR (expectedfms0_static, hfloat, 64, 2) =
+  (hfloat64_t *) VECT_VAR (expectedfms0, float, 64, 2);
+hfloat64_t * VECT_VAR (expectedfms1_static, hfloat, 64, 2) =
+  (hfloat64_t *) VECT_VAR (expectedfms1, float, 64, 2);
+hfloat64_t * VECT_VAR (expectedfms2_static, hfloat, 64, 2) =
+  (hfloat64_t *) VECT_VAR (expectedfms2, float, 64, 2);
+hfloat64_t * VECT_VAR (expectedfms3_static, hfloat, 64, 2) =
+  (hfloat64_t *) VECT_VAR (expectedfms3, float, 64, 2);
+hfloat64_t * VECT_VAR (expectedfma0_static, hfloat, 64, 2) =
+  (hfloat64_t *) VECT_VAR (expectedfma0, float, 64, 2);
+hfloat64_t * VECT_VAR (expectedfma1_static, hfloat, 64, 2) =
+  (hfloat64_t *) VECT_VAR (expectedfma1, float, 64, 2);
+hfloat64_t * VECT_VAR (expectedfma2_static, hfloat, 64, 2) =
+  (hfloat64_t *) VECT_VAR (expectedfma2, float, 64, 2);
+hfloat64_t * VECT_VAR (expectedfma3_static, hfloat, 64, 2) =
+  (hfloat64_t *) VECT_VAR (expectedfma3, float, 64, 2);
+
+VECT_VAR_DECL(expectedfms0, float, 64, 1) [] = {DA0 + -DB0 * DE0};
+VECT_VAR_DECL(expectedfms1, float, 64, 1) [] = {DA2 + -DB2 * DE1};
+VECT_VAR_DECL(expectedfms2, float, 64, 1) [] = {DA4 + -DB4 * DE2};
+VECT_VAR_DECL(expectedfms3, float, 64, 1) [] = {DA6 + -DB6 * DE3};
+VECT_VAR_DECL(expectedfma0, float, 64, 1) [] = {DA0 + DB0 * DE0};
+VECT_VAR_DECL(expectedfma1, float, 64, 1) [] = {DA2 + DB2 * DE1};
+VECT_VAR_DECL(expectedfma2, float, 64, 1) [] = {DA4 + DB4 * DE2};
+VECT_VAR_DECL(expectedfma3, float, 64, 1) [] = {DA6 + DB6 * DE3};
+
+hfloat64_t * VECT_VAR (expectedfms0_static, hfloat, 64, 1) =
+  (hfloat64_t *) VECT_VAR (expectedfms0, float, 64, 1);
+hfloat64_t * VECT_VAR (expectedfms1_static, hfloat, 64, 1) =
+  (hfloat64_t *) VECT_VAR (expectedfms1, float, 64, 1);
+hfloat64_t * VECT_VAR (expectedfms2_static, hfloat, 64, 1) =
+  (hfloat64_t *) VECT_VAR (expectedfms2, float, 64, 1);
+hfloat64_t * VECT_VAR (expectedfms3_static, hfloat, 64, 1) =
+  (hfloat64_t *) VECT_VAR (expectedfms3, float, 64, 1);
+hfloat64_t * VECT_VAR (expectedfma0_static, hfloat, 64, 1) =
+  (hfloat64_t *) VECT_VAR (expectedfma0, float, 64, 1);
+hfloat64_t * VECT_VAR (expectedfma1_static, hfloat, 64, 1) =
+  (hfloat64_t *) VECT_VAR (expectedfma1, float, 64, 1);
+hfloat64_t * VECT_VAR (expectedfma2_static, hfloat, 64, 1) =
+  (hfloat64_t *) VECT_VAR (expectedfma2, float, 64, 1);
+hfloat64_t * VECT_VAR (expectedfma3_static, hfloat, 64, 1) =
+  (hfloat64_t *) VECT_VAR (expectedfma3, float, 64, 1);
+
+void exec_vfma_vfms_n (void)
+{
+#undef TEST_MSG
+#define TEST_MSG "VFMS_VFMA_N (FP32)"
+  clean_results ();
+
+  DECL_VARIABLE(vsrc_1, float, 32, 2);
+  DECL_VARIABLE(vsrc_2, float, 32, 2);
+  VECT_VAR_DECL (buf_src_1, float, 32, 2) [] = {A0, A1};
+  VECT_VAR_DECL (buf_src_2, float, 32, 2) [] = {B0, B1};
+  VLOAD (vsrc_1, buf_src_1, , float, f, 32, 2);
+  VLOAD (vsrc_2, buf_src_2, , float, f, 32, 2);
+  DECL_VARIABLE (vector_res, float, 32, 2) =
+    vfms_n_f32 (VECT_VAR (vsrc_1, float, 32, 2),
+        VECT_VAR (vsrc_2, float, 32, 2), elem0);
+  vst1_f32 (VECT_VAR (result, float, 32, 2),
+        VECT_VAR (vector_res, float, 32, 2));
+  CHECK_FP (TEST_MSG, float, 32, 2, PRIx16, expectedfms0_static, "");
+  VECT_VAR (vector_res, float, 32, 2) =
+    vfma_n_f32 (VECT_VAR (vsrc_1, float, 32, 2),
+        VECT_VAR (vsrc_2, float, 32, 2), elem0);
+  vst1_f32 (VECT_VAR (result, float, 32, 2),
+        VECT_VAR (vector_res, float, 32, 2));
+  CHECK_FP (TEST_MSG, float, 32, 2, PRIx16, expectedfma0_static, "");
+
+  VECT_VAR_DECL (buf_src_3, float, 32, 2) [] = {A2, A3};
+  VECT_VAR_DECL (buf_src_4, float, 32, 2) [] = {B2, B3};
+  VLOAD (vsrc_1, buf_src_3, , float, f, 32, 2);
+  VLOAD (vsrc_2, buf_src_4, , float, f, 32, 2);
+  VECT_VAR (vector_res, float, 32, 2) =
+    vfms_n_f32 (VECT_VAR (vsrc_1, float, 32, 2),
+        VECT_VAR (vsrc_2, float, 32, 2), elem1);
+  vst1_f32 (VECT_VAR (result, float, 32, 2),
+        VECT_VAR (vector_res, float, 32, 2));
+  CHECK_FP (TEST_MSG, float, 32, 2, PRIx16, expectedfms1_static, "");
+  VECT_VAR (vector_res, float, 32, 2) =
+    vfma_n_f32 (VECT_VAR (vsrc_1, float, 32, 2),
+        VECT_VAR (vsrc_2, float, 32, 2), elem1);
+  vst1_f32 (VECT_VAR (result, float, 32, 2),
+        VECT_VAR (vector_res, float, 32, 2));
+  CHECK_FP (TEST_MSG, float, 32, 2, PRIx16, expectedfma1_static, "");
+
+  VECT_VAR_DECL (buf_src_5, float, 32, 2) [] = {A4, A5};
+  VECT_VAR_DECL (buf_src_6, float, 32, 2) [] = {B4, B5};
+  VLOAD (vsrc_1, buf_src_5, , float, f, 32, 2);
+  VLOAD (vsrc_2, buf_src_6, , float, f, 32, 2);
+  VECT_VAR (vector_res, float, 32, 2) =
+    vfms_n_f32 (VECT_VAR (vsrc_1, float, 32, 2),
+        VECT_VAR (vsrc_2, float, 32, 2), elem2);
+  vst1_f32 (VECT_VAR (result, float, 32, 2),
+        VECT_VAR (vector_res, float, 32, 2));
+  CHECK_FP (TEST_MSG, float, 32, 2, PRIx16, expectedfms2_static, "");
+  VECT_VAR (vector_res, float, 32, 2) =
+    vfma_n_f32 (VECT_VAR (vsrc_1, float, 32, 2),
+        VECT_VAR (vsrc_2, float, 32, 2), elem2);
+  vst1_f32 (VECT_VAR (result, float, 32, 2),
+        VECT_VAR (vector_res, float, 32, 2));
+  CHECK_FP (TEST_MSG, float, 32, 2, PRIx16, expectedfma2_static, "");
+
+  VECT_VAR_DECL (buf_src_7, float, 32, 2) [] = {A6, A7};
+  VECT_VAR_DECL (buf_src_8, float, 32, 2) [] = {B6, B7};
+  VLOAD (vsrc_1, buf_src_7, , float, f, 32, 2);
+  VLOAD (vsrc_2, buf_src_8, , float, f, 32, 2);
+  VECT_VAR (vector_res, float, 32, 2) =
+    vfms_n_f32 (VECT_VAR (vsrc_1, float, 32, 2),
+        VECT_VAR (vsrc_2, float, 32, 2), elem3);
+  vst1_f32 (VECT_VAR (result, float, 32, 2),
+        VECT_VAR (vector_res, float, 32, 2));
+  CHECK_FP (TEST_MSG, float, 32, 2, PRIx16, expectedfms3_static, "");
+  VECT_VAR (vector_res, float, 32, 2) =
+    vfma_n_f32 (VECT_VAR (vsrc_1, float, 32, 2),
+        VECT_VAR (vsrc_2, float, 32, 2), elem3);
+  vst1_f32 (VECT_VAR (result, float, 32, 2),
+        VECT_VAR (vector_res, float, 32, 2));
+  CHECK_FP (TEST_MSG, float, 32, 2, PRIx16, expectedfma3_static, "");
+
+#undef TEST_MSG
+#define TEST_MSG "VFMSQ_VFMAQ_N (FP32)"
+  clean_results ();
+
+  DECL_VARIABLE(vsrc_1, float, 32, 4);
+  DECL_VARIABLE(vsrc_2, float, 32, 4);
+  VECT_VAR_DECL (buf_src_1, float, 32, 4) [] = {A0, A1, A2, A3};
+  VECT_VAR_DECL (buf_src_2, float, 32, 4) [] = {B0, B1, B2, B3};
+  VLOAD (vsrc_1, buf_src_1, q, float, f, 32, 4);
+  VLOAD (vsrc_2, buf_src_2, q, float, f, 32, 4);
+  DECL_VARIABLE (vector_res, float, 32, 4) =
+    vfmsq_n_f32 (VECT_VAR (vsrc_1, float, 32, 4),
+         VECT_VAR (vsrc_2, float, 32, 4), elem0);
+  vst1q_f32 (VECT_VAR (result, float, 32, 4),
+         VECT_VAR (vector_res, float, 32, 4));
+  CHECK_FP (TEST_MSG, float, 32, 4, PRIx16, expectedfms0_static, "");
+  VECT_VAR (vector_res, float, 32, 4) =
+    vfmaq_n_f32 (VECT_VAR (vsrc_1, float, 32, 4),
+         VECT_VAR (vsrc_2, float, 32, 4), elem0);
+  vst1q_f32 (VECT_VAR (result, float, 32, 4),
+         VECT_VAR (vector_res, float, 32, 4));
+  CHECK_FP (TEST_MSG, float, 32, 4, PRIx16, expectedfma0_static, "");
+
+  VECT_VAR_DECL (buf_src_3, float, 32, 4) [] = {A4, A5, A6, A7};
+  VECT_VAR_DECL (buf_src_4, float, 32, 4) [] = {B4, B5, B6, B7};
+  VLOAD (vsrc_1, buf_src_3, q, float, f, 32, 4);
+  VLOAD (vsrc_2, buf_src_4, q, float, f, 32, 4);
+  VECT_VAR (vector_res, float, 32, 4) =
+    vfmsq_n_f32 (VECT_VAR (vsrc_1, float, 32, 4),
+         VECT_VAR (vsrc_2, float, 32, 4), elem1);
+  vst1q_f32 (VECT_VAR (result, float, 32, 4),
+         VECT_VAR (vector_res, float, 32, 4));
+  CHECK_FP (TEST_MSG, float, 32, 4, PRIx16, expectedfms1_static, "");
+  VECT_VAR (vector_res, float, 32, 4) =
+    vfmaq_n_f32 (VECT_VAR (vsrc_1, float, 32, 4),
+         VECT_VAR (vsrc_2, float, 32, 4), elem1);
+  vst1q_f32 (VECT_VAR (result, float, 32, 4),
+         VECT_VAR (vector_res, float, 32, 4));
+  CHECK_FP (TEST_MSG, float, 32, 4, PRIx16, expectedfma1_static, "");
+
+  VECT_VAR_DECL (buf_src_5, float, 32, 4) [] = {A0, A2, A4, A6};
+  VECT_VAR_DECL (buf_src_6, float, 32, 4) [] = {B0, B2, B4, B6};
+  VLOAD (vsrc_1, buf_src_5, q, float, f, 32, 4);
+  VLOAD (vsrc_2, buf_src_6, q, float, f, 32, 4);
+  VECT_VAR (vector_res, float, 32, 4) =
+    vfmsq_n_f32 (VECT_VAR (vsrc_1, float, 32, 4),
+         VECT_VAR (vsrc_2, float, 32, 4), elem2);
+  vst1q_f32 (VECT_VAR (result, float, 32, 4),
+         VECT_VAR (vector_res, float, 32, 4));
+  CHECK_FP (TEST_MSG, float, 32, 4, PRIx16, expectedfms2_static, "");
+  VECT_VAR (vector_res, float, 32, 4) =
+    vfmaq_n_f32 (VECT_VAR (vsrc_1, float, 32, 4),
+         VECT_VAR (vsrc_2, float, 32, 4), elem2);
+  vst1q_f32 (VECT_VAR (result, float, 32, 4),
+         VECT_VAR (vector_res, float, 32, 4));
+  CHECK_FP (TEST_MSG, float, 32, 4, PRIx16, expectedfma2_static, "");
+
+  VECT_VAR_DECL (buf_src_7, float, 32, 4) [] = {A1, A3, A5, A7};
+  VECT_VAR_DECL (buf_src_8, float, 32, 4) [] = {B1, B3, B5, B7};
+  VLOAD (vsrc_1, buf_src_7, q, float, f, 32, 4);
+  VLOAD (vsrc_2, buf_src_8, q, float, f, 32, 4);
+  VECT_VAR (vector_res, float, 32, 4) =
+    vfmsq_n_f32 (VECT_VAR (vsrc_1, float, 32, 4),
+         VECT_VAR (vsrc_2, float, 32, 4), elem3);
+  vst1q_f32 (VECT_VAR (result, float, 32, 4),
+         VECT_VAR (vector_res, float, 32, 4));
+  CHECK_FP (TEST_MSG, float, 32, 4, PRIx16, expectedfms3_static, "");
+  VECT_VAR (vector_res, float, 32, 4) =
+    vfmaq_n_f32 (VECT_VAR (vsrc_1, float, 32, 4),
+         VECT_VAR (vsrc_2, float, 32, 4), elem3);
+  vst1q_f32 (VECT_VAR (result, float, 32, 4),
+         VECT_VAR (vector_res, float, 32, 4));
+  CHECK_FP (TEST_MSG, float, 32, 4, PRIx16, expectedfma3_static, "");
+
+#undef TEST_MSG
+#define TEST_MSG "VFMSQ_VFMAQ_N (FP64)"
+  clean_results ();
+
+  DECL_VARIABLE(vsrc_1, float, 64, 2);
+  DECL_VARIABLE(vsrc_2, float, 64, 2);
+  VECT_VAR_DECL (buf_src_1, float, 64, 2) [] = {DA0, DA1};
+  VECT_VAR_DECL (buf_src_2, float, 64, 2) [] = {DB0, DB1};
+  VLOAD (vsrc_1, buf_src_1, q, float, f, 64, 2);
+  VLOAD (vsrc_2, buf_src_2, q, float, f, 64, 2);
+  DECL_VARIABLE (vector_res, float, 64, 2) =
+    vfmsq_n_f64 (VECT_VAR (vsrc_1, float, 64, 2),
+         VECT_VAR (vsrc_2, float, 64, 2), delem0);
+  vst1q_f64 (VECT_VAR (result, float, 64, 2),
+         VECT_VAR (vector_res, float, 64, 2));
+  CHECK_FP (TEST_MSG, float, 64, 2, PRIx16, expectedfms0_static, "");
+  VECT_VAR (vector_res, float, 64, 2) =
+    vfmaq_n_f64 (VECT_VAR (vsrc_1, float, 64, 2),
+         VECT_VAR (vsrc_2, float, 64, 2), delem0);
+  vst1q_f64 (VECT_VAR (result, float, 64, 2),
+         VECT_VAR (vector_res, float, 64, 2));
+  CHECK_FP (TEST_MSG, float, 64, 2, PRIx16, expectedfma0_static, "");
+
+  VECT_VAR_DECL (buf_src_3, float, 64, 2) [] = {DA2, DA3};
+  VECT_VAR_DECL (buf_src_4, float, 64, 2) [] = {DB2, DB3};
+  VLOAD (vsrc_1, buf_src_3, q, float, f, 64, 2);
+  VLOAD (vsrc_2, buf_src_4, q, float, f, 64, 2);
+  VECT_VAR (vector_res, float, 64, 2) =
+    vfmsq_n_f64 (VECT_VAR (vsrc_1, float, 64, 2),
+         VECT_VAR (vsrc_2, float, 64, 2), delem1);
+  vst1q_f64 (VECT_VAR (result, float, 64, 2),
+         VECT_VAR (vector_res, float, 64, 2));
+  CHECK_FP (TEST_MSG, float, 64, 2, PRIx16, expectedfms1_static, "");
+  VECT_VAR (vector_res, float, 64, 2) =
+    vfmaq_n_f64 (VECT_VAR (vsrc_1, float, 64, 2),
+         VECT_VAR (vsrc_2, float, 64, 2), delem1);
+  vst1q_f64 (VECT_VAR (result, float, 64, 2),
+         VECT_VAR (vector_res, float, 64, 2));
+  CHECK_FP (TEST_MSG, float, 64, 2, PRIx16, expectedfma1_static, "");
+
+  VECT_VAR_DECL (buf_src_5, float, 64, 2) [] = {DA4, DA5};
+  VECT_VAR_DECL (buf_src_6, float, 64, 2) [] = {DB4, DB5};
+  VLOAD (vsrc_1, buf_src_5, q, float, f, 64, 2);
+  VLOAD (vsrc_2, buf_src_6, q, float, f, 64, 2);
+  VECT_VAR (vector_res, float, 64, 2) =
+    vfmsq_n_f64 (VECT_VAR (vsrc_1, float, 64, 2),
+         VECT_VAR (vsrc_2, float, 64, 2), delem2);
+  vst1q_f64 (VECT_VAR (result, float, 64, 2),
+         VECT_VAR (vector_res, float, 64, 2));
+  CHECK_FP (TEST_MSG, float, 64, 2, PRIx16, expectedfms2_static, "");
+  VECT_VAR (vector_res, float, 64, 2) =
+    vfmaq_n_f64 (VECT_VAR (vsrc_1, float, 64, 2),
+         VECT_VAR (vsrc_2, float, 64, 2), delem2);
+  vst1q_f64 (VECT_VAR (result, float, 64, 2),
+         VECT_VAR (vector_res, float, 64, 2));
+  CHECK_FP (TEST_MSG, float, 64, 2, PRIx16, expectedfma2_static, "");
+
+  VECT_VAR_DECL (buf_src_7, float, 64, 2) [] = {DA6, DA7};
+  VECT_VAR_DECL (buf_src_8, float, 64, 2) [] = {DB6, DB7};
+  VLOAD (vsrc_1, buf_src_7, q, float, f, 64, 2);
+  VLOAD (vsrc_2, buf_src_8, q, float, f, 64, 2);
+  VECT_VAR (vector_res, float, 64, 2) =
+    vfmsq_n_f64 (VECT_VAR (vsrc_1, float, 64, 2),
+         VECT_VAR (vsrc_2, float, 64, 2), delem3);
+  vst1q_f64 (VECT_VAR (result, float, 64, 2),
+         VECT_VAR (vector_res, float, 64, 2));
+  CHECK_FP (TEST_MSG, float, 64, 2, PRIx16, expectedfms3_static, "");
+  VECT_VAR (vector_res, float, 64, 2) =
+    vfmaq_n_f64 (VECT_VAR (vsrc_1, float, 64, 2),
+         VECT_VAR (vsrc_2, float, 64, 2), delem3);
+  vst1q_f64 (VECT_VAR (result, float, 64, 2),
+         VECT_VAR (vector_res, float, 64, 2));
+  CHECK_FP (TEST_MSG, float, 64, 2, PRIx16, expectedfma3_static, "");
+
+#undef TEST_MSG
+#define TEST_MSG "VFMS_VFMA_N (FP64)"
+  clean_results ();
+
+  DECL_VARIABLE(vsrc_1, float, 64, 1);
+  DECL_VARIABLE(vsrc_2, float, 64, 1);
+  VECT_VAR_DECL (buf_src_1, float, 64, 1) [] = {DA0};
+  VECT_VAR_DECL (buf_src_2, float, 64, 1) [] = {DB0};
+  VLOAD (vsrc_1, buf_src_1, , float, f, 64, 1);
+  VLOAD (vsrc_2, buf_src_2, , float, f, 64, 1);
+  DECL_VARIABLE (vector_res, float, 64, 1) =
+    vfms_n_f64 (VECT_VAR (vsrc_1, float, 64, 1),
+        VECT_VAR (vsrc_2, float, 64, 1), delem0);
+  vst1_f64 (VECT_VAR (result, float, 64, 1),
+         VECT_VAR (vector_res, float, 64, 1));
+  CHECK_FP (TEST_MSG, float, 64, 1, PRIx16, expectedfms0_static, "");
+  VECT_VAR (vector_res, float, 64, 1) =
+    vfma_n_f64 (VECT_VAR (vsrc_1, float, 64, 1),
+        VECT_VAR (vsrc_2, float, 64, 1), delem0);
+  vst1_f64 (VECT_VAR (result, float, 64, 1),
+         VECT_VAR (vector_res, float, 64, 1));
+  CHECK_FP (TEST_MSG, float, 64, 1, PRIx16, expectedfma0_static, "");
+
+  VECT_VAR_DECL (buf_src_3, float, 64, 1) [] = {DA2};
+  VECT_VAR_DECL (buf_src_4, float, 64, 1) [] = {DB2};
+  VLOAD (vsrc_1, buf_src_3, , float, f, 64, 1);
+  VLOAD (vsrc_2, buf_src_4, , float, f, 64, 1);
+  VECT_VAR (vector_res, float, 64, 1) =
+    vfms_n_f64 (VECT_VAR (vsrc_1, float, 64, 1),
+        VECT_VAR (vsrc_2, float, 64, 1), delem1);
+  vst1_f64 (VECT_VAR (result, float, 64, 1),
+         VECT_VAR (vector_res, float, 64, 1));
+  CHECK_FP (TEST_MSG, float, 64, 1, PRIx16, expectedfms1_static, "");
+  VECT_VAR (vector_res, float, 64, 1) =
+    vfma_n_f64 (VECT_VAR (vsrc_1, float, 64, 1),
+        VECT_VAR (vsrc_2, float, 64, 1), delem1);
+  vst1_f64 (VECT_VAR (result, float, 64, 1),
+         VECT_VAR (vector_res, float, 64, 1));
+  CHECK_FP (TEST_MSG, float, 64, 1, PRIx16, expectedfma1_static, "");
+
+  VECT_VAR_DECL (buf_src_5, float, 64, 1) [] = {DA4};
+  VECT_VAR_DECL (buf_src_6, float, 64, 1) [] = {DB4};
+  VLOAD (vsrc_1, buf_src_5, , float, f, 64, 1);
+  VLOAD (vsrc_2, buf_src_6, , float, f, 64, 1);
+  VECT_VAR (vector_res, float, 64, 1) =
+    vfms_n_f64 (VECT_VAR (vsrc_1, float, 64, 1),
+        VECT_VAR (vsrc_2, float, 64, 1), delem2);
+  vst1_f64 (VECT_VAR (result, float, 64, 1),
+         VECT_VAR (vector_res, float, 64, 1));
+  CHECK_FP (TEST_MSG, float, 64, 1, PRIx16, expectedfms2_static, "");
+  VECT_VAR (vector_res, float, 64, 1) =
+    vfma_n_f64 (VECT_VAR (vsrc_1, float, 64, 1),
+        VECT_VAR (vsrc_2, float, 64, 1), delem2);
+  vst1_f64 (VECT_VAR (result, float, 64, 1),
+         VECT_VAR (vector_res, float, 64, 1));
+  CHECK_FP (TEST_MSG, float, 64, 1, PRIx16, expectedfma2_static, "");
+
+  VECT_VAR_DECL (buf_src_7, float, 64, 1) [] = {DA6};
+  VECT_VAR_DECL (buf_src_8, float, 64, 1) [] = {DB6};
+  VLOAD (vsrc_1, buf_src_7, , float, f, 64, 1);
+  VLOAD (vsrc_2, buf_src_8, , float, f, 64, 1);
+  VECT_VAR (vector_res, float, 64, 1) =
+    vfms_n_f64 (VECT_VAR (vsrc_1, float, 64, 1),
+        VECT_VAR (vsrc_2, float, 64, 1), delem3);
+  vst1_f64 (VECT_VAR (result, float, 64, 1),
+         VECT_VAR (vector_res, float, 64, 1));
+  CHECK_FP (TEST_MSG, float, 64, 1, PRIx16, expectedfms3_static, "");
+  VECT_VAR (vector_res, float, 64, 1) =
+    vfma_n_f64 (VECT_VAR (vsrc_1, float, 64, 1),
+        VECT_VAR (vsrc_2, float, 64, 1), delem3);
+  vst1_f64 (VECT_VAR (result, float, 64, 1),
+         VECT_VAR (vector_res, float, 64, 1));
+  CHECK_FP (TEST_MSG, float, 64, 1, PRIx16, expectedfma3_static, "");
+}
+#endif
+
+int
+main (void)
+{
+#if defined(__aarch64__) && defined(__ARM_FEATURE_FMA)
+  exec_vfma_vfms_n ();
+#endif
+  return 0;
+}
diff --git a/gcc/testsuite/gcc.target/aarch64/fmla_intrinsic_1.c
b/gcc/testsuite/gcc.target/aarch64/fmla_intrinsic_1.c
index
1ba1fed98a0711496815e00d2d702e5bfa2a7d43..5b348827002dcfef1f589900a4cf5ff7ada26697
100644
--- a/gcc/testsuite/gcc.target/aarch64/fmla_intrinsic_1.c
+++ b/gcc/testsuite/gcc.target/aarch64/fmla_intrinsic_1.c
@@ -110,6 +110,6 @@ main (int argc, char **argv)
   /* vfmaq_lane_f64.
      vfma_laneq_f64.
      vfmaq_laneq_f64.  */
-/* { dg-final { scan-assembler-times "fmla\\tv\[0-9\]+\.2d,
v\[0-9\]+\.2d, v\[0-9\]+\.2d\\\[\[0-9\]+\\\]" 3 } } */
+/* { dg-final { scan-assembler-times "fmla\\tv\[0-9\]+\.2d,
v\[0-9\]+\.2d, v\[0-9\]+\.2?d\\\[\[0-9\]+\\\]" 3 } } */
     diff --git a/gcc/testsuite/gcc.target/aarch64/fmls_intrinsic_1.c
b/gcc/testsuite/gcc.target/aarch64/fmls_intrinsic_1.c
index
887ebae10da715c8d301a8494a2225e53f15bd7d..6c194a023d34ebafb4d732edc303985531f92a63
100644
--- a/gcc/testsuite/gcc.target/aarch64/fmls_intrinsic_1.c
+++ b/gcc/testsuite/gcc.target/aarch64/fmls_intrinsic_1.c
@@ -111,6 +111,6 @@ main (int argc, char **argv)
   /* vfmsq_lane_f64.
      vfms_laneq_f64.
      vfmsq_laneq_f64.  */
-/* { dg-final { scan-assembler-times "fmls\\tv\[0-9\]+\.2d,
v\[0-9\]+\.2d, v\[0-9\]+\.2d\\\[\[0-9\]+\\\]" 3 } } */
+/* { dg-final { scan-assembler-times "fmls\\tv\[0-9\]+\.2d,
v\[0-9\]+\.2d, v\[0-9\]+\.2?d\\\[\[0-9\]+\\\]" 3 } } */

























					Reply via email to