aarch64: Fix AdvSIMD libmvec routines for big-endian

Previously many routines used * to load from vector types stored
in the data table. This is emitted as ldr, which byte-swaps the
entire vector register, and causes bugs for big-endian when not
all lanes contain the same value. When a vector is to be used
this way, it has been replaced with an array and the load with an
explicit ld1 intrinsic, which byte-swaps only within lanes.

As well, many routines previously used non-standard GCC syntax
for vector operations such as indexing into vectors types with []
and assembling vectors using {}. This syntax should not be mixed
with ACLE, as the former does not respect endianness whereas the
latter does. Such examples have been replaced with, for instance,
vcombine_* and vgetq_lane* intrinsics. Helpers which only use the
GCC syntax, such as the v_call helpers, do not need changing as
they do not use intrinsics.

Reviewed-by: Szabolcs Nagy <szabolcs.nagy@arm.com>
(cherry picked from commit 90a6ca8b28)

sysdeps/aarch64/fpu/exp10f_advsimd.c

@@ -25,7 +25,8 @@
 static const struct data
 {
   float32x4_t poly[5];
-  float32x4_t log10_2_and_inv, shift;
+  float log10_2_and_inv[4];
+  float32x4_t shift;
 
 #if !WANT_SIMD_EXCEPT
   float32x4_t scale_thresh;
@@ -111,10 +112,11 @@ float32x4_t VPCS_ATTR NOINLINE V_NAME_F1 (exp10) (float32x4_t x)
   /* exp10(x) = 2^n * 10^r = 2^n * (1 + poly (r)),
      with poly(r) in [1/sqrt(2), sqrt(2)] and
      x = r + n * log10 (2), with r in [-log10(2)/2, log10(2)/2].  */
-  float32x4_t z = vfmaq_laneq_f32 (d->shift, x, d->log10_2_and_inv, 0);
+  float32x4_t log10_2_and_inv = vld1q_f32 (d->log10_2_and_inv);
+  float32x4_t z = vfmaq_laneq_f32 (d->shift, x, log10_2_and_inv, 0);
   float32x4_t n = vsubq_f32 (z, d->shift);
-  float32x4_t r = vfmsq_laneq_f32 (x, n, d->log10_2_and_inv, 1);
-  r = vfmsq_laneq_f32 (r, n, d->log10_2_and_inv, 2);
+  float32x4_t r = vfmsq_laneq_f32 (x, n, log10_2_and_inv, 1);
+  r = vfmsq_laneq_f32 (r, n, log10_2_and_inv, 2);
   uint32x4_t e = vshlq_n_u32 (vreinterpretq_u32_f32 (z), 23);
 
   float32x4_t scale = vreinterpretq_f32_u32 (vaddq_u32 (e, ExponentBias));

sysdeps/aarch64/fpu/expm1_advsimd.c

@@ -23,7 +23,9 @@
 static const struct data
 {
   float64x2_t poly[11];
-  float64x2_t invln2, ln2, shift;
+  float64x2_t invln2;
+  double ln2[2];
+  float64x2_t shift;
   int64x2_t exponent_bias;
 #if WANT_SIMD_EXCEPT
   uint64x2_t thresh, tiny_bound;
@@ -92,8 +94,9 @@ float64x2_t VPCS_ATTR V_NAME_D1 (expm1) (float64x2_t x)
      where 2^i is exact because i is an integer.  */
   float64x2_t n = vsubq_f64 (vfmaq_f64 (d->shift, d->invln2, x), d->shift);
   int64x2_t i = vcvtq_s64_f64 (n);
-  float64x2_t f = vfmsq_laneq_f64 (x, n, d->ln2, 0);
-  f = vfmsq_laneq_f64 (f, n, d->ln2, 1);
+  float64x2_t ln2 = vld1q_f64 (&d->ln2[0]);
+  float64x2_t f = vfmsq_laneq_f64 (x, n, ln2, 0);
+  f = vfmsq_laneq_f64 (f, n, ln2, 1);
 
   /* Approximate expm1(f) using polynomial.
      Taylor expansion for expm1(x) has the form:

sysdeps/aarch64/fpu/expm1f_advsimd.c

@@ -23,7 +23,7 @@
 static const struct data
 {
   float32x4_t poly[5];
-  float32x4_t invln2_and_ln2;
+  float invln2_and_ln2[4];
   float32x4_t shift;
   int32x4_t exponent_bias;
 #if WANT_SIMD_EXCEPT
@@ -88,11 +88,12 @@ float32x4_t VPCS_ATTR NOINLINE V_NAME_F1 (expm1) (float32x4_t x)
      and f = x - i * ln2, then f is in [-ln2/2, ln2/2].
      exp(x) - 1 = 2^i * (expm1(f) + 1) - 1
      where 2^i is exact because i is an integer.  */
-  float32x4_t j = vsubq_f32 (
-      vfmaq_laneq_f32 (d->shift, x, d->invln2_and_ln2, 0), d->shift);
+  float32x4_t invln2_and_ln2 = vld1q_f32 (d->invln2_and_ln2);
+  float32x4_t j
+      = vsubq_f32 (vfmaq_laneq_f32 (d->shift, x, invln2_and_ln2, 0), d->shift);
   int32x4_t i = vcvtq_s32_f32 (j);
-  float32x4_t f = vfmsq_laneq_f32 (x, j, d->invln2_and_ln2, 1);
-  f = vfmsq_laneq_f32 (f, j, d->invln2_and_ln2, 2);
+  float32x4_t f = vfmsq_laneq_f32 (x, j, invln2_and_ln2, 1);
+  f = vfmsq_laneq_f32 (f, j, invln2_and_ln2, 2);
 
   /* Approximate expm1(f) using polynomial.
      Taylor expansion for expm1(x) has the form:

sysdeps/aarch64/fpu/log10_advsimd.c

@@ -58,8 +58,10 @@ static inline struct entry
 lookup (uint64x2_t i)
 {
   struct entry e;
-  uint64_t i0 = (i[0] >> (52 - V_LOG10_TABLE_BITS)) & IndexMask;
-  uint64_t i1 = (i[1] >> (52 - V_LOG10_TABLE_BITS)) & IndexMask;
+  uint64_t i0
+      = (vgetq_lane_u64 (i, 0) >> (52 - V_LOG10_TABLE_BITS)) & IndexMask;
+  uint64_t i1
+      = (vgetq_lane_u64 (i, 1) >> (52 - V_LOG10_TABLE_BITS)) & IndexMask;
   float64x2_t e0 = vld1q_f64 (&__v_log10_data.table[i0].invc);
   float64x2_t e1 = vld1q_f64 (&__v_log10_data.table[i1].invc);
   e.invc = vuzp1q_f64 (e0, e1);

sysdeps/aarch64/fpu/log2_advsimd.c

@@ -55,8 +55,10 @@ static inline struct entry
 lookup (uint64x2_t i)
 {
   struct entry e;
-  uint64_t i0 = (i[0] >> (52 - V_LOG2_TABLE_BITS)) & IndexMask;
-  uint64_t i1 = (i[1] >> (52 - V_LOG2_TABLE_BITS)) & IndexMask;
+  uint64_t i0
+      = (vgetq_lane_u64 (i, 0) >> (52 - V_LOG2_TABLE_BITS)) & IndexMask;
+  uint64_t i1
+      = (vgetq_lane_u64 (i, 1) >> (52 - V_LOG2_TABLE_BITS)) & IndexMask;
   float64x2_t e0 = vld1q_f64 (&__v_log2_data.table[i0].invc);
   float64x2_t e1 = vld1q_f64 (&__v_log2_data.table[i1].invc);
   e.invc = vuzp1q_f64 (e0, e1);

sysdeps/aarch64/fpu/log_advsimd.c

@@ -54,17 +54,12 @@ lookup (uint64x2_t i)
 {
   /* Since N is a power of 2, n % N = n & (N - 1).  */
   struct entry e;
-  uint64_t i0 = (i[0] >> (52 - V_LOG_TABLE_BITS)) & IndexMask;
-  uint64_t i1 = (i[1] >> (52 - V_LOG_TABLE_BITS)) & IndexMask;
+  uint64_t i0 = (vgetq_lane_u64 (i, 0) >> (52 - V_LOG_TABLE_BITS)) & IndexMask;
+  uint64_t i1 = (vgetq_lane_u64 (i, 1) >> (52 - V_LOG_TABLE_BITS)) & IndexMask;
   float64x2_t e0 = vld1q_f64 (&__v_log_data.table[i0].invc);
   float64x2_t e1 = vld1q_f64 (&__v_log_data.table[i1].invc);
-#if __BYTE_ORDER == __LITTLE_ENDIAN
   e.invc = vuzp1q_f64 (e0, e1);
   e.logc = vuzp2q_f64 (e0, e1);
-#else
-  e.invc = vuzp1q_f64 (e1, e0);
-  e.logc = vuzp2q_f64 (e1, e0);
-#endif
   return e;
 }
 

sysdeps/aarch64/fpu/tan_advsimd.c

@@ -23,7 +23,8 @@
 static const struct data
 {
   float64x2_t poly[9];
-  float64x2_t half_pi, two_over_pi, shift;
+  double half_pi[2];
+  float64x2_t two_over_pi, shift;
 #if !WANT_SIMD_EXCEPT
   float64x2_t range_val;
 #endif
@@ -81,8 +82,9 @@ float64x2_t VPCS_ATTR V_NAME_D1 (tan) (float64x2_t x)
   /* Use q to reduce x to r in [-pi/4, pi/4], by:
      r = x - q * pi/2, in extended precision.  */
   float64x2_t r = x;
-  r = vfmsq_laneq_f64 (r, q, dat->half_pi, 0);
-  r = vfmsq_laneq_f64 (r, q, dat->half_pi, 1);
+  float64x2_t half_pi = vld1q_f64 (dat->half_pi);
+  r = vfmsq_laneq_f64 (r, q, half_pi, 0);
+  r = vfmsq_laneq_f64 (r, q, half_pi, 1);
   /* Further reduce r to [-pi/8, pi/8], to be reconstructed using double angle
      formula.  */
   r = vmulq_n_f64 (r, 0.5);

sysdeps/aarch64/fpu/tanf_advsimd.c

@@ -23,7 +23,7 @@
 static const struct data
 {
   float32x4_t poly[6];
-  float32x4_t pi_consts;
+  float pi_consts[4];
   float32x4_t shift;
 #if !WANT_SIMD_EXCEPT
   float32x4_t range_val;
@@ -95,16 +95,17 @@ float32x4_t VPCS_ATTR NOINLINE V_NAME_F1 (tan) (float32x4_t x)
 #endif
 
   /* n = rint(x/(pi/2)).  */
-  float32x4_t q = vfmaq_laneq_f32 (d->shift, x, d->pi_consts, 3);
+  float32x4_t pi_consts = vld1q_f32 (d->pi_consts);
+  float32x4_t q = vfmaq_laneq_f32 (d->shift, x, pi_consts, 3);
   float32x4_t n = vsubq_f32 (q, d->shift);
   /* Determine if x lives in an interval, where |tan(x)| grows to infinity.  */
   uint32x4_t pred_alt = vtstq_u32 (vreinterpretq_u32_f32 (q), v_u32 (1));
 
   /* r = x - n * (pi/2)  (range reduction into -pi./4 .. pi/4).  */
   float32x4_t r;
-  r = vfmaq_laneq_f32 (x, n, d->pi_consts, 0);
-  r = vfmaq_laneq_f32 (r, n, d->pi_consts, 1);
-  r = vfmaq_laneq_f32 (r, n, d->pi_consts, 2);
+  r = vfmaq_laneq_f32 (x, n, pi_consts, 0);
+  r = vfmaq_laneq_f32 (r, n, pi_consts, 1);
+  r = vfmaq_laneq_f32 (r, n, pi_consts, 2);
 
   /* If x lives in an interval, where |tan(x)|
      - is finite, then use a polynomial approximation of the form