diff options
| author | Anthony Barbier <anthony.barbier@arm.com> | 2018-08-10 17:36:36 +0100 |
|---|---|---|
| committer | Anthony Barbier <anthony.barbier@arm.com> | 2018-11-02 16:54:54 +0000 |
| commit | 3a6163ed0c2d0ab4cac0456e8f66c704c6ad10c2 (patch) | |
| tree | f2e34b47fb0b8bf1fac18c337a67c0145def7b7b /arm_compute/core/NEON/NEMath.inl | |
| parent | e3f1bd10e71075f7d8b406698809666e1eb4e7ae (diff) | |
| download | ComputeLibrary-3a6163ed0c2d0ab4cac0456e8f66c704c6ad10c2.tar.gz | |
COMPMID-1499: Fixed issues to build for FP16 on Android
Change-Id: I7cd15e9115b5c6f544005528d69061751286be11
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/143708
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele DiGiorgio <michele.digiorgio@arm.com>
Diffstat (limited to 'arm_compute/core/NEON/NEMath.inl')
| -rw-r--r-- | arm_compute/core/NEON/NEMath.inl | 56 |
1 files changed, 28 insertions, 28 deletions
diff --git a/arm_compute/core/NEON/NEMath.inl b/arm_compute/core/NEON/NEMath.inl index 84154020a5..61d25d115c 100644 --- a/arm_compute/core/NEON/NEMath.inl +++ b/arm_compute/core/NEON/NEMath.inl @@ -173,35 +173,7 @@ inline float32x4_t vpowq_f32(float32x4_t val, float32x4_t n) #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC /** Exponent polynomial coefficients */ -const std::array<float16x8_t, 8> exp_tab_f16 = -{ - { - vdupq_n_f16(1.f), - vdupq_n_f16(0.0416598916054f), - vdupq_n_f16(0.500000596046f), - vdupq_n_f16(0.0014122662833f), - vdupq_n_f16(1.00000011921f), - vdupq_n_f16(0.00833693705499f), - vdupq_n_f16(0.166665703058f), - vdupq_n_f16(0.000195780929062f), - } -}; - /** Logarithm polynomial coefficients */ -const std::array<float16x8_t, 8> log_tab_f16 = -{ - { - vdupq_n_f16(-2.29561495781f), - vdupq_n_f16(-2.47071170807f), - vdupq_n_f16(-5.68692588806f), - vdupq_n_f16(-0.165253549814f), - vdupq_n_f16(5.17591238022f), - vdupq_n_f16(0.844007015228f), - vdupq_n_f16(4.58445882797f), - vdupq_n_f16(0.0141278216615f), - } -}; - #ifndef DOXYGEN_SKIP_THIS inline float16x4_t vinvsqrt_f16(float16x4_t x) { @@ -264,6 +236,20 @@ inline float16x8_t vtaylor_polyq_f16(float16x8_t x, const std::array<float16x8_t inline float16x8_t vexpq_f16(float16x8_t x) { + static const std::array<float16x8_t, 8> exp_tab_f16 = + { + { + vdupq_n_f16(1.f), + vdupq_n_f16(0.0416598916054f), + vdupq_n_f16(0.500000596046f), + vdupq_n_f16(0.0014122662833f), + vdupq_n_f16(1.00000011921f), + vdupq_n_f16(0.00833693705499f), + vdupq_n_f16(0.166665703058f), + vdupq_n_f16(0.000195780929062f), + } + }; + static const float16x8_t CONST_LN2 = vdupq_n_f16(0.6931471805f); // ln(2) static const float16x8_t CONST_INV_LN2 = vdupq_n_f16(1.4426950408f); // 1/ln(2) static const float16x8_t CONST_0 = vdupq_n_f16(0.f); @@ -285,6 +271,20 @@ inline float16x8_t vexpq_f16(float16x8_t x) inline float16x8_t vlogq_f16(float16x8_t x) { + static const std::array<float16x8_t, 8> log_tab_f16 = + { + { + vdupq_n_f16(-2.29561495781f), + vdupq_n_f16(-2.47071170807f), + vdupq_n_f16(-5.68692588806f), + vdupq_n_f16(-0.165253549814f), + vdupq_n_f16(5.17591238022f), + vdupq_n_f16(0.844007015228f), + vdupq_n_f16(4.58445882797f), + vdupq_n_f16(0.0141278216615f), + } + }; + static const int16x8_t CONST_127 = vdupq_n_s16(127); // 127 static const float16x8_t CONST_LN2 = vdupq_n_f16(0.6931471805f); // ln(2) |