MCPcopy Create free account
hub / github.com/ARM-software/ComputeLibrary / vdequantize

Function vdequantize

src/core/NEON/NESymm.h:171–182  ·  view source on GitHub ↗

Dequantize a neon vector holding 16 16-bit quantized values. * * @param[in] qv Input values to be dequantized. * @param[in] qi Quantization information to be used in the computation. * * @return Dequantized values in a neon vector */

Source from the content-addressed store, hash-verified

169 * @return Dequantized values in a neon vector
170 */
171inline float32x4x4_t vdequantize(const int16x8x2_t &qv, const UniformQuantizationInfo &qi)
172{
173 const float scale = qi.scale;
174 const float32x4_t vscale = vdupq_n_f32(scale);
175 const float32x4x4_t vdequantized_input = {{
176 vmulq_f32(vcvtq_f32_s32(vmovl_s16(vget_low_s16(qv.val[0]))), vscale),
177 vmulq_f32(vcvtq_f32_s32(vmovl_s16(vget_high_s16(qv.val[0]))), vscale),
178 vmulq_f32(vcvtq_f32_s32(vmovl_s16(vget_low_s16(qv.val[1]))), vscale),
179 vmulq_f32(vcvtq_f32_s32(vmovl_s16(vget_high_s16(qv.val[1]))), vscale),
180 }};
181 return vdequantized_input;
182}
183
184/** Quantize a neon vector holding 16 floating point values.
185 *

Callers 14

batch_concatFunction · 0.50
run_opMethod · 0.50
run_opMethod · 0.50
mul_saturate_quantized_8Function · 0.50
depth_concatFunction · 0.50
elementwise_op<int8_t>Function · 0.50
elementwise_op<uint8_t>Function · 0.50

Calls

no outgoing calls

Tested by

no test coverage detected