xmos-ai-tools 1.1.2.dev263__py3-none-macosx_11_0_arm64.whl → 1.1.2.dev276__py3-none-macosx_11_0_arm64.whl

xmos_ai_tools/runtime/include/lib_nn/api/nn_layers.h

@@ -51,7 +51,7 @@ typedef struct {
  * channels)
  * @param job_count [in]    The number of jobs to be initialized.
  */
-void bsign_8_prepare(nn_bsign_8_job_t* jobs, int8_t* zero_point_vect,
+void bsign_8_prepare(nn_bsign_8_job_t *jobs, int8_t *zero_point_vect,
                      const uint32_t N, const int8_t zero_point,
                      const int32_t job_count);
 
@@ -82,8 +82,8 @@ void bsign_8_prepare(nn_bsign_8_job_t* jobs, int8_t* zero_point_vect,
  * @param plan [in]    The @oper{bsign_8} plan to be processed
  * @param job [in]     The @oper{bsign_8} job to be processed
  */
-void bsign_8(bnn_b32_t* Y, const int8_t* X, const int8_t* zero_point_vect,
-             const nn_bsign_8_job_t* job);
+void bsign_8(bnn_b32_t *Y, const int8_t *X, const int8_t *zero_point_vect,
+             const nn_bsign_8_job_t *job);
 
 /**
  * Struct represents the parameters needed by each `pad_run()` job.
@@ -124,8 +124,9 @@ typedef struct padding_sizes_t {
  * @param x                [in]   Look-up table @tensor{T}
  * @param bytes_per_pixel  [in]   Length @math{N} of input and output vectors
  */
-C_API void pad_prepare(nn_pad_plan_t* plan, const padding_sizes_t* p,
-                 const nn_image_params_t* x, const unsigned bytes_per_pixel);
+C_API void pad_prepare(nn_pad_plan_t *plan, const padding_sizes_t *p,
+                       const nn_image_params_t *x,
+                       const unsigned bytes_per_pixel);
 
 /**
  * @brief Execute @oper{pad_run} job.
@@ -145,19 +146,17 @@ C_API void pad_prepare(nn_pad_plan_t* plan, const padding_sizes_t* p,
  * @param x   [in]     The input vector @tensor{x}
  * @param plan [in]    The prameters describing how to pad.
  */
-void pad_run(char* y, char* x, const nn_pad_plan_t* p, uint32_t pad_value);
+void pad_run(char *y, char *x, const nn_pad_plan_t *p, uint32_t pad_value);
 
-void pad_ref(char* y, char* x, const padding_sizes_t* p,
-             const nn_image_params_t* xp, const unsigned bytes_per_pixel,
+void pad_ref(char *y, char *x, const padding_sizes_t *p,
+             const nn_image_params_t *xp, const unsigned bytes_per_pixel,
              uint32_t pad_value);
 
-
 /**
  * Func to calculate n_3
-*/
-void pad_3_to_4_prepare(uint32_t * n_3, 
-    const unsigned height, 
-    const unsigned width);
+ */
+void pad_3_to_4_prepare(uint32_t *n_3, const unsigned height,
+                        const unsigned width);
 
 /** Function that pads an image with 3-byte values with a 0.
  * The output image must be word aligned. This function solves the general
@@ -169,28 +168,28 @@ void pad_3_to_4_prepare(uint32_t * n_3,
  *
  * @returns  The inner product
  */
-extern void pad_3_to_4_run(int8_t outputs[], int8_t inputs[], uint32_t N_3, uint32_t pad_val);
-extern void pad_3_to_4_ref(int8_t outputs[], int8_t inputs[], uint32_t N_3, uint32_t pad_val);
+extern void pad_3_to_4_run(int8_t outputs[], int8_t inputs[], uint32_t N_3,
+                           uint32_t pad_val);
+extern void pad_3_to_4_ref(int8_t outputs[], int8_t inputs[], uint32_t N_3,
+                           uint32_t pad_val);
 
 typedef struct nn_mul_params_t {
-    int8_t in1_zero_point;
-    int8_t in2_zero_point;
-    int16_t bias;
-    int16_t scalar; 
-    int16_t vlashr_shr;
+  int8_t in1_zero_point;
+  int8_t in2_zero_point;
+  int16_t bias;
+  int16_t scalar;
+  int16_t vlashr_shr;
 } nn_mul_params_t;
 
-void mul_boggle(nn_mul_params_t * params, 
-    double in1Scale, 
-    double in2Scale, 
-    double outputScale,
-    int8_t in1ZeroPoint,
-    int8_t in2ZeroPoint, 
-    int8_t outputZeroPoint);
-void mul_elementwise(int8_t* in1_data, int8_t* in2_data, int element_count, nn_mul_params_t * params, int8_t * out_data);
+void mul_boggle(nn_mul_params_t *params, double in1Scale, double in2Scale,
+                double outputScale, int8_t in1ZeroPoint, int8_t in2ZeroPoint,
+                int8_t outputZeroPoint);
+void mul_elementwise(const int8_t *in1_data, const int8_t *in2_data, int element_count,
+                     nn_mul_params_t *params, int8_t *out_data);
 
 // /**
-//  * Describes the parameters needed for an @oper{add_elementwise} operator. @see add_elementwise().
+//  * Describes the parameters needed for an @oper{add_elementwise} operator.
+//  @see add_elementwise().
 //  */
 // typedef struct {
 //     /**
@@ -219,35 +218,38 @@ void mul_elementwise(int8_t* in1_data, int8_t* in2_data, int element_count, nn_m
 // } nn_add_params_t;
 
 typedef struct {
-        int16_t m1[16];
-        int16_t m2[16];
-        int16_t shift[16];
-        int16_t bias_hi[16];
-        int16_t bias_lo[16];
+  int16_t m1[16];
+  int16_t m2[16];
+  int16_t shift[16];
+  int16_t bias_hi[16];
+  int16_t bias_lo[16];
 } nn_add_params_t;
 
 /**
  * @brief Invoke an @oper{add_elementwise} job.
  *
- * The @oper{add_elementwise} operator adds together two quantized 8-bit input vectors, @tensor{x_0} and @tensor{x_1}
- * element-by-element to produce the output vector @tensor{y}. This function assumes that the input vectors and the 
- * output vector each require different quantization parameters.
+ * The @oper{add_elementwise} operator adds together two quantized 8-bit input
+ * vectors, @tensor{x_0} and @tensor{x_1} element-by-element to produce the
+ * output vector @tensor{y}. This function assumes that the input vectors and
+ * the output vector each require different quantization parameters.
  *
- * In order to add together two quantized vectors, their quantization parameters must match. The contents of `params`
- * indicate how to do this.
+ * In order to add together two quantized vectors, their quantization parameters
+ * must match. The contents of `params` indicate how to do this.
  *
  * @par Parameter Details
  *
  * `Y` points to the output vector @tensor{y} with shape @tensor_shape{N}.
  *
- * `X0` and `X1` respectively point to the first and second input vectors @tensor{x_0} and @tensor{x_1}, each with shape
+ * `X0` and `X1` respectively point to the first and second input vectors
+ * @tensor{x_0} and @tensor{x_1}, each with shape
  * @tensor_shape{N}.
  *
- * `params` describes the parameters @math{s_i}, @math{m_i}, @math{b} and @math{s_{out}} which are applied for each
- * output element.
+ * `params` describes the parameters @math{s_i}, @math{m_i}, @math{b} and
+ * @math{s_{out}} which are applied for each output element.
  *
- * `elm_start` and `elm_count` together specify which output elements @math{y[k]} should be calculated by this
- * invocation. Specifically, this invocation will calculate @math{y[k]} for which `elm_start` @math{\le k \lt}
+ * `elm_start` and `elm_count` together specify which output elements
+ * @math{y[k]} should be calculated by this invocation. Specifically, this
+ * invocation will calculate @math{y[k]} for which `elm_start` @math{\le k \lt}
  * `(elm_start + elm_count)`.
  *
  * @param[out]  Y           The output vector @tensor{y}
@@ -257,30 +259,26 @@ typedef struct {
  * @param[in]   elm_start   Index of first output element to be computed
  * @param[in]   elm_count   Number of output elements to be computed
  */
-void add_elementwise(
-    int8_t Y[],
-    const int8_t X1[],
-    const int8_t X2[],
-    nn_add_params_t *p,
-    const int elm_start,
-    const int elm_count);
+void add_elementwise(int8_t Y[], const int8_t X1[], const int8_t X2[],
+                     nn_add_params_t *p, const int elm_start,
+                     const int elm_count);
 
-/** 
+/**
  * @brief Execute @oper{lookup8} job.
- * 
+ *
  * See @oper_ref{lookup8} for more details about the @oper{lookup8} operator.
- * 
- * Unlike other operators, instances of @oper{lookup8} do not require plans or jobs and no initialization is
- * necessary.
- * 
+ *
+ * Unlike other operators, instances of @oper{lookup8} do not require plans or
+ * jobs and no initialization is necessary.
+ *
  * `Y` points to the output vector @tensor{y} with length @math{N}.
- * 
- * `X` points to the input vector @tensor{x} with length @math{N}. 
- * 
+ *
+ * `X` points to the input vector @tensor{x} with length @math{N}.
+ *
  * `lut` points to the look-up table @math{T} with shape @tensor_shape{256}.
- * 
+ *
  * `N` is the length @math{N} of the input vector @tensor{x}.
- * 
+ *
  * @requires_word_alignment{Y,X}
  *
  * @param Y      [out]  The output vector @tensor{y}
@@ -288,11 +286,20 @@ void add_elementwise(
  * @param lut    [in]   Look-up table @tensor{T}
  * @param N      [in]   Length @math{N} of input and output vectors
  */
-void lookup8(
-    uint8_t* Y,
-    const uint8_t* X,
-    const uint8_t* lut,
-    const unsigned elm_start,
-    const unsigned elm_count);
+void lookup8(uint8_t *Y, const uint8_t *X, const uint8_t *lut,
+             const unsigned elm_start, const unsigned elm_count);
+
+void softmax_exp_sum(float *Y, const int8_t *X, const float *lut,
+                     const unsigned elm_start, const unsigned elm_count);
+
+void softmax_exp_div(int8_t *Y, const int8_t *X, const float *lut,
+                     const float inv_sum, const unsigned elm_start,
+                     const unsigned elm_count);
+
+void softmax_calculate_inv_sum(float *inv_sum, const float sums[]);
+
+void softmax_generate_exp_lut(int zero_point, float scale, float *lut);
 
-#endif  // LAYERS_H_
+void softmax_ref(int8_t *Y, const int8_t *X, const float zero_point,
+                 const float scale, const int length);
+#endif // LAYERS_H_

xmos_ai_tools/runtime/include/lib_nn/api/quadratic_approximation.h

@@ -7,6 +7,7 @@
 #define ACTIVATION_FUNCTION /**/
 #endif
 
+#include "nn_api.h"
 #include <stdint.h>
 
 #define QUADRATIC_APPROXIMATION_MAX_CHUNKS   2048
@@ -47,7 +48,7 @@ typedef float (*float_function_t)(float x);
  * \param max_error     maximum error, ought to be 1
  * \param error         sqrt of sum of squared errors.
  */
-extern void quadratic_approximation_generator(
+C_API void quadratic_approximation_generator(
     quadratic_function_table_t *table,
     ACTIVATION_FUNCTION float_function_t av,
     double input_scaler,
@@ -61,7 +62,7 @@ extern void quadratic_approximation_generator(
  * \param x   the table
  * \returns   The number of bytes in the table
  */
-extern uint32_t quadratic_function_table_number_bytes(quadratic_function_table_t *x);
+C_API uint32_t quadratic_function_table_number_bytes(quadratic_function_table_t *x);
 
 
 /** Function that returns a pointer to the bytes in an approximation table
@@ -69,12 +70,12 @@ extern uint32_t quadratic_function_table_number_bytes(quadratic_function_table_t
  * \param x   the table
  * \returns   Pointer to the bytes in the table
  */
-extern uint8_t *quadratic_function_table_bytes(quadratic_function_table_t *x);
+C_API uint8_t *quadratic_function_table_bytes(quadratic_function_table_t *x);
 
 /** Example functions that can be passed in 
  */
-extern float approximation_function_tanh(float x);
-extern float approximation_function_logistics(float x);
-extern float approximation_function_elu(float x);
+C_API float approximation_function_tanh(float x);
+C_API float approximation_function_logistics(float x);
+C_API float approximation_function_elu(float x);
 
 #endif

xmos_ai_tools/runtime/include/lib_tflite_micro/src/tflite-xcore-kernels/xcore_common.h

@@ -9,7 +9,7 @@ namespace ops {
 namespace micro {
 namespace xcore {
 
-void calculateThreadSplit(int tc, int split_size, int split_start[], int split_end[]);
+void calculateThreadSplit(int &tc, int split_size, int split_start[], int split_end[]);
 
 } // namespace xcore
 } // namespace micro

{xmos_ai_tools-1.1.2.dev263.dist-info → xmos_ai_tools-1.1.2.dev276.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: xmos-ai-tools
-Version: 1.1.2.dev263
+Version: 1.1.2.dev276
 Summary: XMOS AI Tools
 Home-page: https://github.com/xmos/ai_tools
 Author: XMOS
@@ -31,5 +31,5 @@ Requires-Dist: tflite >=2.4.0
 Documentation
 -------------
 
-Click [here](https://github.com/xmos/ai_tools/blob/3c0d86fafcd3d15bc990c408195f713aed12c306/README.md) for documentation on using xmos-ai-tools to deploy AI models on xcore.
+Click [here](https://github.com/xmos/ai_tools/blob/fc0ce8436409aa258396c1cf20a0ff0fdc647745/README.md) for documentation on using xmos-ai-tools to deploy AI models on xcore.
 

{xmos_ai_tools-1.1.2.dev263.dist-info → xmos_ai_tools-1.1.2.dev276.dist-info}/RECORD

@@ -43,7 +43,7 @@ xmos_ai_tools/runtime/include/lib_nn/api/nn_bin_types.h,sha256=DTFX2ecfzwl-_tHlq
 xmos_ai_tools/runtime/include/lib_nn/api/nn_config.h,sha256=KFoA5zyOTTHi6x-qDuIWkF8wAz4Gc4BdOBgUzOArOD0,11558
 xmos_ai_tools/runtime/include/lib_nn/api/nn_conv2d_structs.h,sha256=Kfyq3bD1aLcSnWTB_cijU_D5ThGol5utJ1k-WsjqeMY,2268
 xmos_ai_tools/runtime/include/lib_nn/api/nn_image.h,sha256=gz7NyxmZACZiDED0wT7px2qjEEMcUwfa5TjeOUmYzL4,514
-xmos_ai_tools/runtime/include/lib_nn/api/nn_layers.h,sha256=znneYtZCM-WEK896MB_Zy41MMiX3ROGxUVZ3iWNwAjc,9955
+xmos_ai_tools/runtime/include/lib_nn/api/nn_layers.h,sha256=EUzNW8g-5DiLAbT3w-ZgL_0nX9ptUmB-vaMdbSroERA,10646
 xmos_ai_tools/runtime/include/lib_nn/api/nn_op_helper.h,sha256=VMnBVEcgH0wtiiV91JjtFRG4xkCI0m7WrikEmU6ulss,3716
 xmos_ai_tools/runtime/include/lib_nn/api/nn_op_utils.h,sha256=zoyHr07-ND_DCcS0ovehshCnXTrKBTlYOPX5Ul69JRs,4678
 xmos_ai_tools/runtime/include/lib_nn/api/nn_operator.h,sha256=Nj4jiXVHlnFjVBtV7mCGulTzSyNwCIpitz8Ao9kSIsU,337
@@ -53,7 +53,7 @@ xmos_ai_tools/runtime/include/lib_nn/api/nn_window_params.h,sha256=NeCS7hhE7tdmc
 xmos_ai_tools/runtime/include/lib_nn/api/output_transform_fn_int16.h,sha256=5Na926JTr8Sg0KfeSt3fapzJQdEX4Gmf_oXM_oGsFvM,2122
 xmos_ai_tools/runtime/include/lib_nn/api/output_transform_fn_int16_kernel_transform.h,sha256=9XRx__gL9IuiCsyNh3XrgCa_aWgC6j_2nXCS0a7asSU,390
 xmos_ai_tools/runtime/include/lib_nn/api/output_transform_fn_int16_mappings.h,sha256=claoCFq-IVmD1rDCjMRruum0u1-ADFC-5KE4FPLwLoU,482
-xmos_ai_tools/runtime/include/lib_nn/api/quadratic_approximation.h,sha256=Enia9kZgPM38oToGmRMJqIPOn1ZD0zeOJ8rwbL3Mrec,2699
+xmos_ai_tools/runtime/include/lib_nn/api/quadratic_approximation.h,sha256=VnyV5igZCQml-SjUKg8fRydlZFzqjaIK0OOQU35gi7U,2713
 xmos_ai_tools/runtime/include/lib_nn/api/quadratic_interpolation.h,sha256=qNguTb1oiAJ_n1_wnSFXD4BEE8PTJqePph6FVnkrwac,816
 xmos_ai_tools/runtime/include/lib_nn/api/version.h,sha256=P0otpsAVvhiPNpXQu6uJUrjygoOTaHEJlIegIMyj_W4,303
 xmos_ai_tools/runtime/include/lib_nn/api/vpu_sim.h,sha256=3S1_2eSaEj2q8nhFuRiCefQhwTeYpfothuFu0fC54_s,3811
@@ -70,7 +70,7 @@ xmos_ai_tools/runtime/include/lib_tflite_micro/api/xcore_device_memory.h,sha256=
 xmos_ai_tools/runtime/include/lib_tflite_micro/api/xcore_shared_config.h,sha256=qv3cxHGUHDxdR0xlfdd0qWDOd4V0vwPkmYEVka_j6xw,1015
 xmos_ai_tools/runtime/include/lib_tflite_micro/src/thread_call.h,sha256=KNG-3gNWMSonzYTGbk4L0kT8WU9JD0bOzlk8naz7JO0,4965
 xmos_ai_tools/runtime/include/lib_tflite_micro/src/tflite-xcore-kernels/conv2d_float.h,sha256=XfVVWAKhbv6nByOzO1j6WA4oQ4RTX3PvYx22f-_3JFA,7681
-xmos_ai_tools/runtime/include/lib_tflite_micro/src/tflite-xcore-kernels/xcore_common.h,sha256=tpv6t6mDe5lR2gt_DRsl58RkOcbqBCh7XA_z7qYL48M,358
+xmos_ai_tools/runtime/include/lib_tflite_micro/src/tflite-xcore-kernels/xcore_common.h,sha256=zLIvEU3IrdgACea8jrtNGzWzeLc8SpqyciLyR9NqoLA,359
 xmos_ai_tools/runtime/include/lib_tflite_micro/src/tflite-xcore-kernels/xcore_custom_options.h,sha256=8iD9kjYqpaGpAUfGzfYk7OaPZo1ewpL5bBnZECE6jEg,618
 xmos_ai_tools/runtime/include/lib_tflite_micro/src/tflite-xcore-kernels/xcore_error_reporter.h,sha256=fnSTiuJEL5cstbgtiM9rKwpCCNPHFEZr5b_Qu4PY1Yo,794
 xmos_ai_tools/runtime/include/lib_tflite_micro/src/tflite-xcore-kernels/xcore_interpreter.h,sha256=5-_cbV1uXQDTW51l3F3p06fZB4qx19Z848h9qtoAjE4,1753
@@ -369,17 +369,17 @@ xmos_ai_tools/runtime/include/tensorflow/lite/micro/tools/benchmarking/metrics.h
 xmos_ai_tools/runtime/include/tensorflow/lite/micro/tools/benchmarking/op_resolver.h,sha256=RGL4qxRe7mT-DjxD0NiFhRvVpuDn5xIuwXdrj6I9avs,6053
 xmos_ai_tools/runtime/include/tensorflow/lite/schema/schema_generated.h,sha256=V6KZsZ1QF0xtjxBnXJbk32hvtWYnaBkuURjk_Hs9ivs,1073395
 xmos_ai_tools/runtime/include/tensorflow/lite/schema/schema_utils.h,sha256=Iyv5Rn__Q15pfUm-9DvFU_GFF_8rplBNwor-V5TUi_0,1321
-xmos_ai_tools/runtime/lib/libhost_xtflitemicro.a,sha256=TYRLD-6G08aOCu9s4dbN0skCvw_ON0jS306o53LS42E,956112
-xmos_ai_tools/runtime/lib/libxtflitemicro.a,sha256=XynDOL43IjoGV0pM6S1qxJRnr5in7LMXVbVuC-gP5mY,68177756
+xmos_ai_tools/runtime/lib/libhost_xtflitemicro.a,sha256=ku_d9ivwxZH1Y-HDePNZjXtqshKWkCrFOdskBMYPY1A,958088
+xmos_ai_tools/runtime/lib/libxtflitemicro.a,sha256=zh4DPLFkseuLziV5BwkwI_pyHYs-C5x-GwaVzIPHpKA,68105316
 xmos_ai_tools/xformer/__init__.py,sha256=jA0xba6ZitK9HwWvDJewM8AMU4IZnx_iB1GSBrRUGvU,1627
 xmos_ai_tools/xformer/flash.py,sha256=MG4coi_Lvvg-oQmw1pomJD8eeOH4gAMjixjBFvO2BCk,6376
 xmos_ai_tools/xinterpreters/__init__.py,sha256=PFRB9VxOLKaA--j2ZvWGcmesv2C6uNYqJ_kBam68aUI,50
 xmos_ai_tools/xinterpreters/exceptions.py,sha256=HOjADxHYMPI9mN0YIbWxtw9hSeL2B6XWWwqtGtyJdVs,577
 xmos_ai_tools/xinterpreters/host_interpreter.py,sha256=vzRXpXwoObdbR7x-lnmfAxU5jHbneUead0BsKbT9BbI,25174
-xmos_ai_tools/xinterpreters/libs/macos/xtflm_python.1.0.1.dylib,sha256=LPayYyTG-E9MXQTvAZmGsgZgBIOwMwjfLX8iaBlrBP4,945205
-xmos_ai_tools/xinterpreters/libs/macos/xtflm_python.dylib,sha256=LPayYyTG-E9MXQTvAZmGsgZgBIOwMwjfLX8iaBlrBP4,945205
-xmos_ai_tools-1.1.2.dev263.data/data/bin/xcore-opt,sha256=xKfVP3qS5C5RHCxJPQE4BQL1sj5g18CLvrBPOYLDJ0c,143883590
-xmos_ai_tools-1.1.2.dev263.dist-info/METADATA,sha256=8VEZN9bW5dfyM_IoEmlJuZjyBKdM_d2JKwrcAGCypN4,1363
-xmos_ai_tools-1.1.2.dev263.dist-info/WHEEL,sha256=IPDqJGqJLX0u0WXkppbx2qlPtK5mpK7tD_Q6j3xfzzw,107
-xmos_ai_tools-1.1.2.dev263.dist-info/top_level.txt,sha256=YWegea73ll3tMlRWRdHJemUy2VOuEYDdOIaffxu_eF0,14
-xmos_ai_tools-1.1.2.dev263.dist-info/RECORD,,
+xmos_ai_tools/xinterpreters/libs/macos/xtflm_python.1.0.1.dylib,sha256=U5sPzZPXtmrN4CsA4JWO33TukTydAvnm7h5QkQ9lXPM,945605
+xmos_ai_tools/xinterpreters/libs/macos/xtflm_python.dylib,sha256=U5sPzZPXtmrN4CsA4JWO33TukTydAvnm7h5QkQ9lXPM,945605
+xmos_ai_tools-1.1.2.dev276.data/data/bin/xcore-opt,sha256=1bJ7eNJG8fg6qIHAWUzAghTfpPYm-39f6w0ZvdP_6Kg,143883574
+xmos_ai_tools-1.1.2.dev276.dist-info/METADATA,sha256=ZKYWNfX8fXE1rVnYK8P-ksHp7XD0U7_FyOLHTQcYkAE,1363
+xmos_ai_tools-1.1.2.dev276.dist-info/WHEEL,sha256=IPDqJGqJLX0u0WXkppbx2qlPtK5mpK7tD_Q6j3xfzzw,107
+xmos_ai_tools-1.1.2.dev276.dist-info/top_level.txt,sha256=YWegea73ll3tMlRWRdHJemUy2VOuEYDdOIaffxu_eF0,14
+xmos_ai_tools-1.1.2.dev276.dist-info/RECORD,,