xmos-ai-tools 1.2.1.dev24__py3-none-macosx_11_0_arm64.whl → 1.3.0__py3-none-macosx_11_0_arm64.whl

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

@@ -109,50 +109,6 @@ typedef struct padding_sizes_t {
   int32_t right;
 } padding_sizes_t;
 
-/**
- * @brief Execute @oper{pad_prepare} function.
- *
- * `plan` points to the output vector @tensor{y} with length @math{N}.
- *
- * `p` struct describing the padding to be applied to the input tensor.
- *
- * `x` parameters describing the input tensor to be padded.
- *
- * `bytes_per_pixel` the bytes per pixel for tensor x.
- *
- * @param plan             [out]  The output vector @tensor{y}
- * @param p                [in]   The input vector @tensor{x}
- * @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);
-
-/**
- * @brief Execute @oper{pad_run} job.
- *
- * See @oper_ref{pad_run} for more details about the @oper{requantize_16_to_8}
- * operator.
- *
- * `Y` points to the output vector @tensor{y}.
- *
- * `X` points to the input vector @tensor{x}.
- *
- * `plan` points to the (initialized) plan.
- *
- * @requires_word_alignment{Y,X}
- *
- * @param y   [out]    The output vector @tensor{y}
- * @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_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
  */
@@ -336,12 +292,7 @@ void softmax_generate_exp_lut(int zero_point, float scale, float *lut);
 void softmax_ref(int8_t *Y, const int8_t *X, const float zero_point,
                  const float scale, const int length);
 
-void slice_memcpy(int8_t *dst, int8_t *src, int32_t *in_offsets,
-                  int32_t *out_offsets, int32_t *begin, int32_t *end,
-                  void (*memcpy_func)(void *, void *, size_t));
+void softmax_single(int8_t *Y, const int8_t *X, const float *lut,
+                    const int offset);
 
-void slice_memcpy_get_params(int *begin_dst, int *end_dst, int *in_offsets,
-                             int *out_offsets, int *shape_dst, const int *begin,
-                             const int *size, const int *shape,
-                             const int dtype_size, const int rank);
 #endif // LAYERS_H_

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

@@ -0,0 +1,15 @@
+#ifndef _vpu_memmove_word_aligned_h_
+#define _vpu_memmove_word_aligned_h_
+
+/**
+ * Function that copies a block of memory. Both source and destination
+ * address must be word aligned. Any number of bytes can be copied. There
+ * may be an overlap between the destination and source.
+ *
+ * @param     dst         Destination address, must be word aligned.
+ * @param     src         Source address, must be word aligned.
+ * @param     byte_count  Number of bytes to copy - may be zero
+ */
+void vpu_memmove_word_aligned(void * dst, const void * src, unsigned int byte_count);
+
+#endif

xmos_ai_tools/runtime/include/lib_nn/api/vpu_memset_256.h

@@ -0,0 +1,53 @@
+#ifndef _vpu_memset_256_h_
+#define _vpu_memset_256_h_
+
+/**
+ * Function that replicates a vector. The source address must be word
+ * aligned, the destination address is assumed to be aligned with the
+ * replication pattern in the source. Any number of bytes can be copied.
+ * There should not be an overlap between the destination and source.
+ *
+ * It is assumed that the source address contains 32 replicated bytes (if
+ * the destination address is byte aligned), or that it contains 16
+ * replicated shorts (if the destination address is 16-bit aligned), or
+ * that it contains 8 replicated ints.
+ *
+ * broadcast_32_to_256() and BROADCAST_8_TO_32() cane be used to
+ * create the source vector
+ *
+ * @param     dst         Destination address
+ * @param     src         Source address, must be word aligned.
+ * @param     byte_count  Number of bytes to copy - may be zero
+ */
+void vpu_memset_256(void *dst, const void *src, unsigned int byte_count);
+
+/**
+ * Function that replicates an int over a vector. The vector must be
+ * aligned on an 8-byte boundary. In order to replicate a byte or short over
+ * a vector, combine this with a call to BROADCAST_8_TO_32() or
+ * BROADCAST_16_TO_32(). Declare the vector as a uint64_t x[] in order to
+ * guarantee 8-byte alignement.
+ *
+ * @param     dst         Destination address, must be 8-byte aligned
+ * @param     from        Value to be replicated
+ */
+void broadcast_32_to_256(void *dst, uint32_t from);
+
+/**
+ * Macro that replicates a byte over an int.
+ * Use with broadcast_32_to_256() in order to replicate a byte over a vector
+ */
+#define BROADCAST_8_TO_32(f) (((uint8_t)f) * 0x01010101)
+
+/**
+ * Macro that replicates a short over an int
+ * Use with broadcast_32_to_256() in order to replicate a short over a vector
+ */
+#define BROADCAST_16_TO_32(f) (((uint16_t)f) * 0x00010001)
+
+/**
+ * Macro that replicates a byte over a short
+ */
+#define BROADCAST_8_TO_16(f) (((uint8_t)f) * 0x00000101)
+
+#endif

xmos_ai_tools/runtime/include/lib_tflite_micro/api/inference_engine.h

@@ -175,17 +175,17 @@ void inference_engine_unload_model(inference_engine_t *UNSAFE ie);
  * \param ie           pointer to inference engine.
  * \param model_bytes  Number of bytes in the model
  * \param model_data   Pointer to the model (one of data_model or
- * data_tensor_arena passed above) \param c_flash      Optional channel to flash
- * server, to be used for fetching parameter blocks
+ * data_tensor_arena passed above) \param c_flash_or_tile      Optional channel to flash
+ * or tile server
  *
  * \returns            non zero indicates an error
  */
 #ifdef __XC__
-    int inference_engine_load_model(inference_engine_t * UNSAFE ie, uint32_t model_bytes, uint32_t * UNSAFE model_data, chanend ?c_flash);
+    int inference_engine_load_model(inference_engine_t * UNSAFE ie, uint32_t model_bytes, uint32_t * UNSAFE model_data, chanend ?c_flash_or_tile);
 #else
 int inference_engine_load_model(inference_engine_t *UNSAFE ie,
                                 uint32_t model_bytes,
-                                uint32_t *UNSAFE model_data, void *flash_data);
+                                uint32_t *UNSAFE model_data, void *weights_data_ptr);
 #endif
 
     /** Function that invokes the inference engine. This function will create an

xmos_ai_tools/runtime/include/lib_tflite_micro/api/xcore_config.h

@@ -11,7 +11,7 @@ struct xc_context_config_t {
   // calculated in the compiler.
   int model_thread_count;
   thread_info_t thread_info;
-  void *UNSAFE flash_data; // channel to flash reader.
+  void *UNSAFE weights_data_ptr; // DDR ptr or channel to flash/tile server.
 };
 
 #endif // XCORE_CONFIG_H_

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

@@ -21,12 +21,14 @@ constexpr const char *XC_unaryi16_OpCode = "XC_unaryi16";
 constexpr const char *XC_conv2d_v2_OpCode = "XC_conv2d_v2";
 constexpr const char *XC_maxpool2d_OpCode = "XC_maxpool2d";
 constexpr const char *XC_softmax_OpCode = "XC_softmax";
-constexpr const char *XC_ld_flash_OpCode = "XC_ld_flash";
+constexpr const char *XC_batched_softmax_OpCode = "XC_batched_softmax";
+constexpr const char *XC_ld_weights_OpCode = "XC_ld_weights";
 constexpr const char *XC_add_OpCode = "XC_add";
 constexpr const char *XC_slice_OpCode = "XC_slice";
+constexpr const char *XC_broadcast_OpCode = "XC_broadcast";
 constexpr const char *XC_lookup_OpCode = "XC_lookup";
 constexpr const char *XC_pad_OpCode = "XC_pad";
-constexpr const char *XC_pad_v2_OpCode = "XC_pad_v2";
+constexpr const char *XC_concat_OpCode = "XC_concat";
 constexpr const char *XC_pad_3_to_4_OpCode = "XC_pad_3_to_4";
 constexpr const char *XC_mul_OpCode = "XC_mul";
 // Binarized ops
@@ -44,12 +46,14 @@ TFLMRegistration *Register_XC_unaryi16();
 TFLMRegistration *Register_XC_conv2d_v2();
 TFLMRegistration *Register_XC_maxpool2d();
 TFLMRegistration *Register_XC_softmax();
-TFLMRegistration *Register_XC_ld_flash();
+TFLMRegistration *Register_XC_batched_softmax();
+TFLMRegistration *Register_XC_ld_weights();
 TFLMRegistration *Register_XC_add();
 TFLMRegistration *Register_XC_slice();
+TFLMRegistration *Register_XC_broadcast();
 TFLMRegistration *Register_XC_lookup();
 TFLMRegistration *Register_XC_pad();
-TFLMRegistration *Register_XC_pad_v2();
+TFLMRegistration *Register_XC_concat();
 TFLMRegistration *Register_XC_pad_3_to_4();
 TFLMRegistration *Register_XC_mul();
 // Binarized ops

{xmos_ai_tools-1.2.1.dev24.dist-info → xmos_ai_tools-1.3.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: xmos-ai-tools
-Version: 1.2.1.dev24
+Version: 1.3.0
 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/28b6228b76e38a2fc78c9741d323d656083dca6f/README.md) for documentation on using xmos-ai-tools to deploy AI models on xcore.
+Click [here](https://github.com/xmos/ai_tools/blob/de1c819d1109e67ed90a11d3334e851fe97f1361/README.md) for documentation on using xmos-ai-tools to deploy AI models on xcore.
 

{xmos_ai_tools-1.2.1.dev24.dist-info → xmos_ai_tools-1.3.0.dist-info}/RECORD

@@ -49,7 +49,7 @@ 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=JZx4pkwoJrvU3YtNPvoDZLnOHf01ci7-QfNqViQp0W4,12140
+xmos_ai_tools/runtime/include/lib_nn/api/nn_layers.h,sha256=NJXmlCSXA5UzK0dvJhNADqmrAvmNzvp0kbn9NLQyU_Q,10257
 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=iTZcsp941d4fT574C_aUsHDAkLAWYxoFM4EjW7ypm94,4741
 xmos_ai_tools/runtime/include/lib_nn/api/nn_operator.h,sha256=Nj4jiXVHlnFjVBtV7mCGulTzSyNwCIpitz8Ao9kSIsU,337
@@ -64,16 +64,18 @@ xmos_ai_tools/runtime/include/lib_nn/api/quadratic_interpolation.h,sha256=qNguTb
 xmos_ai_tools/runtime/include/lib_nn/api/quantize_int16.h,sha256=THKX9qg6ob1Cj4LiRmqlogaEAozc2qEIiENBmRS8rgs,736
 xmos_ai_tools/runtime/include/lib_nn/api/quantize_int16_transform.h,sha256=Rg2XOvVL1P2UneJ33X9b-ATwaCIiNAyp_ipdAu1SXyk,1029
 xmos_ai_tools/runtime/include/lib_nn/api/version.h,sha256=P0otpsAVvhiPNpXQu6uJUrjygoOTaHEJlIegIMyj_W4,303
+xmos_ai_tools/runtime/include/lib_nn/api/vpu_memmove_word_aligned.h,sha256=-OgMM0JMG1kIlL6YSiEg-6o6t43t1LUDwAS_inW-r_8,579
+xmos_ai_tools/runtime/include/lib_nn/api/vpu_memset_256.h,sha256=Cb93K-ie4Dp-lqCW0rjzKrIBnEGlKf9xrRPck67sJ1o,1946
 xmos_ai_tools/runtime/include/lib_nn/api/vpu_sim.h,sha256=3S1_2eSaEj2q8nhFuRiCefQhwTeYpfothuFu0fC54_s,3811
 xmos_ai_tools/runtime/include/lib_nn/api/xs3_vpu.h,sha256=xuEpNayaSWeEIoL1--EfQAXkr4TidwDASFYgbshJ-Mw,6227
 xmos_ai_tools/runtime/include/lib_nn/api/xs3a_registers.h,sha256=5i6Tue1VplRI2VlIes-gaD2BXOKreuUit51OYTsbgFs,139774
 xmos_ai_tools/runtime/include/lib_nn/src/asm/asm_constants.h,sha256=VB8pk_H4vI4S18xYfPFCOGZoLG6S0FYW5uVfG3u9glU,1018
 xmos_ai_tools/runtime/include/lib_nn/src/asm/window_op_plan.h,sha256=4I5u3jkbPOXQnj073h24346uJTYApf0A2oegNc0TKjc,704
 xmos_ai_tools/runtime/include/lib_tflite_micro/api/fast_flash.h,sha256=IFDYtVg3AodzlFEQw_MFWs2fNQ43aMEXNVwpaG27sko,1984
-xmos_ai_tools/runtime/include/lib_tflite_micro/api/inference_engine.h,sha256=yX_NVE1dsAhQ67uj3Qz5DiY7oSQG6n3RDgjKN16TJPk,9261
+xmos_ai_tools/runtime/include/lib_tflite_micro/api/inference_engine.h,sha256=qSNg6rIzz-dKdIUoxk9VnqyxoypO4Aqf5h06K7rM82k,9249
 xmos_ai_tools/runtime/include/lib_tflite_micro/api/memory_parallel_transport.h,sha256=P6o4-yWfE3GW_R08zf_kTsg-h4589eAhg9lNvJA7ZCM,1932
 xmos_ai_tools/runtime/include/lib_tflite_micro/api/version.h,sha256=zwy-Yay7mJg09OQpzKyNfITtjfe0Jy3IgErrPdH31I0,318
-xmos_ai_tools/runtime/include/lib_tflite_micro/api/xcore_config.h,sha256=KToE82wnv1DUW72bCeiWGuW3uNsvyu9yURXLjs3OLTM,497
+xmos_ai_tools/runtime/include/lib_tflite_micro/api/xcore_config.h,sha256=6KOImWQXzY6FXIepK746QlkQllmCo3eH5FD3Vp1x7PQ,519
 xmos_ai_tools/runtime/include/lib_tflite_micro/api/xcore_device_memory.h,sha256=bLWcRDNrzClLh8_eR3XRRz3sA2pEAzzxGLDoPTsNp8A,1917
 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
@@ -82,7 +84,7 @@ xmos_ai_tools/runtime/include/lib_tflite_micro/src/tflite-xcore-kernels/xcore_co
 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
-xmos_ai_tools/runtime/include/lib_tflite_micro/src/tflite-xcore-kernels/xcore_ops.h,sha256=7dJTjMUId_e9_L1n-fs3PMKBwGFv8tTc_mlAATMt3nU,2327
+xmos_ai_tools/runtime/include/lib_tflite_micro/src/tflite-xcore-kernels/xcore_ops.h,sha256=Po7Im0S-LaasVyFhJuJPieMO6iTdqPROexKODPodjBA,2557
 xmos_ai_tools/runtime/include/lib_tflite_micro/src/tflite-xcore-kernels/xcore_profiler.h,sha256=5wo0BpSfxebxo-NpAoJYsheULF9u_Nz3xbhjUs2K2Hg,1215
 xmos_ai_tools/runtime/include/lib_tflite_micro/src/tflite-xcore-kernels/xcore_utils.h,sha256=kynLZ8kJMx4sGQNOkIqoF7zjey1mKVeEbgdGS-UAX_o,4717
 xmos_ai_tools/runtime/include/lib_xud/lib_xud/api/xud.h,sha256=CpBh7idogG_eZXEe29i306iXbz3Jcnvd5CgptIFG8Lo,20680
@@ -377,17 +379,17 @@ 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=OS9Ri-QlojRkRKjjb4OfjDX0tRO1sF8-wblt3Ng-sN8,985832
-xmos_ai_tools/runtime/lib/libxtflitemicro.a,sha256=eRwbQz7UZ8Npo1SDbvpdSyxCRCnAZ0iC-D9cSdOrxsc,68882352
+xmos_ai_tools/runtime/lib/libhost_xtflitemicro.a,sha256=Llu9vTvZHrV-7BmqkpD0Z-rsj9EQqAzFKpRDizFPJ1M,1057704
+xmos_ai_tools/runtime/lib/libxtflitemicro.a,sha256=rYiKTYKeRhGRKX07cmE-HPpOVxWe3zY--DTlqDGXL5s,69456482
 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=No-g29_twbIaJ4S4fzU5TGpKd4WPVEE0737HtL877sw,25692
-xmos_ai_tools/xinterpreters/libs/macos/xtflm_python.1.0.1.dylib,sha256=IBVBusDfGIhGQnVkQRhS9e7EVGd-WN1lVJeLnSbau1I,965909
-xmos_ai_tools/xinterpreters/libs/macos/xtflm_python.dylib,sha256=IBVBusDfGIhGQnVkQRhS9e7EVGd-WN1lVJeLnSbau1I,965909
-xmos_ai_tools-1.2.1.dev24.data/data/bin/xcore-opt,sha256=2zLQrLVZDIgJ8r6Shl0WcMpEY1qPxaqi7wywSYKWX5w,144896582
-xmos_ai_tools-1.2.1.dev24.dist-info/METADATA,sha256=sziCvDgvhrNqNwl_EHoISq7pMgI8NnqGPziAQcMIOs0,1362
-xmos_ai_tools-1.2.1.dev24.dist-info/WHEEL,sha256=mbj6ZioQtbYSI9Enon6Nfaq9PMeblAOX3NoP8Ads9Yw,107
-xmos_ai_tools-1.2.1.dev24.dist-info/top_level.txt,sha256=YWegea73ll3tMlRWRdHJemUy2VOuEYDdOIaffxu_eF0,14
-xmos_ai_tools-1.2.1.dev24.dist-info/RECORD,,
+xmos_ai_tools/xinterpreters/libs/macos/xtflm_python.1.0.1.dylib,sha256=0xOjM-8fZr333_MACVjIxWcRHqs1MVTDN3svjm2jwpo,1055240
+xmos_ai_tools/xinterpreters/libs/macos/xtflm_python.dylib,sha256=0xOjM-8fZr333_MACVjIxWcRHqs1MVTDN3svjm2jwpo,1055240
+xmos_ai_tools-1.3.0.data/data/bin/xcore-opt,sha256=EXyY70weFXnRCGDOXxvBgCRiupHPA2IVZLAd08XIDKA,142062968
+xmos_ai_tools-1.3.0.dist-info/METADATA,sha256=T2oTA3mLLgPInhVvZfqco9MLtPsr70Y2xvUOV1w-IFo,1356
+xmos_ai_tools-1.3.0.dist-info/WHEEL,sha256=mbj6ZioQtbYSI9Enon6Nfaq9PMeblAOX3NoP8Ads9Yw,107
+xmos_ai_tools-1.3.0.dist-info/top_level.txt,sha256=YWegea73ll3tMlRWRdHJemUy2VOuEYDdOIaffxu_eF0,14
+xmos_ai_tools-1.3.0.dist-info/RECORD,,