Add input_signed to neureka

pulp-platform · Jan 19, 2024 · d6ba620 · d6ba620
1 parent dc2409c
commit d6ba620
Show file tree

Hide file tree

Showing 9 changed files with 128 additions and 8 deletions.
diff --git a/neureka/hal/neureka_task.c b/neureka/hal/neureka_task.c
@@ -47,10 +47,9 @@ void neureka_task_init(neureka_task_t *task, const uint8_t kernel_shape,
                        const neureka_weight_offset_mode_e weights_offset_mode,
                        const uint32_t weights_offset_factor,
                        neureka_quant_t quant, neureka_norm_t norm,
-                       const uint8_t stride) {
+                       const uint8_t flag_input_signed) {
   *task = (neureka_task_t){.outbytes = output_bits / 8,
                            .qw = weights_bits,
-                           .stride_shift = stride == 2 ? 1 : 0,
                            .output_channel_throughput =
                                depthwise ? NEUREKA_INPUT_CHANNEL_THROUGHPUT_3x3
                                          : NEUREKA_OUTPUT_CHANNEL_THROUGHPUT,
@@ -67,6 +66,7 @@ void neureka_task_init(neureka_task_t *task, const uint8_t kernel_shape,
                                           : NEUREKA_FLAG_MODE_3x3;
 
   task->data.cfg.conf0 |=
+      flag_input_signed << NEUREKA_SHIFT_FLAG_INPUT_SIGNED |
       NEUREKA_FLAG_NORM_QUANT | quant.function | quant.mode |
       (quant.shift_amount << 16) |
       quant.flag_rounding << NEUREKA_SHIFT_ROUNDING | norm.mode |

diff --git a/neureka/hal/neureka_task.h b/neureka/hal/neureka_task.h
@@ -111,9 +111,7 @@ typedef struct neureka_task_data_t {
 typedef struct neureka_task_t {
   neureka_task_data_t data;
   uint8_t outbytes;
-  uint8_t weight_d0_stride;
   uint8_t qw;
-  uint8_t stride_shift;
   uint8_t output_channel_throughput;
   uint8_t input_channel_throughput;
   uint8_t kernel_shape;
@@ -127,7 +125,7 @@ void neureka_task_init(neureka_task_t *task, const uint8_t kernel_shape,
                        const neureka_weight_offset_mode_e weights_offset_mode,
                        const uint32_t weights_offset_factor,
                        neureka_quant_t quant, neureka_norm_t norm,
-                       const uint8_t stride);
+                       const uint8_t flag_input_signed);
 uint32_t neureka_get_tile_padding(uint32_t padding, uint32_t i_height,
                                   uint32_t i_width, uint32_t n_height,
                                   uint32_t n_width);

diff --git a/neureka/hal/neureka_task_defs.h b/neureka/hal/neureka_task_defs.h
@@ -61,14 +61,15 @@
 
 /*  SHIFT  */
 
+#define NEUREKA_SHIFT_FLAG_INPUT_SIGNED (26)
 #define NEUREKA_SHIFT_FLAG_NORM_BIAS (25)
 #define NEUREKA_SHIFT_FLAG_NORM_SHIFT (24)
 #define NEUREKA_SHIFT_QUANT_SHIFT (16)
 #define NEUREKA_SHIFT_ROUNDING (11)
 
 /*  CONF0 FLAGS */
 
-#define NEUREKA_FLAG_SIGNED_ACTIVATION (1 << 26)
+#define NEUREKA_FLAG_INPUT_SIGNED (1 << 26)
 #define NEUREKA_FLAG_NORM_BIAS (1 << 25)
 #define NEUREKA_FLAG_NORM_SHIFT (1 << 24)
 #define NEUREKA_FLAG_QUANT_FUNCTION_IDENTITY (1 << 23)

diff --git a/test/NeurekaTestConf.py b/test/NeurekaTestConf.py
@@ -50,7 +50,7 @@ def _check_type(
     @field_validator("in_type")
     @classmethod
     def check_valid_in_type(cls, v: IntegerType) -> IntegerType:
-        NeurekaTestConf._check_type("in_type", v, ["uint8"])
+        NeurekaTestConf._check_type("in_type", v, ["uint8", "int8"])
         return v
 
     @field_validator("out_type")

diff --git a/test/app/src/nnx_layer.c b/test/app/src/nnx_layer.c
@@ -113,7 +113,12 @@ static void task_prepare(nnx_task_t *task) {
       (nnx_norm_t){.mode = normMode8Bit,
                    .flag_bias = HAS_BIAS ? nnxTaskFlagTrue : nnxTaskFlagFalse,
                    .flag_shift = nnxTaskFlagFalse},
-      STRIDE_HEIGHT);
+#ifdef NNX_NE16
+      STRIDE_HEIGHT
+#elif NNX_NEUREKA
+      INPUT_SIGNED
+#endif
+  );
 
 #if STRIDE_HEIGHT == 2 && STRIDE_WIDTH == 2
   nnx_task_set_dims_stride2x2(

diff --git a/test/tests/test_102/conf.json b/test/tests/test_102/conf.json
@@ -0,0 +1,29 @@
+{
+    "in_height": 4,
+    "in_width": 3,
+    "in_channel": 8,
+    "out_channel": 8,
+    "padding": {
+        "top": 0,
+        "bottom": 0,
+        "left": 0,
+        "right": 0
+    },
+    "kernel_shape": {
+        "height": 3,
+        "width": 3
+    },
+    "depthwise": false,
+    "stride": {
+        "height": 1,
+        "width": 1
+    },
+    "in_type": "int8",
+    "out_type": "uint8",
+    "weight_type": "int8",
+    "scale_type": "uint8",
+    "bias_type": "int32",
+    "has_norm_quant": true,
+    "has_bias": true,
+    "has_relu": true
+}
diff --git a/test/tests/test_103/conf.json b/test/tests/test_103/conf.json
@@ -0,0 +1,29 @@
+{
+    "in_height": 20,
+    "in_width": 15,
+    "in_channel": 40,
+    "out_channel": 25,
+    "padding": {
+        "top": 0,
+        "bottom": 0,
+        "left": 0,
+        "right": 0
+    },
+    "kernel_shape": {
+        "height": 3,
+        "width": 3
+    },
+    "depthwise": false,
+    "stride": {
+        "height": 1,
+        "width": 1
+    },
+    "in_type": "int8",
+    "out_type": "uint8",
+    "weight_type": "int8",
+    "scale_type": "uint8",
+    "bias_type": "int32",
+    "has_norm_quant": true,
+    "has_bias": true,
+    "has_relu": true
+}
diff --git a/test/tests/test_104/conf.json b/test/tests/test_104/conf.json
@@ -0,0 +1,29 @@
+{
+    "in_height": 20,
+    "in_width": 15,
+    "in_channel": 40,
+    "out_channel": 25,
+    "padding": {
+        "top": 0,
+        "bottom": 0,
+        "left": 0,
+        "right": 0
+    },
+    "kernel_shape": {
+        "height": 1,
+        "width": 1
+    },
+    "depthwise": false,
+    "stride": {
+        "height": 1,
+        "width": 1
+    },
+    "in_type": "int8",
+    "out_type": "uint8",
+    "weight_type": "int8",
+    "scale_type": "uint8",
+    "bias_type": "int32",
+    "has_norm_quant": true,
+    "has_bias": true,
+    "has_relu": true
+}
diff --git a/test/tests/test_105/conf.json b/test/tests/test_105/conf.json
@@ -0,0 +1,29 @@
+{
+    "in_height": 20,
+    "in_width": 15,
+    "in_channel": 40,
+    "out_channel": 40,
+    "padding": {
+        "top": 0,
+        "bottom": 0,
+        "left": 0,
+        "right": 0
+    },
+    "kernel_shape": {
+        "height": 3,
+        "width": 3
+    },
+    "depthwise": true,
+    "stride": {
+        "height": 1,
+        "width": 1
+    },
+    "in_type": "int8",
+    "out_type": "uint8",
+    "weight_type": "int8",
+    "scale_type": "uint8",
+    "bias_type": "int32",
+    "has_norm_quant": true,
+    "has_bias": true,
+    "has_relu": true
+}