code cleanup: fix for unit tests

openvinotoolkit · Dec 1, 2024 · 5ca8940 · 5ca8940
1 parent d96c9ed
commit 5ca8940
Show file tree

Hide file tree

Showing 4 changed files with 25 additions and 20 deletions.
diff --git a/src/plugins/intel_gpu/include/intel_gpu/primitives/dynamic_quantize.hpp b/src/plugins/intel_gpu/include/intel_gpu/primitives/dynamic_quantize.hpp
@@ -27,7 +27,7 @@ struct dynamic_quantize : public primitive_base<dynamic_quantize> {
     dynamic_quantize(const primitive_id& id,
            const input_info& input,
            const Attributes& attrs,
-           const size_t input_size = 2)
+           const size_t input_size = 3)
            : primitive_base(id, {input})
            , attrs(attrs)
            , input_size(input_size) {

diff --git a/src/plugins/intel_gpu/src/kernel_selector/cl_kernels/dynamic_quantize_gpu_ref.cl b/src/plugins/intel_gpu/src/kernel_selector/cl_kernels/dynamic_quantize_gpu_ref.cl
@@ -6,6 +6,14 @@
 
 #define UINT64_MAX 0xFFFFFFFFFFFFFFFF
 
+#if ASYMMETRIC_QUANTIZATION && UNSIGNED_OUTPUT
+    #define TO_OUTPUT_TYPE_RTE(val)  convert_uchar_rte(val)
+    #define TO_OUTPUT_VEC_TYPE_RTE(val)  convert_uchar8_rte(val)
+#else
+    #define TO_OUTPUT_TYPE_RTE(val)  convert_char_rte(val)
+    #define TO_OUTPUT_VEC_TYPE_RTE(val)  convert_char8_rte(val)
+#endif
+
 #if OUTPUT_DIMS != 4
 #error "dynamic_quantize_gpu_ref.cl: Unsupported output dimension"
 #endif
@@ -93,8 +101,12 @@ KERNEL(dynamic_quantize_gpu_ref)(
 #if ASYMMETRIC_QUANTIZATION
     // need to support output data precision of i8 and u8
     OUTPUT1_TYPE scale = (OUTPUT1_TYPE)((CHAR_MAX - CHAR_MIN) / (max_val - min_val));
+#   if UNSIGNED_OUTPUT
     OUTPUT1_TYPE zp = (OUTPUT1_TYPE)(-min_val * scale);
-#else
+#   else // SIGNED_OUTPUT
+    OUTPUT1_TYPE zp = (OUTPUT1_TYPE)(-min_val * scale) - CHAR_MAX;
+#   endif
+#else  // !ASYMMETRIC_QUANTIZATION
     max_val = work_group_reduce_max(max_val);
     OUTPUT1_TYPE scale = 127.0h / max_val;
 #endif
@@ -107,43 +119,31 @@ KERNEL(dynamic_quantize_gpu_ref)(
         const uint out_offset = OUTPUT_GET_INDEX(b + b_off, f + f_off, y + y_off, x);
 
         half val = input[in_offset];
-#if ASYMMETRIC_QUANTIZATION
         val *= scale;
+#if ASYMMETRIC_QUANTIZATION
         val += zp;
-        // printf("1 val %f\n", val);
-        output[out_offset] = convert_uchar_rte(val);
-#else
-        val *= scale;
-        output[out_offset] = convert_char_rte(val);
 #endif
+        output[out_offset] = TO_OUTPUT_TYPE_RTE(val);
 #else
         const uint in_offset = INPUT0_GET_INDEX(b + b_off, f + f_off, y + y_off, 0);
         const uint out_offset = OUTPUT_GET_INDEX(b + b_off, f + f_off, y + y_off, 0);
         int x;
         for (x = 0; x < INPUT0_SIZE_X / 8; x++) {
             half8 val = as_half8(vload8(0, (ushort*)input + in_offset + x * 8));
-#if ASYMMETRIC_QUANTIZATION
             val *= scale;
+#if ASYMMETRIC_QUANTIZATION
             val += zp;
-            // printf("2 val %f\n", val);
-            vstore8(convert_uchar8_rte(val), 0, output + out_offset + x * 8);
-#else
-            val *= scale;
-            vstore8(convert_char8_rte(val), 0, output + out_offset + x * 8);
 #endif
+            vstore8(TO_OUTPUT_VEC_TYPE_RTE(val), 0, output + out_offset + x * 8);
         }
         x *= 8;
         for (; x < INPUT0_SIZE_X; x++) {
             half val = input[in_offset + x];
-#if ASYMMETRIC_QUANTIZATION
             val *= scale;
+#if ASYMMETRIC_QUANTIZATION
             val += zp;
-            // printf("3 val %f zp %f\n", val, zp);
-            output[out_offset + x] = convert_uchar_rte(val);
-#else
-            val *= scale;
-            output[out_offset + x] = convert_char_rte(val);
 #endif
+            output[out_offset + x] = TO_OUTPUT_TYPE_RTE(val);
         }
 #endif
     }

diff --git a/...ns/intel_gpu/src/kernel_selector/kernels/dynamic_quantize/dynamic_quantize_kernel_ref.cpp b/...ns/intel_gpu/src/kernel_selector/kernels/dynamic_quantize/dynamic_quantize_kernel_ref.cpp
@@ -54,6 +54,7 @@ JitConstants DynamicQuantizeKernelRef::GetJitConstants(const dynamic_quantize_pa
 
     jit.AddConstant(MakeJitConstant("ASYMMETRIC_QUANTIZATION", params.use_asymmetric_quantization));
     jit.AddConstant(MakeJitConstant("GROUP_SCALES_WITH_ZP", params.combine_scales_and_zp));
+    jit.AddConstant(MakeJitConstant("UNSIGNED_OUTPUT", params.outputs[0].GetDType() == Datatype::UINT8 ? 1 : 0));
 
     auto group_sizes = params.group_sizes;
     group_sizes.resize(std::min((size_t)4, group_sizes.size()), 1);

diff --git a/src/plugins/intel_gpu/tests/unit/test_cases/dynamic_quantize_gpu_test.cpp b/src/plugins/intel_gpu/tests/unit/test_cases/dynamic_quantize_gpu_test.cpp
@@ -147,6 +147,10 @@ TEST_F(dynamic_quantization_gpu_tests, simple_quantizing_large_size) {
     this->test_dynamic_quantization(false, {11, 1, 1, 4096}, {2048, 1, 1, 4096});
 }
 
+TEST_F(dynamic_quantization_gpu_tests, simple_quantizing_large_size_dynamic) {
+    this->test_dynamic_quantization(false, {-1, 1, 1, 4096}, {2048, 1, 1, 4096});
+}
+
 TEST_F(dynamic_quantization_gpu_tests, simple_quantizing_small_size) {
     this->test_dynamic_quantization(false, {1, 1, 1, 4096}, {64, 1, 1, 4096});
 }