From ee2566914d3476b8103b88915f3b81bda8490b44 Mon Sep 17 00:00:00 2001
From: James Ward <james.ward@arm.com>
Date: Tue, 15 Nov 2022 11:36:47 +0000
Subject: FP16 improvements

* Update FP16 resize to newest spec version
* Correct casting to fp16 for graphs of >1 ops

Change-Id: Iedff9a71eb7f72948b3c00a635bb0fd07d414bcd
Signed-off-by: James Ward <james.ward@arm.com>
---
 reference_model/src/arith_util.h | 15 ++++++++++-----
 1 file changed, 10 insertions(+), 5 deletions(-)

(limited to 'reference_model/src/arith_util.h')

diff --git a/reference_model/src/arith_util.h b/reference_model/src/arith_util.h
index 33bdeed..a75d7a3 100644
--- a/reference_model/src/arith_util.h
+++ b/reference_model/src/arith_util.h
@@ -30,17 +30,18 @@
 #include <fenv.h>
 #include <math.h>
 #define __STDC_LIMIT_MACROS    //enable min/max of plain data type
-#include "func_debug.h"
 #include "func_config.h"
+#include "func_debug.h"
+#include "half.hpp"
 #include "inttypes.h"
 #include "tosa_generated.h"
+#include <Eigen/Core>
+#include <bitset>
 #include <cassert>
 #include <iostream>
 #include <limits>
 #include <stdint.h>
 #include <typeinfo>
-#include <Eigen/Core>
-#include <bitset>
 
 using namespace tosa;
 using namespace std;
@@ -269,8 +270,12 @@ float fpTrunc(float f_in)
             truncateFloatToBFloat(&f_in, 1);
             break;
         case DType_FP16:
-            // TODO(jw): implement FP16 truncate function (no-op placeholder for now)
-            break;
+            // Cast to temporary float16 value before casting back to float32
+            {
+                half_float::half h = half_float::half_cast<half_float::half, float>(f_in);
+                f_in               = half_float::half_cast<float, half_float::half>(h);
+                break;
+            }
         case DType_FP32:
             // No-op for fp32
             break;
-- 
cgit v1.2.1