🔖 Release 0.5.5

Author: Stefan Mach

docs/CHANGELOG.md

@@ -9,11 +9,18 @@ Versions of the IP in the same major relase are "pin-compatible" with each other
 
 
 ## [Unreleased]
+
 ### Added
 ### Changed
 ### Fixed
 
 
+## [0.5.5] - 2019-06-02
+
+### Fixed
+- UF flag handling according to IEEE754-2008 (#11)
+
+
 ## [0.5.4] - 2019-06-02
 
 ### Added

src/fpnew_cast_multi.sv

@@ -557,7 +557,7 @@ module fpnew_cast_multi #(
   end
 
   // Classification after rounding select by destination format
-  assign uf_after_round = fmt_uf_after_round[dst_fmt_q2] & ~result_true_zero; // zero is not UF
+  assign uf_after_round = fmt_uf_after_round[dst_fmt_q2];
   assign of_after_round = fmt_of_after_round[dst_fmt_q2];
 
   // Negative integer result needs to be brought into two's complement
@@ -669,7 +669,7 @@ module fpnew_cast_multi #(
     NV: src_is_int_q & (of_before_round | of_after_round), // overflow is invalid for I2F casts
     DZ: 1'b0, // no divisions
     OF: ~src_is_int_q & (~info_q2.is_inf & (of_before_round | of_after_round)), // inf casts no OF
-    UF: uf_after_round,
+    UF: uf_after_round & fp_regular_status.NX,
     NX: src_is_int_q ? (| fp_round_sticky_bits) // overflow is invalid in i2f
             : (| fp_round_sticky_bits) | (~info_q2.is_inf & (of_before_round | of_after_round))
   };

src/fpnew_f2fcast.sv

@@ -319,7 +319,7 @@ module fpnew_f2fcast #(
     NV: 1'b0, // only valid cases are handled in regular path
     DZ: 1'b0, // no divisions
     OF: ~info_a.is_inf & (of_before_round | of_after_round), // rounding can introduce new overflow
-    UF: uf_after_round,                           // true zero results don't count as underflow
+    UF: uf_after_round & regular_status.NX,                  // only inexact results raise UF
     NX: (| round_sticky_bits) | (~info_a.is_inf & (of_before_round | of_after_round))
   };
 

src/fpnew_fma.sv

@@ -616,7 +616,7 @@ module fpnew_fma #(
     NV: 1'b0, // only valid cases are handled in regular path
     DZ: 1'b0, // no divisions
     OF: of_before_round | of_after_round,         // rounding can introduce new overflow
-    UF: uf_after_round & ~result_zero,            // true zero results don't count as underflow
+    UF: uf_after_round & regular_status.NX,       // only inexact results raise UF
     NX: (| round_sticky_bits) | of_before_round | of_after_round // RS bits mean loss in precision
   };
 

src/fpnew_fma_multi.sv

@@ -741,7 +741,7 @@ module fpnew_fma_multi #(
   end
 
   // Classification after rounding select by destination format
-  assign uf_after_round = fmt_uf_after_round[dst_fmt_q2] & ~result_zero; // zero is not UF
+  assign uf_after_round = fmt_uf_after_round[dst_fmt_q2];
   assign of_after_round = fmt_of_after_round[dst_fmt_q2];
 
 
@@ -757,7 +757,7 @@ module fpnew_fma_multi #(
     NV: 1'b0, // only valid cases are handled in regular path
     DZ: 1'b0, // no divisions
     OF: of_before_round | of_after_round,         // rounding can introduce new overflow
-    UF: uf_after_round & ~result_zero,            // true zero results don't count as underflow
+    UF: uf_after_round & regular_status.NX,                      // only inexact results raise UF
     NX: (| round_sticky_bits) | of_before_round | of_after_round // RS bits mean loss in precision
   };