aboutsummaryrefslogtreecommitdiffstats
path: root/libraries
diff options
context:
space:
mode:
authorTristan Gingold <gingold@adacore.com>2015-12-20 13:56:08 +0100
committerTristan Gingold <gingold@adacore.com>2015-12-20 13:58:11 +0100
commit92ac42fcf39d5dac4529d939a49437b2a7afe4c2 (patch)
tree2ec07dca7950bdc9ab67bba23700ccc42f8c2385 /libraries
parentd3623693949f9bcfe7917f9d059e839a95843519 (diff)
downloadghdl-92ac42fcf39d5dac4529d939a49437b2a7afe4c2.tar.gz
ghdl-92ac42fcf39d5dac4529d939a49437b2a7afe4c2.tar.bz2
ghdl-92ac42fcf39d5dac4529d939a49437b2a7afe4c2.zip
openieee: complete math_real (and fix uniform).
Diffstat (limited to 'libraries')
-rw-r--r--libraries/openieee/math_real-body.vhdl155
-rw-r--r--libraries/openieee/math_real.vhdl96
2 files changed, 241 insertions, 10 deletions
diff --git a/libraries/openieee/math_real-body.vhdl b/libraries/openieee/math_real-body.vhdl
index 1d7a154af..e825a7801 100644
--- a/libraries/openieee/math_real-body.vhdl
+++ b/libraries/openieee/math_real-body.vhdl
@@ -46,22 +46,60 @@ package body MATH_REAL is
function TRUNC (X : REAL) return REAL is
begin
assert false severity failure;
- end TRUNC;
+ end;
+
+ function fmod (X, Y : REAL) return REAL;
+ attribute foreign of fmod : function is "VHPIDIRECT fmod";
+
+ function fmod (X, Y : REAL) return REAL is
+ begin
+ assert false severity failure;
+ end;
+
+ function "mod" (X, Y : REAL) return REAL
+ is
+ variable res : real;
+ begin
+ assert y /= 0.0 report "ieee.math_real.""mod"": dividend is 0.0"
+ severity failure;
+ res := fmod (x, y);
+ if res /= 0.0 then
+ if x > 0.0 xor y > 0.0 then
+ res := res + y;
+ end if;
+ end if;
+ return res;
+ end "mod";
+
+ function REALMAX (X, Y : REAL) return REAL is
+ begin
+ assert false severity failure;
+ end;
+
+ function REALMIN (X, Y : REAL) return REAL is
+ begin
+ assert false severity failure;
+ end;
procedure UNIFORM (SEED1, SEED2 : inout POSITIVE; X : out REAL)
is
variable z, k : Integer;
+ variable s1, s2 : Integer;
begin
k := seed1 / 53668;
- seed1 := 40014 * (seed1 - k * 53668) - k * 12211;
- if seed1 < 0 then
- seed1 := seed1 + 2147483563;
+ s1 := 40014 * (seed1 - k * 53668) - k * 12211;
+ if s1 < 0 then
+ seed1 := s1 + 2147483563;
+ else
+ seed1 := s1;
end if;
k := seed2 / 52774;
- seed2 := 40692 * (seed2 - k * 52774) - k * 3791;
- if seed2 < 0 then
- seed2 := seed2 + 2147483399;
+ s2 := 40692 * (seed2 - k * 52774) - k * 3791;
+ if s2 < 0 then
+ seed2 := s2 + 2147483399;
+ else
+ seed2 := s2;
end if;
z := seed1 - seed2;
@@ -72,15 +110,114 @@ package body MATH_REAL is
x := real (z) * 4.656613e-10;
end UNIFORM;
+ function SQRT (X : REAL) return REAL is
+ begin
+ assert false severity failure;
+ end;
+
+ function CBRT (X : REAL) return REAL is
+ begin
+ assert false severity failure;
+ end;
+
+ function "**" (X : INTEGER; Y : REAL) return REAL is
+ begin
+ return real (x) ** y;
+ end "**";
+
+ function "**" (X : REAL; Y : REAL) return REAL is
+ begin
+ assert false severity failure;
+ end;
+
+ function EXP (X : REAL) return REAL is
+ begin
+ assert false severity failure;
+ end;
+
+ function LOG (X : REAL) return REAL is
+ begin
+ assert false severity failure;
+ end;
+
+ function LOG2 (X : REAL) return REAL is
+ begin
+ assert false severity failure;
+ end;
+
+ function LOG10 (X : REAL) return REAL is
+ begin
+ assert false severity failure;
+ end;
+
+ function LOG (X : REAL; BASE : REAL) return REAL is
+ begin
+ return log (x) / log (base);
+ end log;
+
function SIN (X : REAL) return REAL is
begin
assert false severity failure;
- end SIN;
+ end;
function COS (X : REAL) return REAL is
begin
assert false severity failure;
- end COS;
+ end;
+
+ function TAN (X : REAL) return REAL is
+ begin
+ assert false severity failure;
+ end;
+
+ function ARCSIN (X : REAL) return REAL is
+ begin
+ assert false severity failure;
+ end;
+
+ function ARCCOS (X : REAL) return REAL is
+ begin
+ assert false severity failure;
+ end;
+
+ function ARCTAN (Y : REAL) return REAL is
+ begin
+ assert false severity failure;
+ end;
+
+ function ARCTAN (Y, X : REAL) return REAL is
+ begin
+ assert false severity failure;
+ end;
+
+ function SINH (X : REAL) return REAL is
+ begin
+ assert false severity failure;
+ end;
+
+ function COSH (X : REAL) return REAL is
+ begin
+ assert false severity failure;
+ end;
+
+ function TANH (X : REAL) return REAL is
+ begin
+ assert false severity failure;
+ end;
+ function ARCSINH (X : REAL) return REAL is
+ begin
+ assert false severity failure;
+ end;
+
+ function ARCCOSH (X : REAL) return REAL is
+ begin
+ assert false severity failure;
+ end;
+
+ function ARCTANH (Y : REAL) return REAL is
+ begin
+ assert false severity failure;
+ end;
end MATH_REAL;
diff --git a/libraries/openieee/math_real.vhdl b/libraries/openieee/math_real.vhdl
index b2814f534..b8c2150d9 100644
--- a/libraries/openieee/math_real.vhdl
+++ b/libraries/openieee/math_real.vhdl
@@ -17,7 +17,31 @@
-- <http://www.gnu.org/licenses/>.
package MATH_REAL is
- constant math_pi : real := 3.14159_26535_89793_23846;
+ -- The values were computed with at least 40 digits and rounded to
+ -- 20 digits after the dot. They were checked with the original ieee
+ -- specification (log2_of_e has an extra digit from the spec).
+ constant math_e : real := 2.71828_18284_59045_23536;
+ constant math_1_over_e : real := 0.36787_94411_71442_321596;
+
+ constant math_pi : real := 3.14159_26535_89793_23846;
+ constant math_2_pi : real := 6.28318_53071_79586_47693;
+ constant math_pi_over_2 : real := 1.57079_63267_94896_61923;
+ constant math_pi_over_3 : real := 1.04719_75511_96597_74615;
+ constant math_pi_over_4 : real := 0.78539_81633_97448_30962;
+ constant math_3_pi_over_2 : real := 4.71238_89803_84689_85769;
+
+ constant math_log_of_2 : real := 0.69314_71805_59945_30942;
+ constant math_log_of_10 : real := 2.30258_50929_94045_68402;
+ constant math_log2_of_e : real := 1.44269_50408_88963_40736;
+ constant math_log10_of_e : real := 0.43429_44819_03251_82765;
+
+ constant math_sqrt_2 : real := 1.41421_35623_73095_04880;
+ constant math_1_over_sqrt_2 : real := 0.70710_67811_86547_52440;
+ constant math_sqrt_pi : real := 1.77245_38509_05516_02730;
+
+ constant math_deg_to_rad : real := 0.01745_32925_19943_29577;
+ constant math_rad_to_deg : real := 57.29577_95130_82320_87680;
+
function SIGN (X : REAL) return REAL;
function CEIL (X : REAL) return REAL;
@@ -32,13 +56,83 @@ package MATH_REAL is
function TRUNC (X : REAL) return REAL;
attribute foreign of trunc : function is "VHPIDIRECT trunc";
+ function "mod" (X, Y : REAL) return REAL;
+ -- Contrary to fmod, the sign of the result is the sign of Y.
+
+ function REALMAX (X, Y : REAL) return REAL;
+ attribute foreign of REALMAX : function is "VHPIDIRECT fmax";
+
+ function REALMIN (X, Y : REAL) return REAL;
+ attribute foreign of REALMIN : function is "VHPIDIRECT fmin";
+
procedure UNIFORM (SEED1, SEED2 : inout POSITIVE; X : out REAL);
-- Algorithm from: Pierre L'Ecuyer, CACM June 1988 Volume 31 Number 6
-- page 747 figure 3.
+ function SQRT (X : REAL) return REAL;
+ attribute foreign of SQRT : function is "VHPIDIRECT sqrt";
+
+ function CBRT (X : REAL) return REAL;
+ attribute foreign of CBRT : function is "VHPIDIRECT cbrt";
+
+ function "**" (X : INTEGER; Y : REAL) return REAL;
+
+ function "**" (X : REAL; Y : REAL) return REAL;
+ attribute foreign of "**" [ REAL, REAL return REAL ]: function is
+ "VHPIDIRECT pow";
+
+ function EXP (X : REAL) return REAL;
+ attribute foreign of EXP : function is "VHPIDIRECT exp";
+
+ function LOG (X : REAL) return REAL;
+ attribute foreign of LOG [ REAL return REAL ] : function is "VHPIDIRECT log";
+
+ function LOG2 (X : REAL) return REAL;
+ attribute foreign of LOG2 : function is "VHPIDIRECT log2";
+
+ function LOG10 (X : REAL) return REAL;
+ attribute foreign of LOG10 : function is "VHPIDIRECT log10";
+
+ function LOG (X : REAL; BASE : REAL) return REAL;
+
function SIN (X : REAL) return REAL;
attribute foreign of SIN : function is "VHPIDIRECT sin";
function COS (X : REAL) return REAL;
attribute foreign of COS : function is "VHPIDIRECT cos";
+
+ function TAN (X : REAL) return REAL;
+ attribute foreign of TAN : function is "VHPIDIRECT tan";
+
+ function ARCSIN (X : REAL) return REAL;
+ attribute foreign of ARCSIN : function is "VHPIDIRECT asin";
+
+ function ARCCOS (X : REAL) return REAL;
+ attribute foreign of ARCCOS : function is "VHPIDIRECT acos";
+
+ function ARCTAN (Y : REAL) return REAL;
+ attribute foreign of ARCTAN [ REAL return REAL ]: function is
+ "VHPIDIRECT atan";
+
+ function ARCTAN (Y, X : REAL) return REAL;
+ attribute foreign of ARCTAN [ REAL, REAL return REAL ]: function is
+ "VHPIDIRECT atan2";
+
+ function SINH (X : REAL) return REAL;
+ attribute foreign of SINH : function is "VHPIDIRECT sinh";
+
+ function COSH (X : REAL) return REAL;
+ attribute foreign of COSH : function is "VHPIDIRECT cosh";
+
+ function TANH (X : REAL) return REAL;
+ attribute foreign of TANH : function is "VHPIDIRECT tanh";
+
+ function ARCSINH (X : REAL) return REAL;
+ attribute foreign of ARCSINH : function is "VHPIDIRECT asinh";
+
+ function ARCCOSH (X : REAL) return REAL;
+ attribute foreign of ARCCOSH : function is "VHPIDIRECT acosh";
+
+ function ARCTANH (Y : REAL) return REAL;
+ attribute foreign of ARCTANH : function is "VHPIDIRECT atanh";
end MATH_REAL;