summaryrefslogtreecommitdiff
path: root/pr/tests/lltest.c
diff options
context:
space:
mode:
Diffstat (limited to 'pr/tests/lltest.c')
-rw-r--r--pr/tests/lltest.c236
1 files changed, 118 insertions, 118 deletions
diff --git a/pr/tests/lltest.c b/pr/tests/lltest.c
index 7e21e76f..36e67984 100644
--- a/pr/tests/lltest.c
+++ b/pr/tests/lltest.c
@@ -2,7 +2,7 @@
/* This Source Code Form is subject to the terms of the Mozilla Public
* License, v. 2.0. If a copy of the MPL was not distributed with this
* file, You can obtain one at http://mozilla.org/MPL/2.0/. */
-
+
/*
** testll.c -- test suite for 64bit integer (longlong) operations
**
@@ -15,11 +15,11 @@
**
** Description:
** lltest.c tests the functions defined in NSPR 2.0's prlong.h.
-**
+**
** Successive tests begin to depend on other LL functions working
** correctly. So, ... Do not change the order of the tests as run
** from main().
-**
+**
** Caveats:
** Do not even begin to think that this is an exhaustive test!
**
@@ -54,8 +54,8 @@ static PRBool verboseMode = PR_FALSE;
const PRInt64 bigZero = LL_INIT( 0, 0 );
const PRInt64 bigOne = LL_INIT( 0, 1 );
const PRInt64 bigTwo = LL_INIT( 0, 2 );
-const PRInt64 bigSixTeen = LL_INIT( 0, 16 );
-const PRInt64 bigThirtyTwo = LL_INIT( 0, 32 );
+const PRInt64 bigSixTeen = LL_INIT( 0, 16 );
+const PRInt64 bigThirtyTwo = LL_INIT( 0, 32 );
const PRInt64 bigMinusOne = LL_INIT( 0xffffffff, 0xffffffff );
const PRInt64 bigMinusTwo = LL_INIT( 0xffffffff, 0xfffffffe );
const PRInt64 bigNumber = LL_INIT( 0x7fffffff, 0xffffffff );
@@ -107,7 +107,7 @@ ResultFailed( char *what, char *how, PRInt64 expected, PRInt64 got)
PR_fprintf(output, "Expected: 0x%llx Got: 0x%llx\n", expected, got );
}
return;
-}
+}
/*
@@ -127,106 +127,106 @@ static void TestAssignment( void )
/*
** TestComparisons() -- Test the longlong comparison operations
*/
-static void
+static void
TestComparisons( void )
{
ReportProgress("Testing Comparisons Operations\n");
-
- /* test for zero */
+
+ /* test for zero */
if ( !LL_IS_ZERO( bigZero ))
SetFailed( "LL_IS_ZERO", "Zero is not zero" );
-
+
if ( LL_IS_ZERO( bigOne ))
SetFailed( "LL_IS_ZERO", "One tests as zero" );
-
+
if ( LL_IS_ZERO( bigMinusOne ))
SetFailed( "LL_IS_ZERO", "Minus One tests as zero" );
-
+
/* test equal */
if ( !LL_EQ( bigZero, bigZero ))
SetFailed( "LL_EQ", "zero EQ zero");
-
+
if ( !LL_EQ( bigOne, bigOne ))
SetFailed( "LL_EQ", "one EQ one" );
-
+
if ( !LL_EQ( bigNumber, bigNumber ))
SetFailed( "LL_EQ", "bigNumber EQ bigNumber" );
-
+
if ( !LL_EQ( bigMinusOne, bigMinusOne ))
SetFailed( "LL_EQ", "minus one EQ minus one");
-
+
if ( LL_EQ( bigZero, bigOne ))
SetFailed( "LL_EQ", "zero EQ one");
-
+
if ( LL_EQ( bigOne, bigZero ))
SetFailed( "LL_EQ", "one EQ zero" );
-
+
if ( LL_EQ( bigMinusOne, bigOne ))
SetFailed( "LL_EQ", "minus one EQ one");
-
+
if ( LL_EQ( bigNumber, bigOne ))
SetFailed( "LL_EQ", "bigNumber EQ one");
-
+
/* test not equal */
if ( LL_NE( bigZero, bigZero ))
SetFailed( "LL_NE", "0 NE 0");
-
+
if ( LL_NE( bigOne, bigOne ))
SetFailed( "LL_NE", "1 NE 1");
-
+
if ( LL_NE( bigMinusOne, bigMinusOne ))
SetFailed( "LL_NE", "-1 NE -1");
-
+
if ( LL_NE( bigNumber, bigNumber ))
SetFailed( "LL_NE", "n NE n");
-
+
if ( LL_NE( bigMinusNumber, bigMinusNumber ))
SetFailed( "LL_NE", "-n NE -n");
-
+
if ( !LL_NE( bigZero, bigOne))
SetFailed( "LL_NE", "0 NE 1");
-
+
if ( !LL_NE( bigOne, bigMinusNumber))
SetFailed( "LL_NE", "1 NE -n");
-
+
/* Greater than or equal to zero */
if ( !LL_GE_ZERO( bigZero ))
SetFailed( "LL_GE_ZERO", "0");
-
+
if ( !LL_GE_ZERO( bigOne ))
SetFailed( "LL_GE_ZERO", "1");
-
+
if ( !LL_GE_ZERO( bigNumber ))
SetFailed( "LL_GE_ZERO", "n");
-
+
if ( LL_GE_ZERO( bigMinusOne ))
SetFailed( "LL_GE_ZERO", "-1");
-
+
if ( LL_GE_ZERO( bigMinusNumber ))
SetFailed( "LL_GE_ZERO", "-n");
-
+
/* Algebraic Compare two values */
if ( !LL_CMP( bigZero, ==, bigZero ))
SetFailed( "LL_CMP", "0 == 0");
-
+
if ( LL_CMP( bigZero, >, bigZero ))
SetFailed( "LL_CMP", "0 > 0");
-
+
if ( LL_CMP( bigZero, <, bigZero ))
SetFailed( "LL_CMP", "0 < 0");
-
+
if ( LL_CMP( bigNumber, <, bigOne ))
SetFailed( "LL_CMP", "n < 1");
-
+
if ( !LL_CMP( bigNumber, >, bigOne ))
SetFailed( "LL_CMP", "n <= 1");
-
+
if ( LL_CMP( bigOne, >, bigNumber ))
SetFailed( "LL_CMP", "1 > n");
-
+
if ( LL_CMP( bigMinusNumber, >, bigNumber ))
SetFailed( "LL_CMP", "-n > n");
-
+
if ( LL_CMP( bigNumber, !=, bigNumber))
SetFailed( "LL_CMP", "n != n");
@@ -287,22 +287,22 @@ TestComparisons( void )
/* Bitwise Compare two numbers */
if ( !LL_UCMP( bigZero, ==, bigZero ))
SetFailed( "LL_UCMP", "0 == 0");
-
+
if ( LL_UCMP( bigZero, >, bigZero ))
SetFailed( "LL_UCMP", "0 > 0");
-
+
if ( LL_UCMP( bigZero, <, bigZero ))
SetFailed( "LL_UCMP", "0 < 0");
-
+
if ( LL_UCMP( bigNumber, <, bigOne ))
SetFailed( "LL_UCMP", "n < 1");
-
+
if ( !LL_UCMP( bigNumber, >, bigOne ))
SetFailed( "LL_UCMP", "n < 1");
-
+
if ( LL_UCMP( bigOne, >, bigNumber ))
SetFailed( "LL_UCMP", "1 > n");
-
+
if ( LL_UCMP( bigMinusNumber, <, bigNumber ))
SetFailed( "LL_UCMP", "-n < n");
@@ -365,14 +365,14 @@ static void
TestLogicalOperations( void )
{
PRUint64 result, result2;
-
+
ReportProgress("Testing Logical Operations\n");
-
+
/* Test AND */
LL_AND( result, bigZero, bigZero );
if ( !LL_IS_ZERO( result ))
ResultFailed( "LL_AND", "0 & 0", bigZero, result );
-
+
LL_AND( result, bigOne, bigOne );
if ( LL_IS_ZERO( result ))
ResultFailed( "LL_AND", "1 & 1", bigOne, result );
@@ -384,47 +384,47 @@ TestLogicalOperations( void )
LL_AND( result, bigMinusOne, bigMinusOne );
if ( !LL_UCMP( result, ==, bigMinusOne ))
ResultFailed( "LL_AND", "-1 & -1", bigMinusOne, result );
-
+
/* test OR */
LL_OR( result, bigZero, bigZero );
if ( !LL_IS_ZERO( result ))
ResultFailed( "LL_OR", "0 | 1", bigZero, result);
-
+
LL_OR( result, bigZero, bigOne );
if ( LL_IS_ZERO( result ))
ResultFailed( "LL_OR", "0 | 1", bigOne, result );
-
+
LL_OR( result, bigZero, bigMinusNumber );
if ( !LL_UCMP( result, ==, bigMinusNumber ))
ResultFailed( "LL_OR", "0 | -n", bigMinusNumber, result);
-
+
LL_OR( result, bigMinusNumber, bigZero );
if ( !LL_UCMP( result, ==, bigMinusNumber ))
ResultFailed( "LL_OR", "-n | 0", bigMinusNumber, result );
-
+
/* test XOR */
LL_XOR( result, bigZero, bigZero );
if ( LL_UCMP( result, !=, bigZero ))
ResultFailed( "LL_XOR", "0 ^ 0", bigZero, result);
-
+
LL_XOR( result, bigOne, bigZero );
if ( LL_UCMP( result, !=, bigOne ))
ResultFailed( "LL_XOR", "1 ^ 0", bigZero, result );
-
+
LL_XOR( result, bigMinusNumber, bigZero );
if ( LL_UCMP( result, !=, bigMinusNumber ))
ResultFailed( "LL_XOR", "-n ^ 0", bigMinusNumber, result );
-
+
LL_XOR( result, bigMinusNumber, bigMinusNumber );
if ( LL_UCMP( result, !=, bigZero ))
ResultFailed( "LL_XOR", "-n ^ -n", bigMinusNumber, result);
-
+
/* test OR2. */
result = bigZero;
LL_OR2( result, bigOne );
if ( LL_UCMP( result, !=, bigOne ))
ResultFailed( "LL_OR2", "(r=0) |= 1", bigOne, result);
-
+
result = bigOne;
LL_OR2( result, bigNumber );
if ( LL_UCMP( result, !=, bigNumber ))
@@ -440,7 +440,7 @@ TestLogicalOperations( void )
LL_NOT( result2, result);
if ( LL_UCMP( result2, !=, bigMinusNumber ))
ResultFailed( "LL_NOT", "r != ~(~-n)", bigMinusNumber, result);
-
+
/* test Negation */
LL_NEG( result, bigMinusNumber );
LL_NEG( result2, result );
@@ -465,79 +465,79 @@ TestConversion( void )
PRUint32 resultU32;
float resultF;
PRFloat64 resultD;
-
+
ReportProgress("Testing Conversion Operations\n");
-
+
/* LL_L2I -- Convert to signed 32bit */
LL_L2I(result32, bigOne );
- if ( result32 != one )
+ if ( result32 != one )
SetFailed( "LL_L2I", "r != 1");
-
+
LL_L2I(result32, bigMinusOne );
- if ( result32 != minusOne )
+ if ( result32 != minusOne )
SetFailed( "LL_L2I", "r != -1");
-
+
/* LL_L2UI -- Convert 64bit to unsigned 32bit */
LL_L2UI( resultU32, bigMinusOne );
if ( resultU32 != (PRUint32) minusOne )
SetFailed( "LL_L2UI", "r != -1");
-
+
LL_L2UI( resultU32, bigOne );
if ( resultU32 != (PRUint32) one )
SetFailed( "LL_L2UI", "r != 1");
-
+
/* LL_L2F -- Convert to 32bit floating point */
LL_L2F( resultF, bigOne );
if ( resultF != 1.0 )
SetFailed( "LL_L2F", "r != 1.0");
-
+
LL_L2F( resultF, bigMinusOne );
if ( resultF != -1.0 )
SetFailed( "LL_L2F", "r != 1.0");
-
+
/* LL_L2D -- Convert to 64bit floating point */
LL_L2D( resultD, bigOne );
if ( resultD != 1.0L )
SetFailed( "LL_L2D", "r != 1.0");
-
+
LL_L2D( resultD, bigMinusOne );
if ( resultD != -1.0L )
SetFailed( "LL_L2D", "r != -1.0");
-
+
/* LL_I2L -- Convert 32bit signed to 64bit signed */
LL_I2L( result, one );
if ( LL_CMP(result, !=, bigOne ))
SetFailed( "LL_I2L", "r != 1");
-
+
LL_I2L( result, minusOne );
if ( LL_CMP(result, !=, bigMinusOne ))
SetFailed( "LL_I2L", "r != -1");
-
+
/* LL_UI2L -- Convert 32bit unsigned to 64bit unsigned */
LL_UI2L( resultU, (PRUint32) one );
if ( LL_CMP(resultU, !=, bigOne ))
SetFailed( "LL_UI2L", "r != 1");
-
- /* [lth.] This did not behave as expected, but it is correct
+
+ /* [lth.] This did not behave as expected, but it is correct
*/
LL_UI2L( resultU, (PRUint32) minusOne );
if ( LL_CMP(resultU, !=, bigZeroFox ))
ResultFailed( "LL_UI2L", "r != -1", bigZeroFox, resultU);
-
+
/* LL_F2L -- Convert 32bit float to 64bit signed */
LL_F2L( result, 1.0 );
if ( LL_CMP(result, !=, bigOne ))
SetFailed( "LL_F2L", "r != 1");
-
+
LL_F2L( result, -1.0 );
if ( LL_CMP(result, !=, bigMinusOne ))
SetFailed( "LL_F2L", "r != -1");
-
+
/* LL_D2L -- Convert 64bit Float to 64bit signed */
LL_D2L( result, 1.0L );
if ( LL_CMP(result, !=, bigOne ))
SetFailed( "LL_D2L", "r != 1");
-
+
LL_D2L( result, -1.0L );
if ( LL_CMP(result, !=, bigMinusOne ))
SetFailed( "LL_D2L", "r != -1");
@@ -561,7 +561,7 @@ static void ShiftCompileOnly()
LL_ISHL(ia, 49, 32);
} /* ShiftCompileOnly */
-
+
/*
** TestShift() -- Test Shifting Operations
@@ -575,21 +575,21 @@ TestShift( void )
PRUint64 resultU;
ReportProgress("Testing Shifting Operations\n");
-
+
/* LL_SHL -- Shift left algebraic */
LL_SHL( result, bigOne, one );
if ( LL_CMP( result, !=, bigTwo ))
ResultFailed( "LL_SHL", "r != 2", bigOne, result );
-
+
LL_SHL( result, bigTwo, thirtyTwo );
if ( LL_CMP( result, !=, largeTwoZero ))
ResultFailed( "LL_SHL", "r != twoZero", largeTwoZero, result);
-
+
/* LL_SHR -- Shift right algebraic */
LL_SHR( result, bigFoxZero, thirtyTwo );
if ( LL_CMP( result, !=, bigMinusOne ))
ResultFailed( "LL_SHR", "r != -1", bigMinusOne, result);
-
+
LL_SHR( result, bigTwo, one );
if ( LL_CMP( result, !=, bigOne ))
ResultFailed( "LL_SHR", "r != 1", bigOne, result);
@@ -597,31 +597,31 @@ TestShift( void )
LL_SHR( result, bigFoxFox, thirtyTwo );
if ( LL_CMP( result, !=, bigMinusOne ))
ResultFailed( "LL_SHR", "r != -1 (was ff,ff)", bigMinusOne, result);
-
+
/* LL_USHR -- Logical shift right */
LL_USHR( resultU, bigZeroFox, thirtyTwo );
if ( LL_UCMP( resultU, !=, bigZero ))
ResultFailed( "LL_USHR", "r != 0 ", bigZero, result);
-
+
LL_USHR( resultU, bigFoxFox, thirtyTwo );
if ( LL_UCMP( resultU, !=, bigZeroFox ))
ResultFailed( "LL_USHR", "r != 0 ", bigZeroFox, result);
-
+
/* LL_ISHL -- Shift a 32bit integer into a 64bit result */
LL_ISHL( resultU, minusOne, thirtyTwo );
if ( LL_UCMP( resultU, !=, bigFoxZero ))
ResultFailed( "LL_ISHL", "r != ff,00 ", bigFoxZero, result);
-
+
LL_ISHL( resultU, one, sixtyThree );
if ( LL_UCMP( resultU, !=, bigEightZero ))
ResultFailed( "LL_ISHL", "r != 80,00 ", bigEightZero, result);
-
+
LL_ISHL( resultU, one, sixteen );
if ( LL_UCMP( resultU, !=, big64K ))
ResultFailed( "LL_ISHL", "r != 64K ", big64K, resultU);
-
+
return;
-}
+}
/*
@@ -637,10 +637,10 @@ TestArithmetic( void )
PRInt64 largeMultCand = LL_INIT( 0x00000000, 0x7fffffff );
PRInt64 largeMinusMultCand = LL_INIT( 0xffffffff, 0x10000001 );
PRInt64 largeMultCandx64K = LL_INIT( 0x00007fff, 0xffff0000 );
- PRInt64 largeNumSHL5 = LL_INIT( 0x0000001f, 0xffffffe0 );
+ PRInt64 largeNumSHL5 = LL_INIT( 0x0000001f, 0xffffffe0 );
PRInt64 result, result2;
- /* Addition */
+ /* Addition */
LL_ADD( result, bigOne, bigOne );
if ( LL_CMP( result, !=, bigTwo ))
ResultFailed( "LL_ADD", "r != 1 + 1", bigTwo, result);
@@ -652,30 +652,30 @@ TestArithmetic( void )
LL_ADD( result, largeVal, bigOne );
if ( LL_CMP( result, !=, largeValPlusOne ))
ResultFailed( "LL_ADD", "lVP1 != lV + 1", largeValPlusOne, result);
-
+
/* Subtraction */
LL_SUB( result, bigOne, bigOne );
if ( LL_CMP( result, !=, bigZero ))
ResultFailed( "LL_SUB", "r != 1 - 1", bigZero, result);
-
+
LL_SUB( result, bigTwo, bigOne );
if ( LL_CMP( result, !=, bigOne ))
ResultFailed( "LL_SUB", "r != 2 - 1", bigOne, result);
-
+
LL_SUB( result, largeValPlusOne, bigOne );
if ( LL_CMP( result, !=, largeVal ))
ResultFailed( "LL_SUB", "r != lVP1 - 1", largeVal, result);
-
-
+
+
/* Multiply */
LL_MUL( result, largeVal, bigTwo );
if ( LL_CMP( result, !=, largeValTimesTwo ))
ResultFailed( "LL_MUL", "r != lV*2", largeValTimesTwo, result);
-
+
LL_MUL( result, largeMultCand, big64K );
if ( LL_CMP( result, !=, largeMultCandx64K ))
ResultFailed( "LL_MUL", "r != lV*64K", largeMultCandx64K, result);
-
+
LL_NEG( result2, largeMultCand );
LL_MUL( result, largeMultCand, bigMinusOne );
if ( LL_CMP( result, !=, result2 ))
@@ -686,13 +686,13 @@ TestArithmetic( void )
if ( LL_CMP( result, !=, largeNumSHL5 ))
ResultFailed( "LL_MUL", "r != 0f<<5", largeNumSHL5, result );
-
+
/* LL_DIV() Division */
LL_DIV( result, bigOne, bigOne);
if ( LL_CMP( result, !=, bigOne ))
ResultFailed( "LL_DIV", "1 != 1", bigOne, result);
-
+
LL_DIV( result, bigNumber, bigOne );
if ( LL_CMP( result, !=, bigNumber ))
ResultFailed( "LL_DIV", "r != n / 1", bigNumber, result);
@@ -704,32 +704,32 @@ TestArithmetic( void )
LL_DIV( result, bigMinusNumber, bigMinusOne );
if ( LL_CMP( result, !=, bigNumber ))
ResultFailed( "LL_DIV", "r != -n / -1", bigNumber, result);
-
+
LL_SHL( result2, bigZeroFox, 5 );
LL_DIV( result, result2, bigOne );
if ( LL_CMP( result, !=, result2 ))
ResultFailed( "LL_DIV", "0f<<5 != 0f<<5", result2, result);
-
+
LL_SHL( result2, bigZeroFox, 5 );
LL_NEG( result2, result2 );
LL_DIV( result, result2, bigOne );
if ( LL_CMP( result, !=, result2 ))
ResultFailed( "LL_DIV", "-0f<<5 != -0f<<5", result2, result);
-
+
LL_SHL( result2, bigZeroFox, 17 );
LL_DIV( result, result2, bigMinusOne );
LL_NEG( result2, result2 );
if ( LL_CMP( result, !=, result2 ))
ResultFailed( "LL_DIV", "-0f<<17 != -0f<<17", result2, result);
-
-
+
+
/* LL_MOD() Modulo Division */
LL_ADD( result2, bigThirtyTwo, bigOne );
LL_MOD( result, result2, bigSixTeen );
if ( LL_CMP( result, !=, bigOne ))
ResultFailed( "LL_MOD", "r != 1", bigSixTeen, result);
-
-
+
+
LL_MUL( result2, bigZeroFox, bigThirtyTwo );
LL_ADD( result2, result2, bigSixTeen);
LL_MOD( result, result2, bigThirtyTwo );
@@ -740,10 +740,10 @@ TestArithmetic( void )
LL_DIV( result, bigOne, bigOne);
if ( LL_CMP( result, !=, bigOne ))
ResultFailed( "LL_DIV", "r != 16", bigSixTeen, result);
-
+
return;
-}
+}
static void TestWellknowns(void)
{
@@ -755,7 +755,7 @@ static void TestWellknowns(void)
ResultFailed( "min, mmin", "min != mmin", max, mmin);
if (LL_NE(zero, mzero))
ResultFailed( "zero, mzero", "zero != mzero", zero, mzero);
-} /* TestWellknowns */
+} /* TestWellknowns */
/*
** Initialize() -- Initialize the test case
@@ -771,7 +771,7 @@ Initialize( PRIntn argc, char **argv )
/*
** Parse command line options
- */
+ */
while (PL_OPT_EOL != (os = PL_GetNextOpt(opt)))
{
if (PL_OPT_BAD == os) continue;
@@ -780,12 +780,12 @@ Initialize( PRIntn argc, char **argv )
case 'd': /* set debug mode */
debugMode = PR_TRUE;
break;
-
+
case 'v': /* set verbose mode */
verboseMode = PR_TRUE;
debugMode = PR_TRUE;
break;
-
+
case 'h': /* user wants some guidance */
default:
PR_fprintf(output, "You get help.\n");
@@ -794,7 +794,7 @@ Initialize( PRIntn argc, char **argv )
}
PL_DestroyOptState(opt);
return(0);
-}
+}
int main(int argc, char **argv)
{
@@ -811,14 +811,14 @@ int main(int argc, char **argv)
TestShift();
TestArithmetic();
TestWellknowns();
-
+
/*
** That's all folks!
*/
if ( failedAlready )
{
PR_fprintf(output, "FAIL\n");\
- }
+ }
else
{
PR_fprintf(output, "PASS\n");\