ahmed/OpenCL-CTS
1
0
Fork 0
mirror of https://github.com/KhronosGroup/OpenCL-CTS.git synced 2026-03-21 23:09:01 +00:00
Code Packages Projects Releases Wiki Activity

Synchronise with Khronos-private Gitlab branch

Browse Source 
The maintenance of the conformance tests is moving to Github.

This commit contains all the changes that have been done in
Gitlab since the first public release of the conformance tests.

Signed-off-by: Kevin Petit <kevin.petit@arm.com>
This commit is contained in:
Kevin Petit
2019-02-20 16:36:05 +00:00
committed by Kévin Petit
parent 95196e7fb4
commit d8733efc0f
576 changed files with 212486 additions and 191776 deletions
Download Patch File Download Diff File Expand all files Collapse all files

228
test_conformance/math_brute_force/binary_i.c
View File

@@ -1,6 +1,6 @@
//
// Copyright (c) 2017 The Khronos Group Inc.
//
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
@@ -29,7 +29,7 @@ int TestFunc_Float_Float_Int(const Func *f, MTdata);
int TestFunc_Double_Double_Int(const Func *f, MTdata);
#if defined( __cplusplus)
extern "C"
extern "C"
#endif
const vtbl _binary_i = { "binary_i", TestFunc_Float_Float_Int, TestFunc_Double_Double_Int };
@@ -44,7 +44,7 @@ static int BuildKernel( const char *name, int vectorSize, cl_uint kernel_count,
" out[i] = ", name, "( in1[i], in2[i] );\n"
"}\n"
};
const char *c3[] = { "__kernel void math_kernel", sizeNames[vectorSize], "( __global float* out, __global float* in, __global int* in2)\n"
"{\n"
" size_t i = get_global_id(0);\n"
@@ -87,31 +87,31 @@ static int BuildKernel( const char *name, int vectorSize, cl_uint kernel_count,
const char **kern = c;
size_t kernSize = sizeof(c)/sizeof(c[0]);
if( sizeValues[vectorSize] == 3 )
{
kern = c3;
kernSize = sizeof(c3)/sizeof(c3[0]);
}
char testName[32];
snprintf( testName, sizeof( testName ) -1, "math_kernel%s", sizeNames[vectorSize] );
return MakeKernels(kern, (cl_uint) kernSize, testName, kernel_count, k, p);
return MakeKernels(kern, (cl_uint) kernSize, testName, kernel_count, k, p);
}
static int BuildKernelDouble( const char *name, int vectorSize, cl_uint kernel_count, cl_kernel *k, cl_program *p )
{
const char *c[] = { "#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n",
"__kernel void math_kernel", sizeNames[vectorSize], "( __global double", sizeNames[vectorSize], "* out, __global double", sizeNames[vectorSize], "* in1, __global int", sizeNames[vectorSize], "* in2 )\n"
"__kernel void math_kernel", sizeNames[vectorSize], "( __global double", sizeNames[vectorSize], "* out, __global double", sizeNames[vectorSize], "* in1, __global int", sizeNames[vectorSize], "* in2 )\n"
"{\n"
" int i = get_global_id(0);\n"
" out[i] = ", name, "( in1[i], in2[i] );\n"
"}\n"
};
const char *c3[] = { "#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n",
"__kernel void math_kernel", sizeNames[vectorSize], "( __global double* out, __global double* in, __global int* in2)\n"
"__kernel void math_kernel", sizeNames[vectorSize], "( __global double* out, __global double* in, __global int* in2)\n"
"{\n"
" size_t i = get_global_id(0);\n"
" if( i + 1 < get_global_size(0) )\n"
@@ -153,17 +153,17 @@ static int BuildKernelDouble( const char *name, int vectorSize, cl_uint kernel_c
const char **kern = c;
size_t kernSize = sizeof(c)/sizeof(c[0]);
if( sizeValues[vectorSize] == 3 )
{
kern = c3;
kernSize = sizeof(c3)/sizeof(c3[0]);
}
char testName[32];
snprintf( testName, sizeof( testName ) -1, "math_kernel%s", sizeNames[vectorSize] );
return MakeKernels(kern, (cl_uint) kernSize, testName, kernel_count, k, p);
return MakeKernels(kern, (cl_uint) kernSize, testName, kernel_count, k, p);
}
typedef struct BuildKernelInfo
@@ -193,25 +193,25 @@ static cl_int BuildKernel_DoubleFn( cl_uint job_id, cl_uint thread_id UNUSED, vo
// A table of more difficult cases to get right
static const float specialValuesFloat[] = {
-NAN, -INFINITY, -FLT_MAX, MAKE_HEX_FLOAT(-0x1.000002p64f, -0x1000002L, 40), MAKE_HEX_FLOAT(-0x1.0p64f, -0x1L, 64), MAKE_HEX_FLOAT(-0x1.fffffep63f, -0x1fffffeL, 39), MAKE_HEX_FLOAT(-0x1.000002p63f, -0x1000002L, 39), MAKE_HEX_FLOAT(-0x1.0p63f, -0x1L, 63), MAKE_HEX_FLOAT(-0x1.fffffep62f, -0x1fffffeL, 38),
MAKE_HEX_FLOAT(-0x1.000002p32f, -0x1000002L, 8), MAKE_HEX_FLOAT(-0x1.0p32f, -0x1L, 32), MAKE_HEX_FLOAT(-0x1.fffffep31f, -0x1fffffeL, 7), MAKE_HEX_FLOAT(-0x1.000002p31f, -0x1000002L, 7), MAKE_HEX_FLOAT(-0x1.0p31f, -0x1L, 31), MAKE_HEX_FLOAT(-0x1.fffffep30f, -0x1fffffeL, 6), -1000.f, -100.f, -4.0f, -3.5f,
-3.0f, MAKE_HEX_FLOAT(-0x1.800002p1f, -0x1800002L, -23), -2.5f, MAKE_HEX_FLOAT(-0x1.7ffffep1f, -0x17ffffeL, -23), -2.0f, MAKE_HEX_FLOAT(-0x1.800002p0f, -0x1800002L, -24), -1.5f, MAKE_HEX_FLOAT(-0x1.7ffffep0f, -0x17ffffeL, -24),MAKE_HEX_FLOAT(-0x1.000002p0f, -0x1000002L, -24), -1.0f, MAKE_HEX_FLOAT(-0x1.fffffep-1f, -0x1fffffeL, -25),
MAKE_HEX_FLOAT(-0x1.000002p-1f, -0x1000002L, -25), -0.5f, MAKE_HEX_FLOAT(-0x1.fffffep-2f, -0x1fffffeL, -26), MAKE_HEX_FLOAT(-0x1.000002p-2f, -0x1000002L, -26), -0.25f, MAKE_HEX_FLOAT(-0x1.fffffep-3f, -0x1fffffeL, -27),
MAKE_HEX_FLOAT(-0x1.000002p-126f, -0x1000002L, -150), -FLT_MIN, MAKE_HEX_FLOAT(-0x0.fffffep-126f, -0x0fffffeL, -150), MAKE_HEX_FLOAT(-0x0.000ffep-126f, -0x0000ffeL, -150), MAKE_HEX_FLOAT(-0x0.0000fep-126f, -0x00000feL, -150), MAKE_HEX_FLOAT(-0x0.00000ep-126f, -0x000000eL, -150), MAKE_HEX_FLOAT(-0x0.00000cp-126f, -0x000000cL, -150), MAKE_HEX_FLOAT(-0x0.00000ap-126f, -0x000000aL, -150),
static const float specialValuesFloat[] = {
-NAN, -INFINITY, -FLT_MAX, MAKE_HEX_FLOAT(-0x1.000002p64f, -0x1000002L, 40), MAKE_HEX_FLOAT(-0x1.0p64f, -0x1L, 64), MAKE_HEX_FLOAT(-0x1.fffffep63f, -0x1fffffeL, 39), MAKE_HEX_FLOAT(-0x1.000002p63f, -0x1000002L, 39), MAKE_HEX_FLOAT(-0x1.0p63f, -0x1L, 63), MAKE_HEX_FLOAT(-0x1.fffffep62f, -0x1fffffeL, 38),
MAKE_HEX_FLOAT(-0x1.000002p32f, -0x1000002L, 8), MAKE_HEX_FLOAT(-0x1.0p32f, -0x1L, 32), MAKE_HEX_FLOAT(-0x1.fffffep31f, -0x1fffffeL, 7), MAKE_HEX_FLOAT(-0x1.000002p31f, -0x1000002L, 7), MAKE_HEX_FLOAT(-0x1.0p31f, -0x1L, 31), MAKE_HEX_FLOAT(-0x1.fffffep30f, -0x1fffffeL, 6), -1000.f, -100.f, -4.0f, -3.5f,
-3.0f, MAKE_HEX_FLOAT(-0x1.800002p1f, -0x1800002L, -23), -2.5f, MAKE_HEX_FLOAT(-0x1.7ffffep1f, -0x17ffffeL, -23), -2.0f, MAKE_HEX_FLOAT(-0x1.800002p0f, -0x1800002L, -24), -1.5f, MAKE_HEX_FLOAT(-0x1.7ffffep0f, -0x17ffffeL, -24),MAKE_HEX_FLOAT(-0x1.000002p0f, -0x1000002L, -24), -1.0f, MAKE_HEX_FLOAT(-0x1.fffffep-1f, -0x1fffffeL, -25),
MAKE_HEX_FLOAT(-0x1.000002p-1f, -0x1000002L, -25), -0.5f, MAKE_HEX_FLOAT(-0x1.fffffep-2f, -0x1fffffeL, -26), MAKE_HEX_FLOAT(-0x1.000002p-2f, -0x1000002L, -26), -0.25f, MAKE_HEX_FLOAT(-0x1.fffffep-3f, -0x1fffffeL, -27),
MAKE_HEX_FLOAT(-0x1.000002p-126f, -0x1000002L, -150), -FLT_MIN, MAKE_HEX_FLOAT(-0x0.fffffep-126f, -0x0fffffeL, -150), MAKE_HEX_FLOAT(-0x0.000ffep-126f, -0x0000ffeL, -150), MAKE_HEX_FLOAT(-0x0.0000fep-126f, -0x00000feL, -150), MAKE_HEX_FLOAT(-0x0.00000ep-126f, -0x000000eL, -150), MAKE_HEX_FLOAT(-0x0.00000cp-126f, -0x000000cL, -150), MAKE_HEX_FLOAT(-0x0.00000ap-126f, -0x000000aL, -150),
MAKE_HEX_FLOAT(-0x0.000008p-126f, -0x0000008L, -150), MAKE_HEX_FLOAT(-0x0.000006p-126f, -0x0000006L, -150), MAKE_HEX_FLOAT(-0x0.000004p-126f, -0x0000004L, -150), MAKE_HEX_FLOAT(-0x0.000002p-126f, -0x0000002L, -150), -0.0f,
+NAN, +INFINITY, +FLT_MAX, MAKE_HEX_FLOAT(+0x1.000002p64f, +0x1000002L, 40), MAKE_HEX_FLOAT(+0x1.0p64f, +0x1L, 64), MAKE_HEX_FLOAT(+0x1.fffffep63f, +0x1fffffeL, 39), MAKE_HEX_FLOAT(+0x1.000002p63f, +0x1000002L, 39), MAKE_HEX_FLOAT(+0x1.0p63f, +0x1L, 63), MAKE_HEX_FLOAT(+0x1.fffffep62f, +0x1fffffeL, 38),
MAKE_HEX_FLOAT(+0x1.000002p32f, +0x1000002L, 8), MAKE_HEX_FLOAT(+0x1.0p32f, +0x1L, 32), MAKE_HEX_FLOAT(+0x1.fffffep31f, +0x1fffffeL, 7), MAKE_HEX_FLOAT(+0x1.000002p31f, +0x1000002L, 7), MAKE_HEX_FLOAT(+0x1.0p31f, +0x1L, 31), MAKE_HEX_FLOAT(+0x1.fffffep30f, +0x1fffffeL, 6), +1000.f, +100.f, +4.0f, +3.5f,
+3.0f, MAKE_HEX_FLOAT(+0x1.800002p1f, +0x1800002L, -23), 2.5f, MAKE_HEX_FLOAT(+0x1.7ffffep1f, +0x17ffffeL, -23),+2.0f, MAKE_HEX_FLOAT(+0x1.800002p0f, +0x1800002L, -24), 1.5f, MAKE_HEX_FLOAT(+0x1.7ffffep0f, +0x17ffffeL, -24), MAKE_HEX_FLOAT(+0x1.000002p0f, +0x1000002L, -24), +1.0f, MAKE_HEX_FLOAT(+0x1.fffffep-1f, +0x1fffffeL, -25),
MAKE_HEX_FLOAT(+0x1.000002p-1f, +0x1000002L, -25), +0.5f, MAKE_HEX_FLOAT(+0x1.fffffep-2f, +0x1fffffeL, -26), MAKE_HEX_FLOAT(+0x1.000002p-2f, +0x1000002L, -26), +0.25f, MAKE_HEX_FLOAT(+0x1.fffffep-3f, +0x1fffffeL, -27),
MAKE_HEX_FLOAT(0x1.000002p-126f, 0x1000002L, -150), +FLT_MIN, MAKE_HEX_FLOAT(+0x0.fffffep-126f, +0x0fffffeL, -150), MAKE_HEX_FLOAT(+0x0.000ffep-126f, +0x0000ffeL, -150), MAKE_HEX_FLOAT(+0x0.0000fep-126f, +0x00000feL, -150), MAKE_HEX_FLOAT(+0x0.00000ep-126f, +0x000000eL, -150), MAKE_HEX_FLOAT(+0x0.00000cp-126f, +0x000000cL, -150), MAKE_HEX_FLOAT(+0x0.00000ap-126f, +0x000000aL, -150),
MAKE_HEX_FLOAT(+0x0.000008p-126f, +0x0000008L, -150), MAKE_HEX_FLOAT(+0x0.000006p-126f, +0x0000006L, -150), MAKE_HEX_FLOAT(+0x0.000004p-126f, +0x0000004L, -150), MAKE_HEX_FLOAT(+0x0.000002p-126f, +0x0000002L, -150), +0.0f
+NAN, +INFINITY, +FLT_MAX, MAKE_HEX_FLOAT(+0x1.000002p64f, +0x1000002L, 40), MAKE_HEX_FLOAT(+0x1.0p64f, +0x1L, 64), MAKE_HEX_FLOAT(+0x1.fffffep63f, +0x1fffffeL, 39), MAKE_HEX_FLOAT(+0x1.000002p63f, +0x1000002L, 39), MAKE_HEX_FLOAT(+0x1.0p63f, +0x1L, 63), MAKE_HEX_FLOAT(+0x1.fffffep62f, +0x1fffffeL, 38),
MAKE_HEX_FLOAT(+0x1.000002p32f, +0x1000002L, 8), MAKE_HEX_FLOAT(+0x1.0p32f, +0x1L, 32), MAKE_HEX_FLOAT(+0x1.fffffep31f, +0x1fffffeL, 7), MAKE_HEX_FLOAT(+0x1.000002p31f, +0x1000002L, 7), MAKE_HEX_FLOAT(+0x1.0p31f, +0x1L, 31), MAKE_HEX_FLOAT(+0x1.fffffep30f, +0x1fffffeL, 6), +1000.f, +100.f, +4.0f, +3.5f,
+3.0f, MAKE_HEX_FLOAT(+0x1.800002p1f, +0x1800002L, -23), 2.5f, MAKE_HEX_FLOAT(+0x1.7ffffep1f, +0x17ffffeL, -23),+2.0f, MAKE_HEX_FLOAT(+0x1.800002p0f, +0x1800002L, -24), 1.5f, MAKE_HEX_FLOAT(+0x1.7ffffep0f, +0x17ffffeL, -24), MAKE_HEX_FLOAT(+0x1.000002p0f, +0x1000002L, -24), +1.0f, MAKE_HEX_FLOAT(+0x1.fffffep-1f, +0x1fffffeL, -25),
MAKE_HEX_FLOAT(+0x1.000002p-1f, +0x1000002L, -25), +0.5f, MAKE_HEX_FLOAT(+0x1.fffffep-2f, +0x1fffffeL, -26), MAKE_HEX_FLOAT(+0x1.000002p-2f, +0x1000002L, -26), +0.25f, MAKE_HEX_FLOAT(+0x1.fffffep-3f, +0x1fffffeL, -27),
MAKE_HEX_FLOAT(0x1.000002p-126f, 0x1000002L, -150), +FLT_MIN, MAKE_HEX_FLOAT(+0x0.fffffep-126f, +0x0fffffeL, -150), MAKE_HEX_FLOAT(+0x0.000ffep-126f, +0x0000ffeL, -150), MAKE_HEX_FLOAT(+0x0.0000fep-126f, +0x00000feL, -150), MAKE_HEX_FLOAT(+0x0.00000ep-126f, +0x000000eL, -150), MAKE_HEX_FLOAT(+0x0.00000cp-126f, +0x000000cL, -150), MAKE_HEX_FLOAT(+0x0.00000ap-126f, +0x000000aL, -150),
MAKE_HEX_FLOAT(+0x0.000008p-126f, +0x0000008L, -150), MAKE_HEX_FLOAT(+0x0.000006p-126f, +0x0000006L, -150), MAKE_HEX_FLOAT(+0x0.000004p-126f, +0x0000004L, -150), MAKE_HEX_FLOAT(+0x0.000002p-126f, +0x0000002L, -150), +0.0f
};
static size_t specialValuesFloatCount = sizeof( specialValuesFloat ) / sizeof( specialValuesFloat[0] );
static const int specialValuesInt[] = { 0, 1, 2, 3, 126, 127, 128, 0x02000001, 0x04000001, 1465264071, 1488522147,
static const int specialValuesInt[] = { 0, 1, 2, 3, 126, 127, 128, 0x02000001, 0x04000001, 1465264071, 1488522147,
-1, -2, -3, -126, -127, -128, -0x02000001, -0x04000001, -1465264071, -1488522147 };
static size_t specialValuesIntCount = sizeof( specialValuesInt ) / sizeof( specialValuesInt[0] );
@@ -327,11 +327,11 @@ int TestFunc_Float_Float_Int(const Func *f, MTdata d)
BuildKernelInfo build_info = { gMinVectorSizeIndex, test_info.threadCount, test_info.k, test_info.programs, f->nameInCode };
if( (error = ThreadPool_Do( BuildKernel_FloatFn, gMaxVectorSizeIndex - gMinVectorSizeIndex, &build_info ) ))
goto exit;
// Run the kernels
error = ThreadPool_Do( TestFloat, (cl_uint) ((1ULL<<32) / test_info.step), &test_info );
// Accumulate the arithmetic errors
for( i = 0; i < test_info.threadCount; i++ )
{
@@ -342,7 +342,7 @@ int TestFunc_Float_Float_Int(const Func *f, MTdata d)
maxErrorVal2 = test_info.tinfo[i].maxErrorValue2;
}
}
if( error )
goto exit;
@@ -353,7 +353,7 @@ int TestFunc_Float_Float_Int(const Func *f, MTdata d)
else
vlog( "passed." );
}
if( gMeasureTimes )
{
@@ -404,7 +404,7 @@ int TestFunc_Float_Float_Int(const Func *f, MTdata d)
vlog_error( "Error %d at clFinish\n", error );
goto exit;
}
uint64_t endTime = GetTime();
double time = SubtractTime( endTime, startTime );
sum += time;
@@ -418,7 +418,7 @@ int TestFunc_Float_Float_Int(const Func *f, MTdata d)
vlog_perf( clocksPerOp, LOWER_IS_BETTER, "clocks / element", "%sf%s", f->name, sizeNames[j] );
}
}
if( ! gSkipCorrectnessTesting )
vlog( "\t%8.2f @ {%a, %d}", maxError, maxErrorVal, maxErrorVal2 );
vlog( "\n" );
@@ -447,7 +447,7 @@ exit:
clReleaseMemObject(test_info.tinfo[i].outBuf[j]);
clReleaseCommandQueue(test_info.tinfo[i].tQueue);
}
free( test_info.tinfo );
}
@@ -480,28 +480,28 @@ static cl_int TestFloat( cl_uint job_id, cl_uint thread_id, void *data )
{
vlog_error( "Error: clEnqueueMapBuffer %d failed! err: %d\n", j, error );
return error;
}
}
}
// Get that moving
if( (error = clFlush(tinfo->tQueue) ))
vlog( "clFlush failed\n" );
//Init input array
cl_uint *p = (cl_uint *)gIn + thread_id * buffer_elements;
cl_uint *p2 = (cl_uint *)gIn2 + thread_id * buffer_elements;
j = 0;
int totalSpecialValueCount = specialValuesFloatCount * specialValuesIntCount;
int indx = (totalSpecialValueCount - 1) / buffer_elements;
if( job_id <= (cl_uint)indx )
int totalSpecialValueCount = specialValuesFloatCount * specialValuesIntCount;
int indx = (totalSpecialValueCount - 1) / buffer_elements;
if( job_id <= (cl_uint)indx )
{ // test edge cases
float *fp = (float *)p;
cl_int *ip2 = (cl_int *)p2;
uint32_t x, y;
x = (job_id * buffer_elements) % specialValuesFloatCount;
y = (job_id * buffer_elements) / specialValuesFloatCount;
uint32_t x, y;
x = (job_id * buffer_elements) % specialValuesFloatCount;
y = (job_id * buffer_elements) / specialValuesFloatCount;
for( ; j < buffer_elements; j++ )
{
fp[j] = specialValuesFloat[x];
@@ -522,7 +522,7 @@ static cl_int TestFloat( cl_uint job_id, cl_uint thread_id, void *data )
p[j] = genrand_int32(d);
p2[j] = genrand_int32(d);
}
if( (error = clEnqueueWriteBuffer( tinfo->tQueue, tinfo->inBuf, CL_FALSE, 0, buffer_size, p, 0, NULL, NULL) ))
{
vlog_error( "Error: clEnqueueWriteBuffer failed! err: %d\n", error );
@@ -540,15 +540,15 @@ static cl_int TestFloat( cl_uint job_id, cl_uint thread_id, void *data )
//Wait for the map to finish
if( (error = clWaitForEvents(1, e + j) ))
{
vlog_error( "Error: clWaitForEvents failed! err: %d\n", error );
vlog_error( "Error: clWaitForEvents failed! err: %d\n", error );
goto exit;
}
if( (error = clReleaseEvent( e[j] ) ))
{
vlog_error( "Error: clReleaseEvent failed! err: %d\n", error );
vlog_error( "Error: clReleaseEvent failed! err: %d\n", error );
goto exit;
}
// Fill the result buffer with garbage, so that old results don't carry over
uint32_t pattern = 0xffffdead;
memset_pattern4(out[j], &pattern, buffer_size);
@@ -562,7 +562,7 @@ static cl_int TestFloat( cl_uint job_id, cl_uint thread_id, void *data )
size_t vectorCount = (buffer_elements + sizeValues[j] - 1) / sizeValues[j];
cl_kernel kernel = job->k[j][thread_id]; //each worker thread has its own copy of the cl_kernel
cl_program program = job->programs[j];
if( ( error = clSetKernelArg( kernel, 0, sizeof( tinfo->outBuf[j] ), &tinfo->outBuf[j] ))){ LogBuildError(program); return error; }
if( ( error = clSetKernelArg( kernel, 1, sizeof( tinfo->inBuf ), &tinfo->inBuf ) )) { LogBuildError(program); return error; }
if( ( error = clSetKernelArg( kernel, 2, sizeof( tinfo->inBuf2 ), &tinfo->inBuf2 ) )) { LogBuildError(program); return error; }
@@ -596,7 +596,7 @@ static cl_int TestFloat( cl_uint job_id, cl_uint thread_id, void *data )
{
vlog_error( "Error: clEnqueueMapBuffer %d failed! err: %d\n", j, error );
goto exit;
}
}
}
// Wait for the last buffer
@@ -605,7 +605,7 @@ static cl_int TestFloat( cl_uint job_id, cl_uint thread_id, void *data )
{
vlog_error( "Error: clEnqueueMapBuffer %d failed! err: %d\n", j, error );
goto exit;
}
}
//Verify data
cl_uint *t = (cl_uint *)r;
@@ -614,7 +614,7 @@ static cl_int TestFloat( cl_uint job_id, cl_uint thread_id, void *data )
for( k = gMinVectorSizeIndex; k < gMaxVectorSizeIndex; k++ )
{
cl_uint *q = out[k];
// If we aren't getting the correctly rounded result
if( t[j] != q[j] )
{
@@ -622,7 +622,7 @@ static cl_int TestFloat( cl_uint job_id, cl_uint thread_id, void *data )
double correct = func.f_fi( s[j], s2[j] );
float err = Ulp_Error( test, correct );
int fail = ! (fabsf(err) <= ulps);
if( fail && ftz )
{
// retry per section 6.5.3.2
@@ -647,7 +647,7 @@ static cl_int TestFloat( cl_uint job_id, cl_uint thread_id, void *data )
err = err2;
if( fabsf( err3 ) < fabsf(err ) )
err = err3;
// retry per section 6.5.3.4
if( IsFloatResultSubnormal( correct2, ulps ) || IsFloatResultSubnormal( correct3, ulps ) )
{
@@ -657,12 +657,12 @@ static cl_int TestFloat( cl_uint job_id, cl_uint thread_id, void *data )
}
}
}
if( fabsf(err ) > tinfo->maxError )
{
tinfo->maxError = fabsf(err);
tinfo->maxErrorValue = s[j];
tinfo->maxErrorValue2 = s2[j];
tinfo->maxErrorValue = s[j];
tinfo->maxErrorValue2 = s2[j];
}
if( fail )
{
@@ -680,9 +680,9 @@ static cl_int TestFloat( cl_uint job_id, cl_uint thread_id, void *data )
{
vlog_error( "Error: clEnqueueUnmapMemObject %d failed 2! err: %d\n", j, error );
return error;
}
}
}
if( (error = clFlush(tinfo->tQueue) ))
vlog( "clFlush 3 failed\n" );
@@ -695,32 +695,32 @@ static cl_int TestFloat( cl_uint job_id, cl_uint thread_id, void *data )
exit:
return error;
}
// A table of more difficult cases to get right
static const double specialValuesDouble[] = {
-NAN, -INFINITY, -DBL_MAX, MAKE_HEX_DOUBLE(-0x1.0000000000001p64, -0x10000000000001LL, 12), MAKE_HEX_DOUBLE(-0x1.0p64, -0x1LL, 64), MAKE_HEX_DOUBLE(-0x1.fffffffffffffp63, -0x1fffffffffffffLL, 11), MAKE_HEX_DOUBLE(-0x1.0000000000001p63, -0x10000000000001LL, 11), MAKE_HEX_DOUBLE(-0x1.0p63, -0x1LL, 63), MAKE_HEX_DOUBLE(-0x1.fffffffffffffp62, -0x1fffffffffffffLL, 10),
MAKE_HEX_DOUBLE(-0x1.000002p32, -0x1000002LL, 8), MAKE_HEX_DOUBLE(-0x1.0p32, -0x1LL, 32), MAKE_HEX_DOUBLE(-0x1.fffffffffffffp31, -0x1fffffffffffffLL, -21), MAKE_HEX_DOUBLE(-0x1.0000000000001p31, -0x10000000000001LL, -21), MAKE_HEX_DOUBLE(-0x1.0p31, -0x1LL, 31), MAKE_HEX_DOUBLE(-0x1.fffffffffffffp30, -0x1fffffffffffffLL, -22), -1000., -100., -4.0, -3.5,
-3.0, MAKE_HEX_DOUBLE(-0x1.8000000000001p1, -0x18000000000001LL, -51), -2.5, MAKE_HEX_DOUBLE(-0x1.7ffffffffffffp1, -0x17ffffffffffffLL, -51), -2.0, MAKE_HEX_DOUBLE(-0x1.8000000000001p0, -0x18000000000001LL, -52), -1.5, MAKE_HEX_DOUBLE(-0x1.7ffffffffffffp0, -0x17ffffffffffffLL, -52),MAKE_HEX_DOUBLE(-0x1.0000000000001p0, -0x10000000000001LL, -52), -1.0, MAKE_HEX_DOUBLE(-0x1.fffffffffffffp-1, -0x1fffffffffffffLL, -53),
MAKE_HEX_DOUBLE(-0x1.0000000000001p-1, -0x10000000000001LL, -53), -0.5, MAKE_HEX_DOUBLE(-0x1.fffffffffffffp-2, -0x1fffffffffffffLL, -54), MAKE_HEX_DOUBLE(-0x1.0000000000001p-2, -0x10000000000001LL, -54), -0.25, MAKE_HEX_DOUBLE(-0x1.fffffffffffffp-3, -0x1fffffffffffffLL, -55),
MAKE_HEX_DOUBLE(-0x1.0000000000001p-1022, -0x10000000000001LL, -1074), -DBL_MIN, MAKE_HEX_DOUBLE(-0x0.fffffffffffffp-1022, -0x0fffffffffffffLL, -1074), MAKE_HEX_DOUBLE(-0x0.0000000000fffp-1022, -0x00000000000fffLL, -1074), MAKE_HEX_DOUBLE(-0x0.00000000000fep-1022, -0x000000000000feLL, -1074), MAKE_HEX_DOUBLE(-0x0.000000000000ep-1022, -0x0000000000000eLL, -1074), MAKE_HEX_DOUBLE(-0x0.000000000000cp-1022, -0x0000000000000cLL, -1074), MAKE_HEX_DOUBLE(-0x0.000000000000ap-1022, -0x0000000000000aLL, -1074),
MAKE_HEX_DOUBLE(-0x0.0000000000008p-1022, -0x00000000000008LL, -1074), MAKE_HEX_DOUBLE(-0x0.0000000000007p-1022, -0x00000000000007LL, -1074), MAKE_HEX_DOUBLE(-0x0.0000000000006p-1022, -0x00000000000006LL, -1074), MAKE_HEX_DOUBLE(-0x0.0000000000005p-1022, -0x00000000000005LL, -1074), MAKE_HEX_DOUBLE(-0x0.0000000000004p-1022, -0x00000000000004LL, -1074),
static const double specialValuesDouble[] = {
-NAN, -INFINITY, -DBL_MAX, MAKE_HEX_DOUBLE(-0x1.0000000000001p64, -0x10000000000001LL, 12), MAKE_HEX_DOUBLE(-0x1.0p64, -0x1LL, 64), MAKE_HEX_DOUBLE(-0x1.fffffffffffffp63, -0x1fffffffffffffLL, 11), MAKE_HEX_DOUBLE(-0x1.0000000000001p63, -0x10000000000001LL, 11), MAKE_HEX_DOUBLE(-0x1.0p63, -0x1LL, 63), MAKE_HEX_DOUBLE(-0x1.fffffffffffffp62, -0x1fffffffffffffLL, 10),
MAKE_HEX_DOUBLE(-0x1.000002p32, -0x1000002LL, 8), MAKE_HEX_DOUBLE(-0x1.0p32, -0x1LL, 32), MAKE_HEX_DOUBLE(-0x1.fffffffffffffp31, -0x1fffffffffffffLL, -21), MAKE_HEX_DOUBLE(-0x1.0000000000001p31, -0x10000000000001LL, -21), MAKE_HEX_DOUBLE(-0x1.0p31, -0x1LL, 31), MAKE_HEX_DOUBLE(-0x1.fffffffffffffp30, -0x1fffffffffffffLL, -22), -1000., -100., -4.0, -3.5,
-3.0, MAKE_HEX_DOUBLE(-0x1.8000000000001p1, -0x18000000000001LL, -51), -2.5, MAKE_HEX_DOUBLE(-0x1.7ffffffffffffp1, -0x17ffffffffffffLL, -51), -2.0, MAKE_HEX_DOUBLE(-0x1.8000000000001p0, -0x18000000000001LL, -52), -1.5, MAKE_HEX_DOUBLE(-0x1.7ffffffffffffp0, -0x17ffffffffffffLL, -52),MAKE_HEX_DOUBLE(-0x1.0000000000001p0, -0x10000000000001LL, -52), -1.0, MAKE_HEX_DOUBLE(-0x1.fffffffffffffp-1, -0x1fffffffffffffLL, -53),
MAKE_HEX_DOUBLE(-0x1.0000000000001p-1, -0x10000000000001LL, -53), -0.5, MAKE_HEX_DOUBLE(-0x1.fffffffffffffp-2, -0x1fffffffffffffLL, -54), MAKE_HEX_DOUBLE(-0x1.0000000000001p-2, -0x10000000000001LL, -54), -0.25, MAKE_HEX_DOUBLE(-0x1.fffffffffffffp-3, -0x1fffffffffffffLL, -55),
MAKE_HEX_DOUBLE(-0x1.0000000000001p-1022, -0x10000000000001LL, -1074), -DBL_MIN, MAKE_HEX_DOUBLE(-0x0.fffffffffffffp-1022, -0x0fffffffffffffLL, -1074), MAKE_HEX_DOUBLE(-0x0.0000000000fffp-1022, -0x00000000000fffLL, -1074), MAKE_HEX_DOUBLE(-0x0.00000000000fep-1022, -0x000000000000feLL, -1074), MAKE_HEX_DOUBLE(-0x0.000000000000ep-1022, -0x0000000000000eLL, -1074), MAKE_HEX_DOUBLE(-0x0.000000000000cp-1022, -0x0000000000000cLL, -1074), MAKE_HEX_DOUBLE(-0x0.000000000000ap-1022, -0x0000000000000aLL, -1074),
MAKE_HEX_DOUBLE(-0x0.0000000000008p-1022, -0x00000000000008LL, -1074), MAKE_HEX_DOUBLE(-0x0.0000000000007p-1022, -0x00000000000007LL, -1074), MAKE_HEX_DOUBLE(-0x0.0000000000006p-1022, -0x00000000000006LL, -1074), MAKE_HEX_DOUBLE(-0x0.0000000000005p-1022, -0x00000000000005LL, -1074), MAKE_HEX_DOUBLE(-0x0.0000000000004p-1022, -0x00000000000004LL, -1074),
MAKE_HEX_DOUBLE(-0x0.0000000000003p-1022, -0x00000000000003LL, -1074), MAKE_HEX_DOUBLE(-0x0.0000000000002p-1022, -0x00000000000002LL, -1074), MAKE_HEX_DOUBLE(-0x0.0000000000001p-1022, -0x00000000000001LL, -1074), -0.0,
+NAN, +INFINITY, +DBL_MAX, MAKE_HEX_DOUBLE(+0x1.0000000000001p64, +0x10000000000001LL, 12), MAKE_HEX_DOUBLE(+0x1.0p64, +0x1LL, 64), MAKE_HEX_DOUBLE(+0x1.fffffffffffffp63, +0x1fffffffffffffLL, 11), MAKE_HEX_DOUBLE(+0x1.0000000000001p63, +0x10000000000001LL, 11), MAKE_HEX_DOUBLE(+0x1.0p63, +0x1LL, 63), MAKE_HEX_DOUBLE(+0x1.fffffffffffffp62, +0x1fffffffffffffLL, 10),
MAKE_HEX_DOUBLE(+0x1.000002p32, +0x1000002LL, 8), MAKE_HEX_DOUBLE(+0x1.0p32, +0x1LL, 32), MAKE_HEX_DOUBLE(+0x1.fffffffffffffp31, +0x1fffffffffffffLL, -21), MAKE_HEX_DOUBLE(+0x1.0000000000001p31, +0x10000000000001LL, -21), MAKE_HEX_DOUBLE(+0x1.0p31, +0x1LL, 31), MAKE_HEX_DOUBLE(+0x1.fffffffffffffp30, +0x1fffffffffffffLL, -22), +1000., +100., +4.0, +3.5,
+3.0, MAKE_HEX_DOUBLE(+0x1.8000000000001p1, +0x18000000000001LL, -51), +2.5, MAKE_HEX_DOUBLE(+0x1.7ffffffffffffp1, +0x17ffffffffffffLL, -51), +2.0, MAKE_HEX_DOUBLE(+0x1.8000000000001p0, +0x18000000000001LL, -52), +1.5, MAKE_HEX_DOUBLE(+0x1.7ffffffffffffp0, +0x17ffffffffffffLL, -52),MAKE_HEX_DOUBLE(-0x1.0000000000001p0, -0x10000000000001LL, -52), +1.0, MAKE_HEX_DOUBLE(+0x1.fffffffffffffp-1, +0x1fffffffffffffLL, -53),
MAKE_HEX_DOUBLE(+0x1.0000000000001p-1, +0x10000000000001LL, -53), +0.5, MAKE_HEX_DOUBLE(+0x1.fffffffffffffp-2, +0x1fffffffffffffLL, -54), MAKE_HEX_DOUBLE(+0x1.0000000000001p-2, +0x10000000000001LL, -54), +0.25, MAKE_HEX_DOUBLE(+0x1.fffffffffffffp-3, +0x1fffffffffffffLL, -55),
MAKE_HEX_DOUBLE(+0x1.0000000000001p-1022, +0x10000000000001LL, -1074), +DBL_MIN, MAKE_HEX_DOUBLE(+0x0.fffffffffffffp-1022, +0x0fffffffffffffLL, -1074), MAKE_HEX_DOUBLE(+0x0.0000000000fffp-1022, +0x00000000000fffLL, -1074), MAKE_HEX_DOUBLE(+0x0.00000000000fep-1022, +0x000000000000feLL, -1074), MAKE_HEX_DOUBLE(+0x0.000000000000ep-1022, +0x0000000000000eLL, -1074), MAKE_HEX_DOUBLE(+0x0.000000000000cp-1022, +0x0000000000000cLL, -1074), MAKE_HEX_DOUBLE(+0x0.000000000000ap-1022, +0x0000000000000aLL, -1074),
MAKE_HEX_DOUBLE(+0x0.0000000000008p-1022, +0x00000000000008LL, -1074), MAKE_HEX_DOUBLE(+0x0.0000000000007p-1022, +0x00000000000007LL, -1074), MAKE_HEX_DOUBLE(+0x0.0000000000006p-1022, +0x00000000000006LL, -1074), MAKE_HEX_DOUBLE(+0x0.0000000000005p-1022, +0x00000000000005LL, -1074), MAKE_HEX_DOUBLE(+0x0.0000000000004p-1022, +0x00000000000004LL, -1074),
+NAN, +INFINITY, +DBL_MAX, MAKE_HEX_DOUBLE(+0x1.0000000000001p64, +0x10000000000001LL, 12), MAKE_HEX_DOUBLE(+0x1.0p64, +0x1LL, 64), MAKE_HEX_DOUBLE(+0x1.fffffffffffffp63, +0x1fffffffffffffLL, 11), MAKE_HEX_DOUBLE(+0x1.0000000000001p63, +0x10000000000001LL, 11), MAKE_HEX_DOUBLE(+0x1.0p63, +0x1LL, 63), MAKE_HEX_DOUBLE(+0x1.fffffffffffffp62, +0x1fffffffffffffLL, 10),
MAKE_HEX_DOUBLE(+0x1.000002p32, +0x1000002LL, 8), MAKE_HEX_DOUBLE(+0x1.0p32, +0x1LL, 32), MAKE_HEX_DOUBLE(+0x1.fffffffffffffp31, +0x1fffffffffffffLL, -21), MAKE_HEX_DOUBLE(+0x1.0000000000001p31, +0x10000000000001LL, -21), MAKE_HEX_DOUBLE(+0x1.0p31, +0x1LL, 31), MAKE_HEX_DOUBLE(+0x1.fffffffffffffp30, +0x1fffffffffffffLL, -22), +1000., +100., +4.0, +3.5,
+3.0, MAKE_HEX_DOUBLE(+0x1.8000000000001p1, +0x18000000000001LL, -51), +2.5, MAKE_HEX_DOUBLE(+0x1.7ffffffffffffp1, +0x17ffffffffffffLL, -51), +2.0, MAKE_HEX_DOUBLE(+0x1.8000000000001p0, +0x18000000000001LL, -52), +1.5, MAKE_HEX_DOUBLE(+0x1.7ffffffffffffp0, +0x17ffffffffffffLL, -52),MAKE_HEX_DOUBLE(-0x1.0000000000001p0, -0x10000000000001LL, -52), +1.0, MAKE_HEX_DOUBLE(+0x1.fffffffffffffp-1, +0x1fffffffffffffLL, -53),
MAKE_HEX_DOUBLE(+0x1.0000000000001p-1, +0x10000000000001LL, -53), +0.5, MAKE_HEX_DOUBLE(+0x1.fffffffffffffp-2, +0x1fffffffffffffLL, -54), MAKE_HEX_DOUBLE(+0x1.0000000000001p-2, +0x10000000000001LL, -54), +0.25, MAKE_HEX_DOUBLE(+0x1.fffffffffffffp-3, +0x1fffffffffffffLL, -55),
MAKE_HEX_DOUBLE(+0x1.0000000000001p-1022, +0x10000000000001LL, -1074), +DBL_MIN, MAKE_HEX_DOUBLE(+0x0.fffffffffffffp-1022, +0x0fffffffffffffLL, -1074), MAKE_HEX_DOUBLE(+0x0.0000000000fffp-1022, +0x00000000000fffLL, -1074), MAKE_HEX_DOUBLE(+0x0.00000000000fep-1022, +0x000000000000feLL, -1074), MAKE_HEX_DOUBLE(+0x0.000000000000ep-1022, +0x0000000000000eLL, -1074), MAKE_HEX_DOUBLE(+0x0.000000000000cp-1022, +0x0000000000000cLL, -1074), MAKE_HEX_DOUBLE(+0x0.000000000000ap-1022, +0x0000000000000aLL, -1074),
MAKE_HEX_DOUBLE(+0x0.0000000000008p-1022, +0x00000000000008LL, -1074), MAKE_HEX_DOUBLE(+0x0.0000000000007p-1022, +0x00000000000007LL, -1074), MAKE_HEX_DOUBLE(+0x0.0000000000006p-1022, +0x00000000000006LL, -1074), MAKE_HEX_DOUBLE(+0x0.0000000000005p-1022, +0x00000000000005LL, -1074), MAKE_HEX_DOUBLE(+0x0.0000000000004p-1022, +0x00000000000004LL, -1074),
MAKE_HEX_DOUBLE(+0x0.0000000000003p-1022, +0x00000000000003LL, -1074), MAKE_HEX_DOUBLE(+0x0.0000000000002p-1022, +0x00000000000002LL, -1074), MAKE_HEX_DOUBLE(+0x0.0000000000001p-1022, +0x00000000000001LL, -1074), +0.0,
};
static size_t specialValuesDoubleCount = sizeof( specialValuesDouble ) / sizeof( specialValuesDouble[0] );
static const int specialValuesInt2[] = { 0, 1, 2, 3, 1022, 1023, 1024, INT_MIN, INT_MAX,
static const int specialValuesInt2[] = { 0, 1, 2, 3, 1022, 1023, 1024, INT_MIN, INT_MAX,
-1, -2, -3, -1022, -1023, -11024, -INT_MAX };
static size_t specialValuesInt2Count = sizeof( specialValuesInt ) / sizeof( specialValuesInt[0] );
@@ -746,7 +746,7 @@ int TestFunc_Double_Double_Int(const Func *f, MTdata d)
test_info.f = f;
test_info.ulps = f->double_ulps;
test_info.ftz = f->ftz || gForceFTZ;
// cl_kernels aren't thread safe, so we make one for each vector size for every thread
for( i = gMinVectorSizeIndex; i < gMaxVectorSizeIndex; i++ )
{
@@ -809,12 +809,12 @@ int TestFunc_Double_Double_Int(const Func *f, MTdata d)
BuildKernelInfo build_info = { gMinVectorSizeIndex, test_info.threadCount, test_info.k, test_info.programs, f->nameInCode };
if( (error = ThreadPool_Do( BuildKernel_DoubleFn, gMaxVectorSizeIndex - gMinVectorSizeIndex, &build_info ) ))
goto exit;
// Run the kernels
if( !gSkipCorrectnessTesting )
error = ThreadPool_Do( TestDouble, (cl_uint) ((1ULL<<32) / test_info.step), &test_info );
// Accumulate the arithmetic errors
for( i = 0; i < test_info.threadCount; i++ )
{
@@ -825,7 +825,7 @@ int TestFunc_Double_Double_Int(const Func *f, MTdata d)
maxErrorVal2 = test_info.tinfo[i].maxErrorValue2;
}
}
if( error )
goto exit;
@@ -836,7 +836,7 @@ int TestFunc_Double_Double_Int(const Func *f, MTdata d)
else
vlog( "passed." );
}
if( gMeasureTimes )
{
//Init input arrays
@@ -886,7 +886,7 @@ int TestFunc_Double_Double_Int(const Func *f, MTdata d)
vlog_error( "Error %d at clFinish\n", error );
goto exit;
}
uint64_t endTime = GetTime();
double time = SubtractTime( endTime, startTime );
sum += time;
@@ -902,14 +902,14 @@ int TestFunc_Double_Double_Int(const Func *f, MTdata d)
for( ; j < gMaxVectorSizeIndex; j++ )
vlog( "\t -- " );
}
if( ! gSkipCorrectnessTesting )
vlog( "\t%8.2f @ {%a, %d}", maxError, maxErrorVal, maxErrorVal2 );
vlog( "\n" );
exit:
// Release
// Release
for( i = gMinVectorSizeIndex; i < gMaxVectorSizeIndex; i++ )
{
clReleaseProgram(test_info.programs[i]);
@@ -932,7 +932,7 @@ exit:
clReleaseMemObject(test_info.tinfo[i].outBuf[j]);
clReleaseCommandQueue(test_info.tinfo[i].tQueue);
}
free( test_info.tinfo );
}
@@ -966,28 +966,28 @@ static cl_int TestDouble( cl_uint job_id, cl_uint thread_id, void *data )
{
vlog_error( "Error: clEnqueueMapBuffer %d failed! err: %d\n", j, error );
return error;
}
}
}
// Get that moving
if( (error = clFlush(tinfo->tQueue) ))
vlog( "clFlush failed\n" );
//Init input array
cl_ulong *p = (cl_ulong *)gIn + thread_id * buffer_elements;
cl_int *p2 = (cl_int *)gIn2 + thread_id * buffer_elements;
j = 0;
int totalSpecialValueCount = specialValuesDoubleCount * specialValuesInt2Count;
int indx = (totalSpecialValueCount - 1) / buffer_elements;
if( job_id <= (cl_uint)indx )
int totalSpecialValueCount = specialValuesDoubleCount * specialValuesInt2Count;
int indx = (totalSpecialValueCount - 1) / buffer_elements;
if( job_id <= (cl_uint)indx )
{ // test edge cases
cl_double *fp = (cl_double *)p;
cl_int *ip2 = (cl_int *)p2;
uint32_t x, y;
x = (job_id * buffer_elements) % specialValuesDoubleCount;
y = (job_id * buffer_elements) / specialValuesDoubleCount;
uint32_t x, y;
x = (job_id * buffer_elements) % specialValuesDoubleCount;
y = (job_id * buffer_elements) / specialValuesDoubleCount;
for( ; j < buffer_elements; j++ )
{
fp[j] = specialValuesDouble[x];
@@ -1008,7 +1008,7 @@ static cl_int TestDouble( cl_uint job_id, cl_uint thread_id, void *data )
p[j] = DoubleFromUInt32(genrand_int32(d));
p2[j] = genrand_int32(d);
}
if( (error = clEnqueueWriteBuffer( tinfo->tQueue, tinfo->inBuf, CL_FALSE, 0, buffer_size, p, 0, NULL, NULL) ))
{
vlog_error( "Error: clEnqueueWriteBuffer failed! err: %d\n", error );
@@ -1026,15 +1026,15 @@ static cl_int TestDouble( cl_uint job_id, cl_uint thread_id, void *data )
//Wait for the map to finish
if( (error = clWaitForEvents(1, e + j) ))
{
vlog_error( "Error: clWaitForEvents failed! err: %d\n", error );
vlog_error( "Error: clWaitForEvents failed! err: %d\n", error );
goto exit;
}
if( (error = clReleaseEvent( e[j] ) ))
{
vlog_error( "Error: clReleaseEvent failed! err: %d\n", error );
vlog_error( "Error: clReleaseEvent failed! err: %d\n", error );
goto exit;
}
// Fill the result buffer with garbage, so that old results don't carry over
uint32_t pattern = 0xffffdead;
memset_pattern4(out[j], &pattern, buffer_size);
@@ -1048,7 +1048,7 @@ static cl_int TestDouble( cl_uint job_id, cl_uint thread_id, void *data )
size_t vectorCount = (buffer_elements + sizeValues[j] - 1) / sizeValues[j];
cl_kernel kernel = job->k[j][thread_id]; //each worker thread has its own copy of the cl_kernel
cl_program program = job->programs[j];
if( ( error = clSetKernelArg( kernel, 0, sizeof( tinfo->outBuf[j] ), &tinfo->outBuf[j] ))){ LogBuildError(program); return error; }
if( ( error = clSetKernelArg( kernel, 1, sizeof( tinfo->inBuf ), &tinfo->inBuf ) )) { LogBuildError(program); return error; }
if( ( error = clSetKernelArg( kernel, 2, sizeof( tinfo->inBuf2 ), &tinfo->inBuf2 ) )) { LogBuildError(program); return error; }
@@ -1066,14 +1066,14 @@ static cl_int TestDouble( cl_uint job_id, cl_uint thread_id, void *data )
if( gSkipCorrectnessTesting )
return CL_SUCCESS;
//Calculate the correctly rounded reference result
cl_double *r = (cl_double *)gOut_Ref + thread_id * buffer_elements;
cl_double *s = (cl_double *)gIn + thread_id * buffer_elements;
cl_int *s2 = (cl_int *)gIn2 + thread_id * buffer_elements;
for( j = 0; j < buffer_elements; j++ )
r[j] = (cl_double) func.f_fi( s[j], s2[j] );
// Read the data back -- no need to wait for the first N-1 buffers. This is an in order queue.
for( j = gMinVectorSizeIndex; j + 1 < gMaxVectorSizeIndex; j++ )
{
@@ -1082,7 +1082,7 @@ static cl_int TestDouble( cl_uint job_id, cl_uint thread_id, void *data )
{
vlog_error( "Error: clEnqueueMapBuffer %d failed! err: %d\n", j, error );
goto exit;
}
}
}
// Wait for the last buffer
@@ -1091,7 +1091,7 @@ static cl_int TestDouble( cl_uint job_id, cl_uint thread_id, void *data )
{
vlog_error( "Error: clEnqueueMapBuffer %d failed! err: %d\n", j, error );
goto exit;
}
}
//Verify data
cl_ulong *t = (cl_ulong *)r;
@@ -1100,7 +1100,7 @@ static cl_int TestDouble( cl_uint job_id, cl_uint thread_id, void *data )
for( k = gMinVectorSizeIndex; k < gMaxVectorSizeIndex; k++ )
{
cl_ulong *q = out[k];
// If we aren't getting the correctly rounded result
if( t[j] != q[j] )
{
@@ -1108,7 +1108,7 @@ static cl_int TestDouble( cl_uint job_id, cl_uint thread_id, void *data )
long double correct = func.f_fi( s[j], s2[j] );
float err = Ulp_Error_Double( test, correct );
int fail = ! (fabsf(err) <= ulps);
if( fail && ftz )
{
// retry per section 6.5.3.2
@@ -1131,7 +1131,7 @@ static cl_int TestDouble( cl_uint job_id, cl_uint thread_id, void *data )
err = err2;
if( fabsf( err3 ) < fabsf(err ) )
err = err3;
// retry per section 6.5.3.4
if( IsDoubleResultSubnormal( correct2, ulps ) || IsDoubleResultSubnormal( correct3, ulps ) )
{
@@ -1141,12 +1141,12 @@ static cl_int TestDouble( cl_uint job_id, cl_uint thread_id, void *data )
}
}
}
if( fabsf(err ) > tinfo->maxError )
{
tinfo->maxError = fabsf(err);
tinfo->maxErrorValue = s[j];
tinfo->maxErrorValue2 = s2[j];
tinfo->maxErrorValue = s[j];
tinfo->maxErrorValue2 = s2[j];
}
if( fail )
{
@@ -1157,16 +1157,16 @@ static cl_int TestDouble( cl_uint job_id, cl_uint thread_id, void *data )
}
}
}
for( j = gMinVectorSizeIndex; j < gMaxVectorSizeIndex; j++ )
{
if( (error = clEnqueueUnmapMemObject( tinfo->tQueue, tinfo->outBuf[j], out[j], 0, NULL, NULL)) )
{
vlog_error( "Error: clEnqueueUnmapMemObject %d failed 2! err: %d\n", j, error );
return error;
}
}
}
if( (error = clFlush(tinfo->tQueue) ))
vlog( "clFlush 3 failed\n" );
@@ -1179,7 +1179,7 @@ static cl_int TestDouble( cl_uint job_id, cl_uint thread_id, void *data )
exit:
return error;
}