[Beignet] [PATCH] utests: add utest to double convert int16
rander
rander.wang at intel.com
Tue Mar 14 02:13:44 UTC 2017
Signed-off-by: rander <rander.wang at intel.com>
---
kernels/builtin_convert_double2int16.cl | 73 ++++++++++++++
utests/CMakeLists.txt | 3 +-
utests/builtin_convert_double2int16.cpp | 168 ++++++++++++++++++++++++++++++++
3 files changed, 243 insertions(+), 1 deletion(-)
create mode 100644 kernels/builtin_convert_double2int16.cl
create mode 100644 utests/builtin_convert_double2int16.cpp
diff --git a/kernels/builtin_convert_double2int16.cl b/kernels/builtin_convert_double2int16.cl
new file mode 100644
index 0000000..fb3baea
--- /dev/null
+++ b/kernels/builtin_convert_double2int16.cl
@@ -0,0 +1,73 @@
+#pragma OPENCL EXTENSION cl_khr_fp64 : enable
+
+__kernel void builtin_convert_double2int16(__global double *X,
+ __global short *Z,
+ __global ushort *uZ,
+ int max_input)
+{
+ int i = get_global_id(0);
+ int j;
+
+ for(j = 0; j < max_input; j++)
+ Z[i++] = convert_short(X[j]);
+
+ for(j = 0; j < max_input; j++)
+ Z[i++] = convert_short_sat(X[j]);
+
+ for(j = 0; j < max_input; j++)
+ Z[i++] = convert_short_rtz(X[j]);
+
+ for(j = 0; j < max_input; j++)
+ Z[i++] = convert_short_rtn(X[j]);
+
+ for(j = 0; j < max_input; j++)
+ Z[i++] = convert_short_rte(X[j]);
+
+ for(j = 0; j < max_input; j++)
+ Z[i++] = convert_short_rtp(X[j]);
+
+ for(j = 0; j < max_input; j++)
+ Z[i++] = convert_short_sat_rtz(X[j]);
+
+ for(j = 0; j < max_input; j++)
+ Z[i++] = convert_short_sat_rtn(X[j]);
+
+ for(j = 0; j < max_input; j++)
+ Z[i++] = convert_short_sat_rte(X[j]);
+
+ for(j = 0; j < max_input; j++)
+ Z[i++] = convert_short_sat_rtp(X[j]);
+
+ i = 0;
+ for(j = 0; j < max_input; j++)
+ uZ[i++] = convert_ushort(X[j]);
+
+ for(j = 0; j < max_input; j++)
+ uZ[i++] = convert_ushort_sat(X[j]);
+
+ for(j = 0; j < max_input; j++)
+ uZ[i++] = convert_ushort_rtz(X[j]);
+
+ for(j = 0; j < max_input; j++)
+ uZ[i++] = convert_ushort_rtn(X[j]);
+
+ for(j = 0; j < max_input; j++)
+ uZ[i++] = convert_ushort_rte(X[j]);
+
+ for(j = 0; j < max_input; j++)
+ uZ[i++] = convert_ushort_rtp(X[j]);
+
+ for(j = 0; j < max_input; j++)
+ uZ[i++] = convert_ushort_sat_rtz(X[j]);
+
+ for(j = 0; j < max_input; j++)
+ uZ[i++] = convert_ushort_sat_rtn(X[j]);
+
+ for(j = 0; j < max_input; j++)
+ uZ[i++] = convert_ushort_sat_rte(X[j]);
+
+ for(j = 0; j < max_input; j++)
+ uZ[i++] = convert_ushort_sat_rtp(X[j]);
+
+}
+
diff --git a/utests/CMakeLists.txt b/utests/CMakeLists.txt
index f12643f..c18f79d 100644
--- a/utests/CMakeLists.txt
+++ b/utests/CMakeLists.txt
@@ -300,7 +300,8 @@ set (utests_sources
compiler_device_enqueue.cpp
builtin_relation_fp64.cpp
builtin_commonFunc_fp64.cpp
- builtin_convert_double2int8.cpp)
+ builtin_convert_double2int8.cpp
+ builtin_convert_double2int16.cpp)
if (LLVM_VERSION_NODOT VERSION_GREATER 34)
SET(utests_sources
diff --git a/utests/builtin_convert_double2int16.cpp b/utests/builtin_convert_double2int16.cpp
new file mode 100644
index 0000000..1bfc687
--- /dev/null
+++ b/utests/builtin_convert_double2int16.cpp
@@ -0,0 +1,168 @@
+#include "utest_helper.hpp"
+#include <cmath>
+#include <algorithm>
+
+namespace{
+double doubleX[] = {
+ 0x1.0000000001p0,
+ -0x1.0000000001p2,
+ 0x1.1ffp8,
+ -0x1.1fffp8,
+ 0x1.0p7,
+ -0x1.0p7,
+ 0x1.fffp6,
+ -0x1.ffffp6,
+ 0x1.ffffp7,
+ 0x1.1p-64,
+ -0x1.ffffp7,
+ 0x1.00001p8,
+ -0x1.00001p8,
+ 0x1.00000001p5,
+ -0x1.00000001p5,
+ 0x1.0p-32,
+ -0x1.0p-32,
+ 0x1.000000001p16,
+ -0x1.000000001p16,
+ 0x1.ffffffffffp15,
+ -0x1.1p-64
+ -0x1.ffffffffffp15,
+ 0x1.fffffffffp8,
+ -0x1.fffffffffp8,
+ 0x1fffffp128,
+ -0x1ffffffp128,
+ 0x1.1000001p12,
+ -0x1.1000001p12,
+ 0x1.cp13,
+ -0x1.cp13,
+ 0x1.100001p11,
+ -0x1.100001p11,
+};
+
+const char* testFunc[] =
+{
+ " short convert_short(double x)",
+ " short convert_short_sat(double x)",
+ " short convert_short_rtz(double x)",
+ " short convert_short_rtn(double x)",
+ " short convert_short_rte(double x)",
+ " short convert_short_rtp(double x)",
+ " short convert_short_sat_rtz(double x)",
+ " short convert_short_sat_rtn(double x)",
+ " short convert_short_sat_rte(double x)",
+ " short convert_short_sat_rtp(double x)",
+
+ " ushort convert_ushort(double x)",
+ " ushort convert_ushort_sat(double x)",
+ " ushort convert_ushort_rtz(double x)",
+ " ushort convert_ushort_rtn(double x)",
+ " ushort convert_ushort_rte(double x)",
+ " ushort convert_ushort_rtp(double x)",
+ " ushort convert_ushort_sat_rtz(double x)",
+ " ushort convert_ushort_sat_rtn(double x)",
+ " ushort convert_ushort_sat_rte(double x)",
+ " ushort convert_ushort_sat_rtp(double x)"
+};
+
+short expectResultshort[] = {
+ 1, -4, 287, -287, 128, -128, 127, -127, 255, 0, -255, 256, -256, 32, -32, 0,
+ 0, 32767, -32768, 32767, -32768, 511, -511, 32767, -32768, 4352, -4352, 14336, -14336, 2176, -2176, 0,
+ 1, -4, 287, -287, 128, -128, 127, -127, 255, 0, -255, 256, -256, 32, -32, 0,
+ 0, 32767, -32768, 32767, -32768, 511, -511, 32767, -32768, 4352, -4352, 14336, -14336, 2176, -2176, 0,
+ 1, -4, 287, -287, 128, -128, 127, -127, 255, 0, -255, 256, -256, 32, -32, 0,
+ 0, 32767, -32768, 32767, -32768, 511, -511, 32767, -32768, 4352, -4352, 14336, -14336, 2176, -2176, 0,
+ 1, -5, 287, -288, 128, -128, 127, -128, 255, 0, -256, 256, -257, 32, -33, 0,
+ -1, 32767, -32768, 32767, -32768, 511, -512, 32767, -32768, 4352, -4353, 14336, -14336, 2176, -2177, 0,
+ 1, -4, 288, -288, 128, -128, 128, -128, 256, 0, -256, 256, -256, 32, -32, 0,
+ 0, 32767, -32768, 32767, -32768, 512, -512, 32767, -32768, 4352, -4352, 14336, -14336, 2176, -2176, 0,
+ 2, -4, 288, -287, 128, -128, 128, -127, 256, 1, -255, 257, -256, 33, -32, 1,
+ 0, 32767, -32768, 32767, -32768, 512, -511, 32767, -32768, 4353, -4352, 14336, -14336, 2177, -2176, 0,
+ 1, -4, 287, -287, 128, -128, 127, -127, 255, 0, -255, 256, -256, 32, -32, 0,
+ 0, 32767, -32768, 32767, -32768, 511, -511, 32767, -32768, 4352, -4352, 14336, -14336, 2176, -2176, 0,
+ 1, -5, 287, -288, 128, -128, 127, -128, 255, 0, -256, 256, -257, 32, -33, 0,
+ -1, 32767, -32768, 32767, -32768, 511, -512, 32767, -32768, 4352, -4353, 14336, -14336, 2176, -2177, 0,
+ 1, -4, 288, -288, 128, -128, 128, -128, 256, 0, -256, 256, -256, 32, -32, 0,
+ 0, 32767, -32768, 32767, -32768, 512, -512, 32767, -32768, 4352, -4352, 14336, -14336, 2176, -2176, 0,
+ 2, -4, 288, -287, 128, -128, 128, -127, 256, 1, -255, 257, -256, 33, -32, 1,
+ 0, 32767, -32768, 32767, -32768, 512, -511, 32767, -32768, 4353, -4352, 14336, -14336, 2177, -2176, 0,
+
+};
+
+unsigned short expectResultUshort[] = {
+ 1, 0, 287, 0, 128, 0, 127, 0, 255, 0, 0, 256, 0, 32, 0, 0,
+ 0, 65535, 0, 65535, 0, 511, 0, 65535, 0, 4352, 0, 14336, 0, 2176, 0, 0,
+ 1, 0, 287, 0, 128, 0, 127, 0, 255, 0, 0, 256, 0, 32, 0, 0,
+ 0, 65535, 0, 65535, 0, 511, 0, 65535, 0, 4352, 0, 14336, 0, 2176, 0, 0,
+ 1, 0, 287, 0, 128, 0, 127, 0, 255, 0, 0, 256, 0, 32, 0, 0,
+ 0, 65535, 0, 65535, 0, 511, 0, 65535, 0, 4352, 0, 14336, 0, 2176, 0, 0,
+ 1, 0, 287, 0, 128, 0, 127, 0, 255, 0, 0, 256, 0, 32, 0, 0,
+ 0, 65535, 0, 65535, 0, 511, 0, 65535, 0, 4352, 0, 14336, 0, 2176, 0, 0,
+ 1, 0, 288, 0, 128, 0, 128, 0, 256, 0, 0, 256, 0, 32, 0, 0,
+ 0, 65535, 0, 65535, 0, 512, 0, 65535, 0, 4352, 0, 14336, 0, 2176, 0, 0,
+ 2, 0, 288, 0, 128, 0, 128, 0, 256, 1, 0, 257, 0, 33, 0, 1,
+ 0, 65535, 0, 65535, 0, 512, 0, 65535, 0, 4353, 0, 14336, 0, 2177, 0, 0,
+ 1, 0, 287, 0, 128, 0, 127, 0, 255, 0, 0, 256, 0, 32, 0, 0,
+ 0, 65535, 0, 65535, 0, 511, 0, 65535, 0, 4352, 0, 14336, 0, 2176, 0, 0,
+ 1, 0, 287, 0, 128, 0, 127, 0, 255, 0, 0, 256, 0, 32, 0, 0,
+ 0, 65535, 0, 65535, 0, 511, 0, 65535, 0, 4352, 0, 14336, 0, 2176, 0, 0,
+ 1, 0, 288, 0, 128, 0, 128, 0, 256, 0, 0, 256, 0, 32, 0, 0,
+ 0, 65535, 0, 65535, 0, 512, 0, 65535, 0, 4352, 0, 14336, 0, 2176, 0, 0,
+ 2, 0, 288, 0, 128, 0, 128, 0, 256, 1, 0, 257, 0, 33, 0, 1,
+ 0, 65535, 0, 65535, 0, 512, 0, 65535, 0, 4353, 0, 14336, 0, 2177, 0, 0
+};
+
+double *input_data;
+const int count_input = 32;
+const int max_function = 20;
+
+static void builtin_convert_double2int16(void)
+{
+ // Setup kernel and buffers
+ int k, i, index_cur;
+ short gpu_data[max_function * count_input] = {0};
+ float diff;
+ short log[256] = {0};
+
+ OCL_CREATE_KERNEL("builtin_convert_double2int16");
+
+ OCL_CREATE_BUFFER(buf[0], CL_MEM_READ_WRITE, count_input * max_function * sizeof(double), NULL);
+ OCL_CREATE_BUFFER(buf[1], CL_MEM_READ_WRITE, count_input * max_function * sizeof(short), NULL);
+ OCL_CREATE_BUFFER(buf[2], CL_MEM_READ_WRITE, count_input * max_function * sizeof(short), NULL);
+
+ OCL_SET_ARG(0, sizeof(cl_mem), &buf[0]);
+ OCL_SET_ARG(1, sizeof(cl_mem), &buf[1]);
+ OCL_SET_ARG(2, sizeof(cl_mem), &buf[2]);
+ OCL_SET_ARG(3, sizeof(int), &count_input);
+
+ globals[0] = 1;
+ locals[0] = 1;
+
+ input_data = (double *)doubleX;
+ clEnqueueWriteBuffer( queue, buf[0], CL_TRUE, 0, count_input * max_function * sizeof(double), input_data, 0, NULL, NULL);
+
+ // Run the kernel
+ OCL_NDRANGE( 1 );
+
+ clEnqueueReadBuffer( queue, buf[1], CL_TRUE, 0, sizeof(short) * max_function/2 * count_input, gpu_data, 0, NULL, NULL);
+ for (k = 0; (uint)k < count_input*max_function/2; k++)
+ {
+ OCL_ASSERT(gpu_data[k] == expectResultshort[k]);
+ if(gpu_data[k] != expectResultshort[k])
+ {
+ printf("failed at function:%s, expect value: %d, but get :%d \n", testFunc[k/count_input], expectResultshort[k], gpu_data[k]);
+ }
+ }
+
+ clEnqueueReadBuffer( queue, buf[2], CL_TRUE, 0, sizeof(short) * max_function/2 * count_input, gpu_data, 0, NULL, NULL);
+ unsigned short *ugpu_data = (unsigned short *)gpu_data;
+ for (k = 0; (uint)k < count_input*max_function/2; k++)
+ {
+ OCL_ASSERT(ugpu_data[k] == expectResultUshort[k]);
+ if(ugpu_data[k] != expectResultUshort[k])
+ {
+ printf("failed at function:%s, expect value: %d, but get :%d \n", testFunc[k/count_input], expectResultUshort[k], ugpu_data[k]);
+ }
+ }
+}
+
+MAKE_UTEST_FROM_FUNCTION(builtin_convert_double2int16)
+}
--
2.7.4
More information about the Beignet
mailing list