[Piglit] [PATCH] Add vector conversion tests.

Mon Nov 26 13:24:23 PST 2012

On Mon, Nov 26, 2012 at 07:41:52AM -0600, Aaron Watry wrote:
> Adds tests for scalar to vector conversions as well as float -> int,
> char -> int, and int -> char.
> 
> Also adds tests for IEEE rounding modes (user-specified and default).

This looks good to me.

Reviewed-by: Tom Stellard <thomas.stellard at amd.com>

> ---
>  tests/all_cl.tests                            |    1 +
>  tests/cl/program/execute/vector-conversion.cl |  147 +++++++++++++++++++++++++
>  2 files changed, 148 insertions(+)
>  create mode 100644 tests/cl/program/execute/vector-conversion.cl
> 
> diff --git a/tests/all_cl.tests b/tests/all_cl.tests
> index fab760f..14a9bf8 100644
> --- a/tests/all_cl.tests
> +++ b/tests/all_cl.tests
> @@ -121,6 +121,7 @@ add_plain_program_tester_test(program_execute, 'Scalar arithmetic ulong', 'execu
>  add_plain_program_tester_test(program_execute, 'Scalar arithmetic ushort', 'execute/scalar-arithmetic-ushort.cl')
>  add_plain_program_tester_test(program_execute, 'Vector arithmetic int4', 'execute/vector-arithmetic-int4.program_test')
>  add_plain_program_tester_test(program_execute, 'Vector arithmetic float4', 'execute/vector-arithmetic-float4.program_test')
> +add_plain_program_tester_test(program_execute, 'Vector conversions', 'execute/vector-conversion.cl')
>  add_plain_program_tester_test(program_execute, 'Vector load int4', 'execute/vector-load-int4.cl')
>  add_plain_program_tester_test(program_execute, 'Vector store int4', 'execute/vector-store-int4.cl')
>  add_plain_program_tester_test(program_execute, 'Scalar bitwise op int', 'execute/scalar-bitwise-int.cl')
> diff --git a/tests/cl/program/execute/vector-conversion.cl b/tests/cl/program/execute/vector-conversion.cl
> new file mode 100644
> index 0000000..b54685f
> --- /dev/null
> +++ b/tests/cl/program/execute/vector-conversion.cl
> @@ -0,0 +1,147 @@
> +/*!
> +[config]
> +name: Vector Conversion tests
> +clc_version_min: 10
> +
> +dimensions: 1
> +global_size: 1 0 0
> +
> +[test]
> +name: Convert scalar int to vector int4 via cast
> +kernel_name: convert_int_to_int4
> +arg_in:  1 int            5
> +arg_out: 0 buffer int4[1]  5 5 5 5
> +
> +[test]
> +name: Convert char4 to int4
> +kernel_name: convert_char4_int4
> +arg_in:  1 char4 0 1 2 3
> +arg_out: 0 buffer int4[1]  0 1 2 3
> +
> +[test]
> +name: Convert int4 to float4
> +kernel_name: convert_int4_float4
> +arg_in:  1 int4 0 1 2 3
> +arg_out: 0 buffer float4[1]  0 1 2 3
> +
> +[test]
> +name: Convert int4 to saturated char4
> +kernel_name: convert_int4_to_char4_sat
> +arg_in:  1 int4 -100000 1 2 123456
> +arg_out: 0 buffer char4[1]  -128 1 2 127
> +
> +[test]
> +name: Convert float8 to int8, round to zero (trunc)
> +kernel_name: convert_float8_to_int8_rtz
> +arg_in:  1 float8 -23.67 -23.50 -23.35 -23.0 23.0 23.35 23.50 23.67
> +arg_out: 0 buffer int8[1] -23 -23 -23 -23 23 23 23 23
> +
> +[test]
> +name: Convert float8 to int8, round to positive infinity (ceil)
> +kernel_name: convert_float8_to_int8_rtp
> +arg_in:  1 float8 -23.67 -23.50 -23.35 -23.0 23.0 23.35 23.50 23.67
> +arg_out: 0 buffer int8[1] -23 -23 -23 -23 23 24 24 24
> +
> +[test]
> +name: Convert float8 to int8, round to negative infinity (floor)
> +kernel_name: convert_float8_to_int8_rtn
> +arg_in:  1 float8 -23.67 -23.50 -23.35 -23.0 23.0 23.35 23.50 23.67
> +arg_out: 0 buffer int8[1] -24 -24 -24 -23 23 23 23 23
> +
> +[test]
> +name: Convert float8 to int8, round (half) to nearest even
> +kernel_name: convert_float8_to_int8_rte
> +arg_in:  1 float8 -23.67 -23.50 -23.35 -23.0 23.0 23.35 23.50 23.67
> +arg_out: 0 buffer int8[1]  -24 -24 -23 -23 23 23 24 24
> +
> +[test]
> +name: Convert float8 to int8, round (default rounding mode, rtz)
> +kernel_name: convert_float8_to_int8
> +arg_in:  1 float8 -23.67 -23.50 -23.35 -23.0 23.0 23.35 23.50 23.67
> +arg_out: 0 buffer int8[1] -23 -23 -23 -23 23 23 23 23
> +
> +[test]
> +name: Convert float4 to int4, round to zero
> +kernel_name: convert_float4_to_int4_rtz
> +arg_in:  1 float4 -123.1 0.9 1.3 123.456
> +arg_out: 0 buffer int4[1]  -123 0 1 123
> +
> +[test]
> +name: Convert float4 to int4, round to positive infinity
> +kernel_name: convert_float4_to_int4_rtp
> +arg_in:  1 float4 -123.1 0.9 1.3 123.456
> +arg_out: 0 buffer int4[1]  -123 1 2 124
> +
> +[test]
> +name: Convert float4 to int4, round to negative infinity
> +kernel_name: convert_float4_to_int4_rtn
> +arg_in:  1 float4 -123.1 0.9 1.3 123.456
> +arg_out: 0 buffer int4[1]  -124 0 1 123
> +
> +[test]
> +name: Convert int4 to uint4 via cast, no rounding, no saturation
> +kernel_name: int4_as_uint4
> +arg_in:  1 int4 -123456 1 2 123456
> +arg_out: 0 buffer uint4[1]  4294843840 1 2 123456
> +
> +!*/
> +
> +kernel void convert_float8_to_int8_rtz(global int8* out, float8 in){
> +    *out = convert_int8_rtz(in);
> +}
> +
> +kernel void convert_float8_to_int8_rtp(global int8* out, float8 in){
> +    *out = convert_int8_rtp(in);
> +}
> +
> +kernel void convert_float8_to_int8_rtn(global int8* out, float8 in){
> +    *out = convert_int8_rtn(in);
> +}
> +
> +kernel void convert_float8_to_int8_rte(global int8* out, float8 in){
> +    *out = convert_int8_rte(in);
> +}
> +
> +/* 
> +    According to the following, default for float -> int is truncate:
> +    http://www.khronos.org/registry/cl/sdk/1.0/docs/man/xhtml/SELECT_ROUNDING_MODE.html 
> +*/
> +kernel void convert_float8_to_int8(global int8* out, float8 in){
> +    *out = convert_int8(in);
> +}
> +
> +kernel void int4_as_uint4(global uint4* out, int4 in){
> +    *out = as_uint4(in);
> +}
> +
> +kernel void convert_float4_to_int4_rtz(global int4* out, float4 in){
> +    *out = convert_int4_rtz(in);
> +}
> +
> +kernel void convert_float4_to_int4_rtp(global int4* out, float4 in){
> +    *out = convert_int4_rtp(in);
> +}
> +
> +kernel void convert_float4_to_int4_rtn(global int4* out, float4 in){
> +    *out = convert_int4_rtn(in);
> +}
> +
> +kernel void convert_int4_to_char4_sat_rte(global char4* out, int4 in){
> +    *out = convert_char4_sat_rte(in);
> +}
> +
> +kernel void convert_int4_to_char4_sat(global char4* out, int4 in){
> +    *out = convert_char4_sat(in);
> +}
> +
> +kernel void convert_int_to_int4(global int4* out, int in){
> +    *out = (int4) in;
> +}
> +
> +kernel void convert_char4_int4(global int4* out, char4 in){
> +    *out = convert_int4(in);
> +}
> +
> +kernel void convert_int4_float4(global float4* out, int4 in){
> +    *out = convert_float4(in);
> +}
> -- 
> 1.7.10.4
> 
> _______________________________________________
> Piglit mailing list
> Piglit at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/piglit