[Mesa-dev] [PATCH 8/9] llvmpipe: Test more vector lengths.

[Jose Fonseca]

On 18/04/16 16:30, Tom Stellard wrote:
> On Mon, Apr 18, 2016 at 10:14:35AM +0100, Jose Fonseca wrote:
>> All power of two of up native vector length.
>>
>> There is actually a bug in lp_build_round for v2, whereby it doesn't
>> round to nearest.  Fixing is left to the future, but the test is now
>> able to expect it to fail.
>> ---
>>   src/gallium/drivers/llvmpipe/lp_test_arit.c | 43 ++++++++++++++++++++---------
>>   1 file changed, 30 insertions(+), 13 deletions(-)
>>
>> diff --git a/src/gallium/drivers/llvmpipe/lp_test_arit.c b/src/gallium/drivers/llvmpipe/lp_test_arit.c
>> index ba831f3..f3ba5a1 100644
>> --- a/src/gallium/drivers/llvmpipe/lp_test_arit.c
>> +++ b/src/gallium/drivers/llvmpipe/lp_test_arit.c
>> @@ -297,14 +297,16 @@ unary_tests[] = {
>>    */
>>   static LLVMValueRef
>>   build_unary_test_func(struct gallivm_state *gallivm,
>> -                      const struct unary_test_t *test)
>> +                      const struct unary_test_t *test,
>> +                      unsigned length,
>> +                      const char *test_name)
>>   {
>> -   struct lp_type type = lp_type_float_vec(32, lp_native_vector_width);
>> +   struct lp_type type = lp_type_float_vec(32, length * 32);
>>      LLVMContextRef context = gallivm->context;
>>      LLVMModuleRef module = gallivm->module;
>>      LLVMTypeRef vf32t = lp_build_vec_type(gallivm, type);
>>      LLVMTypeRef args[2] = { LLVMPointerType(vf32t, 0), LLVMPointerType(vf32t, 0) };
>> -   LLVMValueRef func = LLVMAddFunction(module, test->name,
>> +   LLVMValueRef func = LLVMAddFunction(module, test_name,
>>                                          LLVMFunctionType(LLVMVoidTypeInContext(context),
>>                                                           args, Elements(args), 0));
>>      LLVMValueRef arg0 = LLVMGetParam(func, 0);
>> @@ -371,14 +373,15 @@ flush_denorm_to_zero(float val)
>>    * Test one LLVM unary arithmetic builder function.
>>    */
>>   static boolean
>> -test_unary(unsigned verbose, FILE *fp, const struct unary_test_t *test)
>> +test_unary(unsigned verbose, FILE *fp, const struct unary_test_t *test, unsigned length)
>>   {
>> +   char test_name[128];
>> +   util_snprintf(test_name, sizeof test_name, "%s.v%u", test->name, length);
>>      struct gallivm_state *gallivm;
>>      LLVMValueRef test_func;
>>      unary_func_t test_func_jit;
>>      boolean success = TRUE;
>>      int i, j;
>> -   int length = lp_native_vector_width / 32;
>>      float *in, *out;
>>
>>      in = align_malloc(length * 4, length * 4);
>> @@ -391,7 +394,7 @@ test_unary(unsigned verbose, FILE *fp, const struct unary_test_t *test)
>>
>>      gallivm = gallivm_create("test_module", LLVMGetGlobalContext());
>
> This is not related to this patch, but the c++ equivalent of
> LLVMGetGlobalContext() has been removed from LLVM, and I think the C
> API may be removed at some point in the future, so these tests should
> be migrated to use LLVMCreateContext().
>
> -Tom

Thanks for the heads up.  I went ahead and fixed this in the tests.

There's still a use in llvmpipe, but it's guarded by 
PIPE_SUBSYSTEM_EMBEDDED, so it shouldn't affect anybody.  We'll deal 
with that when LLVM breaks the C ABI  -- it's just a matter of creating 
the context in llvmpipe_screen


Jose