struct translate *translate_create( const struct translate_key *key )
{
struct translate *translate = NULL;
#if defined(PIPE_ARCH_X86)
translate = translate_sse2_create( key );
if (translate)
return translate;
#else
(void)translate;
#endif
return translate_generic_create( key );
}
static int update_zero_stride( struct svga_context *svga,
unsigned dirty )
{
unsigned i;
svga->curr.zero_stride_vertex_elements = 0;
svga->curr.num_zero_stride_vertex_elements = 0;
for (i = 0; i < svga->curr.num_vertex_elements; i++) {
const struct pipe_vertex_element *vel = &svga->curr.ve[i];
const struct pipe_vertex_buffer *vbuffer = &svga->curr.vb[
vel->vertex_buffer_index];
if (vbuffer->stride == 0) {
unsigned const_idx =
svga->curr.num_zero_stride_vertex_elements;
struct translate *translate;
struct translate_key key;
void *mapped_buffer;
svga->curr.zero_stride_vertex_elements |= (1 << i);
++svga->curr.num_zero_stride_vertex_elements;
key.output_stride = 4 * sizeof(float);
key.nr_elements = 1;
key.element[0].type = TRANSLATE_ELEMENT_NORMAL;
key.element[0].input_format = vel->src_format;
key.element[0].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
key.element[0].input_buffer = vel->vertex_buffer_index;
key.element[0].input_offset = vel->src_offset;
key.element[0].instance_divisor = vel->instance_divisor;
key.element[0].output_offset = const_idx * 4 * sizeof(float);
translate_key_sanitize(&key);
/* translate_generic_create is technically private but
* we don't want to code-generate, just want generic
* translation */
translate = translate_generic_create(&key);
assert(vel->src_offset == 0);
mapped_buffer = pipe_buffer_map_range(svga->pipe.screen,
vbuffer->buffer,
vel->src_offset,
util_format_get_blocksize(vel->src_format),
PIPE_BUFFER_USAGE_CPU_READ);
translate->set_buffer(translate, vel->vertex_buffer_index,
mapped_buffer,
vbuffer->stride);
translate->run(translate, 0, 1, 0,
svga->curr.zero_stride_constants);
pipe_buffer_unmap(svga->pipe.screen,
vbuffer->buffer);
translate->release(translate);
}
}
if (svga->curr.num_zero_stride_vertex_elements)
svga->dirty |= SVGA_NEW_ZERO_STRIDE;
return 0;
}
int main(int argc, char** argv)
{
struct translate *(*create_fn)(const struct translate_key *key) = 0;
struct translate_key key;
unsigned output_format;
unsigned input_format;
unsigned buffer_size = 4096;
unsigned char* buffer[5];
unsigned char* byte_buffer;
float* float_buffer;
double* double_buffer;
uint16_t *half_buffer;
unsigned * elts;
unsigned count = 4;
unsigned i, j, k;
unsigned passed = 0;
unsigned total = 0;
const float error = 0.03125;
create_fn = 0;
util_cpu_detect();
if(argc <= 1)
{}
else if (!strcmp(argv[1], "generic"))
create_fn = translate_generic_create;
else if (!strcmp(argv[1], "x86"))
create_fn = translate_sse2_create;
else if (!strcmp(argv[1], "nosse"))
{
util_cpu_caps.has_sse = 0;
util_cpu_caps.has_sse2 = 0;
util_cpu_caps.has_sse3 = 0;
util_cpu_caps.has_sse4_1 = 0;
create_fn = translate_sse2_create;
}
else if (!strcmp(argv[1], "sse"))
{
if(!util_cpu_caps.has_sse || !rtasm_cpu_has_sse())
{
printf("Error: CPU doesn't support SSE (test with qemu)\n");
return 2;
}
util_cpu_caps.has_sse2 = 0;
util_cpu_caps.has_sse3 = 0;
util_cpu_caps.has_sse4_1 = 0;
create_fn = translate_sse2_create;
}
else if (!strcmp(argv[1], "sse2"))
{
if(!util_cpu_caps.has_sse2 || !rtasm_cpu_has_sse())
{
printf("Error: CPU doesn't support SSE2 (test with qemu)\n");
return 2;
}
util_cpu_caps.has_sse3 = 0;
util_cpu_caps.has_sse4_1 = 0;
create_fn = translate_sse2_create;
}
else if (!strcmp(argv[1], "sse3"))
{
if(!util_cpu_caps.has_sse3 || !rtasm_cpu_has_sse())
{
printf("Error: CPU doesn't support SSE3 (test with qemu)\n");
return 2;
}
util_cpu_caps.has_sse4_1 = 0;
create_fn = translate_sse2_create;
}
else if (!strcmp(argv[1], "sse4.1"))
{
if(!util_cpu_caps.has_sse4_1 || !rtasm_cpu_has_sse())
{
printf("Error: CPU doesn't support SSE4.1 (test with qemu)\n");
return 2;
}
create_fn = translate_sse2_create;
}
if (!create_fn)
{
printf("Usage: ./translate_test [generic|x86|nosse|sse|sse2|sse3|sse4.1]\n");
return 2;
}
for (i = 1; i < Elements(buffer); ++i)
buffer[i] = align_malloc(buffer_size, 4096);
byte_buffer = align_malloc(buffer_size, 4096);
float_buffer = align_malloc(buffer_size, 4096);
double_buffer = align_malloc(buffer_size, 4096);
half_buffer = align_malloc(buffer_size, 4096);
elts = align_malloc(count * sizeof *elts, 4096);
key.nr_elements = 1;
key.element[0].input_buffer = 0;
key.element[0].input_offset = 0;
key.element[0].output_offset = 0;
key.element[0].type = TRANSLATE_ELEMENT_NORMAL;
key.element[0].instance_divisor = 0;
srand(4359025);
/* avoid negative values that work badly when converted to unsigned format*/
for (i = 0; i < buffer_size; ++i)
byte_buffer[i] = rand() & 0x7f7f7f7f;
for (i = 0; i < buffer_size / sizeof(float); ++i)
float_buffer[i] = (float)rand_double();
for (i = 0; i < buffer_size / sizeof(double); ++i)
double_buffer[i] = rand_double();
for (i = 0; i < buffer_size / sizeof(double); ++i)
half_buffer[i] = util_float_to_half((float) rand_double());
for (i = 0; i < count; ++i)
elts[i] = i;
for (output_format = 1; output_format < PIPE_FORMAT_COUNT; ++output_format)
{
const struct util_format_description* output_format_desc = util_format_description(output_format);
unsigned output_format_size;
unsigned output_normalized = 0;
if (!output_format_desc
|| !output_format_desc->fetch_rgba_float
|| !output_format_desc->pack_rgba_float
|| output_format_desc->colorspace != UTIL_FORMAT_COLORSPACE_RGB
|| output_format_desc->layout != UTIL_FORMAT_LAYOUT_PLAIN
|| !translate_is_output_format_supported(output_format))
continue;
for(i = 0; i < output_format_desc->nr_channels; ++i)
{
if(output_format_desc->channel[i].type != UTIL_FORMAT_TYPE_FLOAT)
output_normalized |= (1 << output_format_desc->channel[i].normalized);
}
output_format_size = util_format_get_stride(output_format, 1);
for (input_format = 1; input_format < PIPE_FORMAT_COUNT; ++input_format)
{
const struct util_format_description* input_format_desc = util_format_description(input_format);
unsigned input_format_size;
struct translate* translate[2];
unsigned fail = 0;
unsigned used_generic = 0;
unsigned input_normalized = 0;
boolean input_is_float = FALSE;
if (!input_format_desc
|| !input_format_desc->fetch_rgba_float
|| !input_format_desc->pack_rgba_float
|| input_format_desc->colorspace != UTIL_FORMAT_COLORSPACE_RGB
|| input_format_desc->layout != UTIL_FORMAT_LAYOUT_PLAIN
|| !translate_is_output_format_supported(input_format))
continue;
input_format_size = util_format_get_stride(input_format, 1);
for(i = 0; i < input_format_desc->nr_channels; ++i)
{
if(input_format_desc->channel[i].type == UTIL_FORMAT_TYPE_FLOAT)
{
input_is_float = 1;
input_normalized |= 1 << 1;
}
else
input_normalized |= (1 << input_format_desc->channel[i].normalized);
}
if(((input_normalized | output_normalized) == 3)
|| ((input_normalized & 1) && (output_normalized & 1)
&& input_format_size * output_format_desc->nr_channels > output_format_size * input_format_desc->nr_channels))
continue;
key.element[0].input_format = input_format;
key.element[0].output_format = output_format;
key.output_stride = output_format_size;
translate[0] = create_fn(&key);
if (!translate[0])
continue;
key.element[0].input_format = output_format;
key.element[0].output_format = input_format;
key.output_stride = input_format_size;
translate[1] = create_fn(&key);
if(!translate[1])
{
used_generic = 1;
translate[1] = translate_generic_create(&key);
if(!translate[1])
continue;
}
for(i = 1; i < 5; ++i)
memset(buffer[i], 0xcd - (0x22 * i), 4096);
if(input_is_float && input_format_desc->channel[0].size == 32)
buffer[0] = (unsigned char*)float_buffer;
else if(input_is_float && input_format_desc->channel[0].size == 64)
buffer[0] = (unsigned char*)double_buffer;
else if(input_is_float && input_format_desc->channel[0].size == 16)
buffer[0] = (unsigned char*)half_buffer;
else if(input_is_float)
abort();
else
buffer[0] = byte_buffer;
translate[0]->set_buffer(translate[0], 0, buffer[0], input_format_size, count - 1);
translate[0]->run_elts(translate[0], elts, count, 0, 0, buffer[1]);
translate[1]->set_buffer(translate[1], 0, buffer[1], output_format_size, count - 1);
translate[1]->run_elts(translate[1], elts, count, 0, 0, buffer[2]);
translate[0]->set_buffer(translate[0], 0, buffer[2], input_format_size, count - 1);
translate[0]->run_elts(translate[0], elts, count, 0, 0, buffer[3]);
translate[1]->set_buffer(translate[1], 0, buffer[3], output_format_size, count - 1);
translate[1]->run_elts(translate[1], elts, count, 0, 0, buffer[4]);
for (i = 0; i < count; ++i)
{
float a[4];
float b[4];
input_format_desc->fetch_rgba_float(a, buffer[2] + i * input_format_size, 0, 0);
input_format_desc->fetch_rgba_float(b, buffer[4] + i * input_format_size, 0, 0);
for (j = 0; j < count; ++j)
{
float d = a[j] - b[j];
if (d > error || d < -error)
{
fail = 1;
break;
}
}
}
printf("%s%s: %s -> %s -> %s -> %s -> %s\n",
fail ? "FAIL" : "PASS",
used_generic ? "[GENERIC]" : "",
input_format_desc->name, output_format_desc->name, input_format_desc->name, output_format_desc->name, input_format_desc->name);
if (1)
{
for (i = 0; i < Elements(buffer); ++i)
{
unsigned format_size = (i & 1) ? output_format_size : input_format_size;
printf("%c ", (i == 2 || i == 4) ? '*' : ' ');
for (j = 0; j < count; ++j)
{
for (k = 0; k < format_size; ++k)
{
printf("%02x", buffer[i][j * format_size + k]);
}
printf(" ");
}
printf("\n");
}
}
if (!fail)
++passed;
++total;
if(translate[1])
translate[1]->release(translate[1]);
translate[0]->release(translate[0]);
}
}
printf("%u/%u tests passed for translate_%s\n", passed, total, argv[1]);
return passed != total;
}
