static void
vc5_set_blend_color(struct pipe_context *pctx,
const struct pipe_blend_color *blend_color)
{
struct vc5_context *vc5 = vc5_context(pctx);
vc5->blend_color.f = *blend_color;
for (int i = 0; i < 4; i++) {
vc5->blend_color.hf[i] =
util_float_to_half(blend_color->color[i]);
}
vc5->dirty |= VC5_DIRTY_BLEND_COLOR;
}
/* helper for setting up border-color buffer for a3xx/a4xx: */
void
fd_setup_border_colors(struct fd_texture_stateobj *tex, void *ptr,
unsigned offset)
{
unsigned i, j;
for (i = 0; i < tex->num_samplers; i++) {
struct pipe_sampler_state *sampler = tex->samplers[i];
uint16_t *bcolor = (uint16_t *)((uint8_t *)ptr +
(BORDERCOLOR_SIZE * offset) +
(BORDERCOLOR_SIZE * i));
uint32_t *bcolor32 = (uint32_t *)&bcolor[16];
if (!sampler)
continue;
/*
* XXX HACK ALERT XXX
*
* The border colors need to be swizzled in a particular
* format-dependent order. Even though samplers don't know about
* formats, we can assume that with a GL state tracker, there's a
* 1:1 correspondence between sampler and texture. Take advantage
* of that knowledge.
*/
if (i < tex->num_textures && tex->textures[i]) {
const struct util_format_description *desc =
util_format_description(tex->textures[i]->format);
for (j = 0; j < 4; j++) {
if (desc->swizzle[j] >= 4)
continue;
const struct util_format_channel_description *chan =
&desc->channel[desc->swizzle[j]];
if (chan->pure_integer) {
bcolor32[desc->swizzle[j] + 4] = sampler->border_color.i[j];
bcolor[desc->swizzle[j] + 8] = sampler->border_color.i[j];
} else {
bcolor32[desc->swizzle[j]] = fui(sampler->border_color.f[j]);
bcolor[desc->swizzle[j]] =
util_float_to_half(sampler->border_color.f[j]);
}
}
}
}
}
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;
}
bool radv_format_pack_clear_color(VkFormat format,
uint32_t clear_vals[2],
VkClearColorValue *value)
{
uint8_t r = 0, g = 0, b = 0, a = 0;
const struct vk_format_description *desc = vk_format_description(format);
if (vk_format_get_component_bits(format, VK_FORMAT_COLORSPACE_RGB, 0) <= 8) {
if (desc->colorspace == VK_FORMAT_COLORSPACE_RGB) {
r = float_to_ubyte(value->float32[0]);
g = float_to_ubyte(value->float32[1]);
b = float_to_ubyte(value->float32[2]);
a = float_to_ubyte(value->float32[3]);
} else if (desc->colorspace == VK_FORMAT_COLORSPACE_SRGB) {
r = util_format_linear_float_to_srgb_8unorm(value->float32[0]);
g = util_format_linear_float_to_srgb_8unorm(value->float32[1]);
b = util_format_linear_float_to_srgb_8unorm(value->float32[2]);
a = float_to_ubyte(value->float32[3]);
}
}
switch (format) {
case VK_FORMAT_R8_UNORM:
case VK_FORMAT_R8_SRGB:
clear_vals[0] = r;
clear_vals[1] = 0;
break;
case VK_FORMAT_R8G8_UNORM:
case VK_FORMAT_R8G8_SRGB:
clear_vals[0] = r | g << 8;
clear_vals[1] = 0;
break;
case VK_FORMAT_R8G8B8A8_SRGB:
case VK_FORMAT_R8G8B8A8_UNORM:
clear_vals[0] = r | g << 8 | b << 16 | a << 24;
clear_vals[1] = 0;
break;
case VK_FORMAT_B8G8R8A8_SRGB:
case VK_FORMAT_B8G8R8A8_UNORM:
clear_vals[0] = b | g << 8 | r << 16 | a << 24;
clear_vals[1] = 0;
break;
case VK_FORMAT_A8B8G8R8_UNORM_PACK32:
case VK_FORMAT_A8B8G8R8_SRGB_PACK32:
clear_vals[0] = r | g << 8 | b << 16 | a << 24;
clear_vals[1] = 0;
break;
case VK_FORMAT_R8_UINT:
clear_vals[0] = value->uint32[0] & 0xff;
clear_vals[1] = 0;
break;
case VK_FORMAT_R16_UINT:
clear_vals[0] = value->uint32[0] & 0xffff;
clear_vals[1] = 0;
break;
case VK_FORMAT_R8G8_UINT:
clear_vals[0] = value->uint32[0] & 0xff;
clear_vals[0] |= (value->uint32[1] & 0xff) << 8;
clear_vals[1] = 0;
break;
case VK_FORMAT_R8G8B8A8_UINT:
clear_vals[0] = value->uint32[0] & 0xff;
clear_vals[0] |= (value->uint32[1] & 0xff) << 8;
clear_vals[0] |= (value->uint32[2] & 0xff) << 16;
clear_vals[0] |= (value->uint32[3] & 0xff) << 24;
clear_vals[1] = 0;
break;
case VK_FORMAT_A8B8G8R8_UINT_PACK32:
clear_vals[0] = value->uint32[0] & 0xff;
clear_vals[0] |= (value->uint32[1] & 0xff) << 8;
clear_vals[0] |= (value->uint32[2] & 0xff) << 16;
clear_vals[0] |= (value->uint32[3] & 0xff) << 24;
clear_vals[1] = 0;
break;
case VK_FORMAT_R16G16_UINT:
clear_vals[0] = value->uint32[0] & 0xffff;
clear_vals[0] |= (value->uint32[1] & 0xffff) << 16;
clear_vals[1] = 0;
break;
case VK_FORMAT_R16G16B16A16_UINT:
clear_vals[0] = value->uint32[0] & 0xffff;
clear_vals[0] |= (value->uint32[1] & 0xffff) << 16;
clear_vals[1] = value->uint32[2] & 0xffff;
clear_vals[1] |= (value->uint32[3] & 0xffff) << 16;
break;
case VK_FORMAT_R32_UINT:
clear_vals[0] = value->uint32[0];
clear_vals[1] = 0;
break;
case VK_FORMAT_R32G32_UINT:
clear_vals[0] = value->uint32[0];
clear_vals[1] = value->uint32[1];
break;
case VK_FORMAT_R32_SINT:
clear_vals[0] = value->int32[0];
clear_vals[1] = 0;
break;
case VK_FORMAT_R16_SFLOAT:
clear_vals[0] = util_float_to_half(value->float32[0]);
clear_vals[1] = 0;
break;
case VK_FORMAT_R16G16_SFLOAT:
clear_vals[0] = util_float_to_half(value->float32[0]);
clear_vals[0] |= (uint32_t)util_float_to_half(value->float32[1]) << 16;
clear_vals[1] = 0;
break;
case VK_FORMAT_R16G16B16A16_SFLOAT:
clear_vals[0] = util_float_to_half(value->float32[0]);
clear_vals[0] |= (uint32_t)util_float_to_half(value->float32[1]) << 16;
clear_vals[1] = util_float_to_half(value->float32[2]);
clear_vals[1] |= (uint32_t)util_float_to_half(value->float32[3]) << 16;
break;
case VK_FORMAT_R16_UNORM:
clear_vals[0] = ((uint16_t)util_iround(CLAMP(value->float32[0], 0.0f, 1.0f) * 0xffff)) & 0xffff;
clear_vals[1] = 0;
break;
case VK_FORMAT_R16G16_UNORM:
clear_vals[0] = ((uint16_t)util_iround(CLAMP(value->float32[0], 0.0f, 1.0f) * 0xffff)) & 0xffff;
clear_vals[0] |= ((uint16_t)util_iround(CLAMP(value->float32[1], 0.0f, 1.0f) * 0xffff)) << 16;
clear_vals[1] = 0;
break;
case VK_FORMAT_R16G16B16A16_UNORM:
clear_vals[0] = ((uint16_t)util_iround(CLAMP(value->float32[0], 0.0f, 1.0f) * 0xffff)) & 0xffff;
clear_vals[0] |= ((uint16_t)util_iround(CLAMP(value->float32[1], 0.0f, 1.0f) * 0xffff)) << 16;
clear_vals[1] = ((uint16_t)util_iround(CLAMP(value->float32[2], 0.0f, 1.0f) * 0xffff)) & 0xffff;
clear_vals[1] |= ((uint16_t)util_iround(CLAMP(value->float32[3], 0.0f, 1.0f) * 0xffff)) << 16;
break;
case VK_FORMAT_A2B10G10R10_UNORM_PACK32:
/* TODO */
return false;
case VK_FORMAT_R32G32_SFLOAT:
clear_vals[0] = fui(value->float32[0]);
clear_vals[1] = fui(value->float32[1]);
break;
case VK_FORMAT_R32_SFLOAT:
clear_vals[1] = 0;
clear_vals[0] = fui(value->float32[0]);
break;
default:
fprintf(stderr, "failed to fast clear %d\n", format);
return false;
}
return true;
}
