radv: Do not hardcode fast clear formats.
authorBas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Mon, 4 Jun 2018 22:52:34 +0000 (00:52 +0200)
committerBas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Tue, 5 Jun 2018 18:53:21 +0000 (20:53 +0200)
except for the odd one out.

This should support many more formats.

Reviewed-by: Dave Airlie <airlied@redhat.com>
src/amd/vulkan/radv_formats.c

index aafe331f27d603ed3366c78d8754325c50fa4cb9..50ec904d514b7f0b8295212733f23470a02c58e1 100644 (file)
@@ -880,194 +880,87 @@ 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_R8_SINT:
-               clear_vals[0] = value->int32[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_R8G8_SINT:
-               clear_vals[0] = value->int32[0] & 0xff;
-               clear_vals[0] |= (value->int32[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_R8G8B8A8_SINT:
-               clear_vals[0] = value->int32[0] & 0xff;
-               clear_vals[0] |= (value->int32[1] & 0xff) << 8;
-               clear_vals[0] |= (value->int32[2] & 0xff) << 16;
-               clear_vals[0] |= (value->int32[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_R16G16B16A16_SNORM:
-               clear_vals[0] = ((uint16_t)util_iround(CLAMP(value->float32[0], -1.0f, 1.0f) * 0x7fff)) & 0xffff;
-               clear_vals[0] |= ((uint16_t)util_iround(CLAMP(value->float32[1], -1.0f, 1.0f) * 0x7fff)) << 16;
-               clear_vals[1] = ((uint16_t)util_iround(CLAMP(value->float32[2], -1.0f, 1.0f) * 0x7fff)) & 0xffff;
-               clear_vals[1] |= ((uint16_t)util_iround(CLAMP(value->float32[3], -1.0f, 1.0f) * 0x7fff)) << 16;
-               break;
-       case VK_FORMAT_A2B10G10R10_UNORM_PACK32:
-               clear_vals[0] = ((uint16_t)util_iround(CLAMP(value->float32[0], 0.0f, 1.0f) * 0x3ff)) & 0x3ff;
-               clear_vals[0] |= (((uint16_t)util_iround(CLAMP(value->float32[1], 0.0f, 1.0f) * 0x3ff)) & 0x3ff) << 10;
-               clear_vals[0] |= (((uint16_t)util_iround(CLAMP(value->float32[2], 0.0f, 1.0f) * 0x3ff)) & 0x3ff) << 20;
-               clear_vals[0] |= (((uint16_t)util_iround(CLAMP(value->float32[3], 0.0f, 1.0f) * 0x3)) & 0x3) << 30;
-               clear_vals[1] = 0;
-               return true;
-       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;
-       case VK_FORMAT_B10G11R11_UFLOAT_PACK32:
+       if (format == VK_FORMAT_B10G11R11_UFLOAT_PACK32) {
                clear_vals[0] = float3_to_r11g11b10f(value->float32);
                clear_vals[1] = 0;
-               break;
-       case VK_FORMAT_R32G32B32A32_SFLOAT:
-               if (value->float32[0] != value->float32[1] ||
-                   value->float32[0] != value->float32[2])
-                       return false;
-               clear_vals[0] = fui(value->float32[0]);
-               clear_vals[1] = fui(value->float32[3]);
-               break;
-       case VK_FORMAT_R32G32B32A32_UINT:
-               if (value->uint32[0] != value->uint32[1] ||
-                   value->uint32[0] != value->uint32[2])
-                       return false;
+               return true;
+       }
+
+       if (desc->layout != VK_FORMAT_LAYOUT_PLAIN) {
+               fprintf(stderr, "failed to fast clear for non-plain format %d\n", format);
+               return false;
+       }
+
+       if (!util_is_power_of_two_or_zero(desc->block.bits)) {
+               fprintf(stderr, "failed to fast clear for NPOT format %d\n", format);
+               return false;
+       }
+
+       if (desc->block.bits > 64) {
+               /*
+                * We have a 128 bits format, check if the first 3 components are the same.
+                * Every elements has to be 32 bits since we don't support 64-bit formats,
+                * and we can skip swizzling checks as alpha always comes last for these and
+                * we do not care about the rest as they have to be the same.
+                */
+               if (desc->channel[0].type == VK_FORMAT_TYPE_FLOAT) {
+                       if (value->float32[0] != value->float32[1] ||
+                           value->float32[0] != value->float32[2])
+                               return false;
+               } else {
+                       if (value->uint32[0] != value->uint32[1] ||
+                           value->uint32[0] != value->uint32[2])
+                               return false;
+               }
                clear_vals[0] = value->uint32[0];
                clear_vals[1] = value->uint32[3];
-               break;
-       case VK_FORMAT_R32G32B32A32_SINT:
-               if (value->int32[0] != value->int32[1] ||
-                   value->int32[0] != value->int32[2])
+               return true;
+       }
+       uint64_t clear_val = 0;
+
+       for (unsigned c = 0; c < 4; ++c) {
+               if (desc->swizzle[c] < 0 || desc->swizzle[c] >= 4)
+                       continue;
+
+               const struct vk_format_channel_description *channel = &desc->channel[desc->swizzle[c]];
+               assert(channel->size);
+
+               uint64_t v = 0;
+               if (channel->pure_integer) {
+                       v = value->uint32[c]  & ((1ULL << channel->size) - 1);
+               } else if (channel->normalized) {
+                       if (channel->type == VK_FORMAT_TYPE_UNSIGNED &&
+                           desc->swizzle[c] < 3 &&
+                           desc->colorspace == VK_FORMAT_COLORSPACE_SRGB) {
+                               assert(channel->size == 8);
+
+                               v = util_format_linear_float_to_srgb_8unorm(value->float32[c]);
+                       } else if (channel->type == VK_FORMAT_TYPE_UNSIGNED) {
+                               v = MAX2(MIN2(value->float32[c], 1.0f), 0.0f) * ((1ULL << channel->size) - 1);
+                       } else  {
+                               v = MAX2(MIN2(value->float32[c], 1.0f), -1.0f) * ((1ULL << (channel->size - 1)) - 1);
+                       }
+               } else if (channel->type == VK_FORMAT_TYPE_FLOAT) {
+                       if (channel->size == 32) {
+                               memcpy(&v, &value->float32[c], 4);
+                       } else if(channel->size == 16) {
+                               v = util_float_to_half(value->float32[c]);
+                       } else {
+                               fprintf(stderr, "failed to fast clear for unhandled float size in format %d\n", format);
+                               return false;
+                       }
+               } else {
+                       fprintf(stderr, "failed to fast clear for unhandled component type in format %d\n", format);
                        return false;
-               clear_vals[0] = value->int32[0];
-               clear_vals[1] = value->int32[3];
-               break;
-       default:
-               fprintf(stderr, "failed to fast clear %d\n", format);
-               return false;
+               }
+               clear_val |= (v & ((1ULL << channel->size) - 1)) << channel->shift;
        }
+
+       clear_vals[0] = clear_val;
+       clear_vals[1] = clear_val >> 32;
+
        return true;
 }