shifathosn
/
KytyPS5
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

graphics: decode packed unsigned float CMASK register clears 

Co-authored-by: Ali Almohaya <AliAlmohaya@gmail.com>
This commit is contained in:
nmzik 2026-10-13 04:32:37 +02:00
parent 5f6dc0a8d0
commit ff6dffe298
2 changed files with 27 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
src/graphics/host_gpu/renderer/image/imageInfo.h
View File

@ -435,6 +435,11 @@ inline constexpr std::array<VideoOutFormatPolicy, 8> VIDEO_OUT_FORMAT_POLICIES {
(mantissa << 13u));
};
switch (format) {
case vk::Format::eB10G11R11UfloatPack32:
next.float32[0] = float16((packed & 0x7ffu) << 4u);
next.float32[1] = float16(((packed >> 11u) & 0x7ffu) << 4u);
next.float32[2] = float16((packed >> 22u) << 5u);
break;
case vk::Format::eR16G16B16A16Sfloat:
next.float32[2] = float16(static_cast<uint32_t>(clear_words >> 32u));
next.float32[3] = float16(static_cast<uint32_t>(clear_words >> 48u));

26
tests/ShaderRecompilerComputeTests.cpp
View File

@ -10869,6 +10869,13 @@ public:
std::array<float, 4> channels;
};
const std::array packed_cases{
PackedCase{vk::Format::eB10G11R11UfloatPack32, 0u, {}},
PackedCase{vk::Format::eB10G11R11UfloatPack32, 0x801c03c0u, {1.0f, 0.5f, 2.0f}},
PackedCase{vk::Format::eB10G11R11UfloatPack32, 1u | (0x7bfu << 11u) | (1u << 22u),
{0x1p-20f, 65024.0f, 0x1p-19f}},
PackedCase{vk::Format::eB10G11R11UfloatPack32, 0x7c0u | (0x7c0u << 11u) | (0x3e0u << 22u),
{std::numeric_limits<float>::infinity(), std::numeric_limits<float>::infinity(),
std::numeric_limits<float>::infinity()}},
PackedCase{vk::Format::eR16Sfloat, 0x7bff7bffu, {65504.0f}},
PackedCase{vk::Format::eR16G16Sfloat, 0xbc003c00u, {1.0f, -1.0f}},
PackedCase{vk::Format::eR16G16B16A16Sfloat, 0x40003800bc003c00ull,
@ -10888,17 +10895,23 @@ public:
};
for (const auto &test : packed_cases) {
vk::ClearColorValue clear{};
Require(name, "packed 16-bit clear",
Require(name, "packed color clear",
DecodePackedColorClear(test.format, test.word, clear) &&
std::bit_cast<std::array<uint32_t, 4>>(clear.float32) ==
std::bit_cast<std::array<uint32_t, 4>>(test.channels),
"16-bit clear lost channel order, sign, range or unused-channel zeros");
"packed clear lost channel order, sign, range or unused-channel zeros");
}
vk::ClearColorValue clear{};
Require(name, "half NaN clear",
DecodePackedColorClear(vk::Format::eR16Sfloat, 0x7e00u, clear) &&
std::isnan(clear.float32[0]),
"half NaN did not remain NaN");
Require(name, "packed unsigned NaN clear",
DecodePackedColorClear(vk::Format::eB10G11R11UfloatPack32,
0x7c1u | (0x7c1u << 11u) | (0x3e1u << 22u), clear) &&
std::isnan(clear.float32[0]) && std::isnan(clear.float32[1]) &&
std::isnan(clear.float32[2]),
"packed unsigned NaNs did not remain NaN");
for (const auto format : {vk::Format::eR16Sfloat, vk::Format::eR16Unorm}) {
Require(name, "R16 DWORD fill",
!DecodeColorDwordFill(format, 0xbc003c00u, clear) &&
@ -10930,6 +10943,10 @@ public:
.eliminate_fast_clear = true},
FillCase{.fill = 0, .texel = {0, 0x3c000000u}, .cmask = true,
.eliminate_fast_clear = true},
FillCase{.fill = 0, .texel = {0, 0}, .cmask = true,
.layout = Prospero::ChannelLayout::k11_11_10, .eliminate_fast_clear = true},
FillCase{.fill = 0, .texel = {0x801c03c0u, 0}, .cmask = true,
.layout = Prospero::ChannelLayout::k11_11_10, .eliminate_fast_clear = true},
FillCase{0x40404040u, {0, 0x3c000000u}, false, false, true},
FillCase{.fill = 0x20202020u, .texel = {0x7bff7bffu},
.layout = Prospero::ChannelLayout::k16, .clear_word = 0x7bff7bffu},
@ -11033,7 +11050,7 @@ public:
registers.SetColorDccAddr(0, {.addr = dcc_address});
registers.SetColorCmask(0, {.addr = dcc_address});
registers.SetColorClearWord0(0, {.word0 = fill_case.cmask ? expected[0] : fill_case.clear_word});
registers.SetColorClearWord1(0, {.word1 = fill_case.eliminate_fast_clear ? expected[1] : 0});
registers.SetColorClearWord1(0, {.word1 = fill_case.cmask && expected.size() > 1u ? expected[1] : 0});
registers.SetRenderTargetMask(0x0f);
scheduler.Begin(registers, user_config, shaders);
@ -11141,7 +11158,8 @@ public:
color.image_id == preexisting_id &&
texture_cache.FindTexture(color.image_id, sampled) != nullptr,
"FCE failed to retain the sampled allocation");
const std::vector<u32> stale{0x00003c00u, 0x3c003c00u};
const std::vector<u32> stale = layout == Prospero::ChannelLayout::k11_11_10
? std::vector<u32>{0x780003c0u} : std::vector<u32>{0x00003c00u, 0x3c003c00u};
Require(name, "FCE before graphics shader or target binding",
ReadCachedTexel(name, context, color.image_id, {}, {1, 1, 1}, 1) == expected &&
ReadCachedTexel(name, context, color.image_id, {511, 255, 0},