Improve msvc codegen for hot-path array LUTs
In some constexpr functions, msvc is building the LUT at runtime (pushing each element onto the stack) out of an abundance of caution. Moving the arrays into be file-scoped constexpr's avoids this and turns the functions into simple look-ups as intended.
This commit is contained in:
parent
465f486160
commit
7853e6b5d4
|
@ -125,16 +125,6 @@ enum class SurfaceTarget {
|
|||
TextureCubeArray,
|
||||
};
|
||||
|
||||
/**
|
||||
* Gets the compression factor for the specified PixelFormat. This applies to just the
|
||||
* "compressed width" and "compressed height", not the overall compression factor of a
|
||||
* compressed image. This is used for maintaining proper surface sizes for compressed
|
||||
* texture formats.
|
||||
*/
|
||||
static constexpr u32 GetCompressionFactor(PixelFormat format) {
|
||||
if (format == PixelFormat::Invalid)
|
||||
return 0;
|
||||
|
||||
constexpr std::array<u32, MaxPixelFormat> compression_factor_table = {{
|
||||
1, // ABGR8U
|
||||
1, // ABGR8S
|
||||
|
@ -204,13 +194,20 @@ static constexpr u32 GetCompressionFactor(PixelFormat format) {
|
|||
1, // Z32FS8
|
||||
}};
|
||||
|
||||
/**
|
||||
* Gets the compression factor for the specified PixelFormat. This applies to just the
|
||||
* "compressed width" and "compressed height", not the overall compression factor of a
|
||||
* compressed image. This is used for maintaining proper surface sizes for compressed
|
||||
* texture formats.
|
||||
*/
|
||||
static constexpr u32 GetCompressionFactor(PixelFormat format) {
|
||||
if (format == PixelFormat::Invalid)
|
||||
return 0;
|
||||
|
||||
ASSERT(static_cast<std::size_t>(format) < compression_factor_table.size());
|
||||
return compression_factor_table[static_cast<std::size_t>(format)];
|
||||
}
|
||||
|
||||
static constexpr u32 GetDefaultBlockWidth(PixelFormat format) {
|
||||
if (format == PixelFormat::Invalid)
|
||||
return 0;
|
||||
constexpr std::array<u32, MaxPixelFormat> block_width_table = {{
|
||||
1, // ABGR8U
|
||||
1, // ABGR8S
|
||||
|
@ -279,14 +276,15 @@ static constexpr u32 GetDefaultBlockWidth(PixelFormat format) {
|
|||
1, // S8Z24
|
||||
1, // Z32FS8
|
||||
}};
|
||||
|
||||
static constexpr u32 GetDefaultBlockWidth(PixelFormat format) {
|
||||
if (format == PixelFormat::Invalid)
|
||||
return 0;
|
||||
|
||||
ASSERT(static_cast<std::size_t>(format) < block_width_table.size());
|
||||
return block_width_table[static_cast<std::size_t>(format)];
|
||||
}
|
||||
|
||||
static constexpr u32 GetDefaultBlockHeight(PixelFormat format) {
|
||||
if (format == PixelFormat::Invalid)
|
||||
return 0;
|
||||
|
||||
constexpr std::array<u32, MaxPixelFormat> block_height_table = {{
|
||||
1, // ABGR8U
|
||||
1, // ABGR8S
|
||||
|
@ -356,14 +354,14 @@ static constexpr u32 GetDefaultBlockHeight(PixelFormat format) {
|
|||
1, // Z32FS8
|
||||
}};
|
||||
|
||||
static constexpr u32 GetDefaultBlockHeight(PixelFormat format) {
|
||||
if (format == PixelFormat::Invalid)
|
||||
return 0;
|
||||
|
||||
ASSERT(static_cast<std::size_t>(format) < block_height_table.size());
|
||||
return block_height_table[static_cast<std::size_t>(format)];
|
||||
}
|
||||
|
||||
static constexpr u32 GetFormatBpp(PixelFormat format) {
|
||||
if (format == PixelFormat::Invalid)
|
||||
return 0;
|
||||
|
||||
constexpr std::array<u32, MaxPixelFormat> bpp_table = {{
|
||||
32, // ABGR8U
|
||||
32, // ABGR8S
|
||||
|
@ -433,6 +431,10 @@ static constexpr u32 GetFormatBpp(PixelFormat format) {
|
|||
64, // Z32FS8
|
||||
}};
|
||||
|
||||
static constexpr u32 GetFormatBpp(PixelFormat format) {
|
||||
if (format == PixelFormat::Invalid)
|
||||
return 0;
|
||||
|
||||
ASSERT(static_cast<std::size_t>(format) < bpp_table.size());
|
||||
return bpp_table[static_cast<std::size_t>(format)];
|
||||
}
|
||||
|
|
Reference in New Issue