spirv: Work around OpLogicalNot being unsupported for ints

glsl: Move misc face/position variables to after main, fixes ES3 compatibility

glsl: Fix support for PSIZE1 and up

spirv: Fix PointCoord input using vec4 instead of vec2

Replace SWAPDBL loop with memcpy, fixes MinGW Win64

Fix caller-defined swizzles for Shader Model 1 shaders.

vulkan: Decouple block size from block increment

D3D11: Increase bool uniform buffer size (thanks @jeffphilp!)

Missed an increment...

D3D11: bool constants are stored as int4 (thanks @jeffphilp!)

opengl: Don't assert if vs_flip_loc == -1. This can be -1 if the GLSL compiler removed it, or if you bound a program without a vertex shader at all.

glsl: better fix for incorrect SM1.1 defines.

Make sure attribute registers are declared correctly in weird SM1 case.

glsl: for SM1.1, don't add incorrect gl_TexCoord defines. We handled this elsewhere.

utils: Added testglcompile.c, to see if the GL accepts our generated shaders.

assembler: Don't lose the TEXBEM opcode in the special ps_1_1 TEX detection.

d3d11: Fix possible leaks after compiling shaders

vulkan: Add support for multithreaded UBOs

spirv: Separate pixel shader outputs from inputs, fix linker accordingly

vulkan: Fix overlapping pixel/vertex shaders where one shader is NULL

Write locations for SPIR-V pixel shader outputs

Handle MOJOSHADER_USAGE_UNKNOWN in SPIR-V linker

Fix SM1 shaders on SPIR-V after TEX* opcodes changes: - Fix implicit input attribute on SPIR-V profile - Remove texcoord attribs from other TEX* opcodes

assembler: set the default source arg swizzle to the correct value.

metal: Fix RCP and RSQ output to match GLSL.

d3d11: Only build the D3D11 glue code on Windows.

cmake: build the HLSL profile by default.

Rebuilt the lexer with a much newer re2c.

glsl: Correct RCP and RSQ output.

glsl: Fix compiler warning.

RSQ opcode requires replicate swizzle.

The assembler needs to accept an implicit ".x" swizzle on RCP's source arg.

Minor replicate_swizzle optimization. Do it in two tests instead of three, and with two shifts instead of five.

Clean up assembler writemask parsing a little.

List default writemasks for instructions. This lets this line of assembly generate correct code: m3x3 r7, r4, c28 As this is legal, but the bytecode validator will throw it out for having a writemask of .xyzw on r7 when it requires .xyz. The assembler needs to know what the implicit writemask should be in this circumstance. (Some instructions might still be incorrect, but are left at the default of a full .xyzw writemask, so they continue to act as before until corrected.)

Fixed typo in error message.

Assembler now accepts "c[5]" as equivalent to "c5". This only accepts a constant integers, you can't currently do something like "c[2+3]" even if the constants could be completely folded during assembly. I don't know if Microsoft's tools allow that, will have to revisit later if so.

Assembler now deals with scalar registers specifying a write mask better.

Don't make the MOJOSHADER_DO_INSTRUCTION_TABLE caller undef everything after.

For SM1.1 TEX, don't add texcoord attributes. Other parts of the system will generate the tX registers as global variables, assign gl_TexCoord[X] to them, and overwrite them with the sampled pixels during the TEX instruction. Setting these as texcoord attributes causes it to _also_ generate a generic vertex attribute and cause all sorts of problems.

Correct the reported usage on SM1 vertex shader input registers. They are hardcoded to mean specific things (v0 is POSITION0, v1 is BLENDWEIGHT, etc).

VS2010 buildfix

Metal uniform buffer overhaul. 1. Metal now uses a single, ~1MB uniform buffer per frame that is shared among all shaders. From my testing this should be more than enough room for the FNA catalog (aside from possibly Little Racers Street with its memory leak). This is a major boon for Metal memory management since it only requires one buffer instead of 3,000 (in LRS's case)! And it simplifies the internal workings of mojoshader_metal dramatically. 2. Metal now has a MOJOSHADER_mtlMakeContextCurrent API to match GL and Vulkan. Other parts of the context API have been rewritten as well to match the behavior you'd expect. 3. The MOJOSHADER_mtlGetUniformBuffers function has been replaced with MOJOSHADER_mtlGetUniformData. Now that we only use a single uniform buffer, there's no need to return one for each of the shaders. 4. The MOJOSHADER_mtlGetFunctionHandle function no longer allocates memory. Instead, all MTLFunction handles are cached in the MOJOSHADER_mtlShader structs at library creation time. This removes the burden of memory management from the user. 5. A variety of formatting and documentation updates.

spirv: Fix bogus detection of PSIZE0

Add dynamic linking support for SPIR-V modules

vulkan: Rework UBO allocation to use a single monolithic buffer

vulkan: Fix uniform buffer copies for bools

spirv: Fix support for FOG1

spirv: Fix texcoord use before load

Fix include path for vulkan.h

Add support for Vulkan rendering. Co-authors include: - Caleb Cornett <caleb.cornett@outlook.com> - Martin Krošlák <kroslakma@gmail.com> - Ethan Lee <flibitijibibo@flibitijibibo.com>

Add getError to effectShaderContext

Improve error handling in Effect implementation

spirv: Fix asserts for pixel shader POSITION/PSIZE lengths

spirv: Rework attrib offset/length organization

Note the usage index range for MOJOSHADER_usage enum

spirv: Fix support for POSITION1+

spirv: Have separate VPOS paths for GL/VK

spirv: Allow COLOR usages to go up to 4, not just 2

Disable D3D11 HLSL optimizations, as they break program behavior