adding GLSL std 450 base
All checks were successful
Build / build (push) Successful in 1m44s
Test / build (push) Successful in 6m46s

This commit is contained in:
2026-01-24 02:46:02 +01:00
parent 37da19ed43
commit 96ad7f12f9
14 changed files with 1501 additions and 740 deletions

View File

@@ -4,8 +4,8 @@ const spv = @import("spv");
const shader_source = @embedFile("shader.spv");
const screen_width = 1250;
const screen_height = 720;
const screen_width = 200;
const screen_height = 200;
pub fn main() !void {
{
@@ -36,10 +36,16 @@ pub fn main() !void {
}
for (0..screen_height) |_| {
var rt = try spv.Runtime.init(allocator, &module);
(try runner_cache.addOne(allocator)).* = .{
.allocator = allocator,
.surface = surface,
.rt = try spv.Runtime.init(allocator, &module),
.rt = rt,
.entry = try rt.getEntryPointByName("main"),
.color = try rt.getResultByName("color"),
.time = try rt.getResultByName("time"),
.pos = try rt.getResultByName("pos"),
.res = try rt.getResultByName("res"),
};
}
@@ -48,9 +54,11 @@ pub fn main() !void {
.allocator = allocator,
});
var timer = try std.time.Timer.start();
var quit = false;
while (!quit) {
try surface.clear(.{ .r = 0.0, .g = 0.0, .b = 0.0, .a = 0.0 });
try surface.clear(.{ .r = 0.0, .g = 0.0, .b = 0.0, .a = 1.0 });
while (sdl3.events.poll()) |event|
switch (event) {
@@ -65,17 +73,19 @@ pub fn main() !void {
const pixel_map: [*]u32 = @as([*]u32, @ptrCast(@alignCast((surface.getPixels() orelse return).ptr)));
var timer = try std.time.Timer.start();
var frame_timer = try std.time.Timer.start();
defer {
const ns = timer.lap();
const ns = frame_timer.lap();
const ms = @as(f32, @floatFromInt(ns)) / std.time.ns_per_s;
std.log.info("Took {d:.3}s - {d:.3}fps to render", .{ ms, 1.0 / ms });
}
const delta: f32 = @as(f32, @floatFromInt(timer.read())) / std.time.ns_per_s;
var wait_group: std.Thread.WaitGroup = .{};
for (0..screen_height) |y| {
const runner = &runner_cache.items[y];
thread_pool.spawnWg(&wait_group, Runner.run, .{ runner, y, pixel_map });
thread_pool.spawnWg(&wait_group, Runner.runWrapper, .{ runner, y, pixel_map, delta });
}
thread_pool.waitAndWork(&wait_group);
}
@@ -92,23 +102,33 @@ const Runner = struct {
allocator: std.mem.Allocator,
surface: sdl3.surface.Surface,
rt: spv.Runtime,
entry: spv.SpvWord,
color: spv.SpvWord,
time: spv.SpvWord,
pos: spv.SpvWord,
res: spv.SpvWord,
fn run(self: *Self, y: usize, pixel_map: [*]u32) void {
fn runWrapper(self: *Self, y: usize, pixel_map: [*]u32, timer: f32) void {
@call(.always_inline, Self.run, .{ self, y, pixel_map, timer }) catch |err| {
std.log.err("{s}", .{@errorName(err)});
if (@errorReturnTrace()) |trace| {
std.debug.dumpStackTrace(trace.*);
}
std.process.abort();
};
}
fn run(self: *Self, y: usize, pixel_map: [*]u32, timer: f32) !void {
var rt = self.rt; // Copy to avoid pointer access of `self` at runtime. Okay as Runtime contains only pointers and trivially copyable fields
const entry = rt.getEntryPointByName("main") catch |err| std.debug.panic("Catch error {s}", .{@errorName(err)});
const color = rt.getResultByName("color") catch |err| std.debug.panic("Catch error {s}", .{@errorName(err)});
const time = rt.getResultByName("time") catch |err| std.debug.panic("Catch error {s}", .{@errorName(err)});
const pos = rt.getResultByName("pos") catch |err| std.debug.panic("Catch error {s}", .{@errorName(err)});
const res = rt.getResultByName("res") catch |err| std.debug.panic("Catch error {s}", .{@errorName(err)});
var output: [4]f32 = undefined;
for (0..screen_width) |x| {
rt.writeInput(f32, &.{@as(f32, @floatFromInt(std.time.milliTimestamp()))}, time) catch |err| std.debug.panic("Catch error {s}", .{@errorName(err)});
rt.writeInput(f32, &.{ @floatFromInt(screen_width), @floatFromInt(screen_height) }, res) catch |err| std.debug.panic("Catch error {s}", .{@errorName(err)});
rt.writeInput(f32, &.{ @floatFromInt(x), @floatFromInt(y) }, pos) catch |err| std.debug.panic("Catch error {s}", .{@errorName(err)});
rt.callEntryPoint(self.allocator, entry) catch |err| std.debug.panic("Catch error {s}", .{@errorName(err)});
rt.readOutput(f32, output[0..], color) catch |err| std.debug.panic("Catch error {s}", .{@errorName(err)});
try rt.writeInput(f32, &.{timer}, self.time);
try rt.writeInput(f32, &.{ @floatFromInt(screen_width), @floatFromInt(screen_height) }, self.res);
try rt.writeInput(f32, &.{ @floatFromInt(x), @floatFromInt(y) }, self.pos);
try rt.callEntryPoint(self.allocator, self.entry);
try rt.readOutput(f32, output[0..], self.color);
const rgba = self.surface.mapRgba(
@truncate(@as(u32, @intFromFloat(output[0] * 255.0))),

View File

@@ -16,12 +16,53 @@ struct FragOut
[entry(frag)]
fn main(input: FragIn) -> FragOut
{
const I: i32 = 32;
const A: f32 = 7.5;
const MA: f32 = 20.0;
const MI: f32 = 0.001;
let uv0 = input.pos / input.res * 2.0 - vec2[f32](1.0, 1.0);
let uv = vec2[f32](uv0.x * (input.res.x / input.res.y), uv0.y);
let col = vec3[f32](0.0, 0.0, 0.0);
let ro = vec3[f32](0.0, 0.0, -2.0);
let rd = vec3[f32](uv.x, uv.y, 1.0);
let dt = 0.0;
let ds = 0.0;
let dm = -1.0;
let p = ro;
let c = vec3[f32](0.0, 0.0, 0.0);
let l = vec3[f32](0.0, sin(input.time * 0.2) * 4.0, cos(input.time * 0.2) * 4.0);
for i in 0 -> I
{
p = ro + rd * dt;
ds = length(c - p) - 1.0;
dt += ds;
if (dm == -1.0 || ds < dm)
dm = ds;
if (ds <= MI)
{
let value = max(dot(normalize(c - p), normalize(p - l)) - 0.35, 0.0);
col = vec3[f32](value, value, value);
break;
}
if (ds >= MA)
{
if (dot(normalize(rd), normalize(l - ro)) <= 1.0)
{
let value = max(dot(normalize(rd), normalize(l - ro)) + 0.15, 0.05)/ 1.15 * (1.0 - dm * A);
col = vec3[f32](value, value, value);
}
break;
}
}
let output: FragOut;
output.color = vec4[f32](
input.pos.x / input.res.x,
input.pos.y / input.res.y,
1.0,
1.0
);
output.color = vec4[f32](col.x, col.y, col.z, 1.0);
return output;
}

Binary file not shown.

View File

@@ -1,11 +1,12 @@
Version 1.0
Generator: 2560130
Bound: 50
Bound: 203
Schema: 0
OpCapability Capability(Shader)
%42 = OpExtInstImport "GLSL.std.450"
OpMemoryModel AddressingModel(Logical) MemoryModel(GLSL450)
OpEntryPoint ExecutionModel(Fragment) %24 "main" %5 %11 %14 %20
OpExecutionMode %24 ExecutionMode(OriginUpperLeft)
OpEntryPoint ExecutionModel(Fragment) %43 "main" %5 %11 %14 %20
OpExecutionMode %43 ExecutionMode(OriginUpperLeft)
OpSource SourceLanguage(NZSL) 4198400
OpSourceExtension "Version: 1.1"
OpName %16 "FragIn"
@@ -18,7 +19,7 @@ Schema: 0
OpName %11 "res"
OpName %14 "pos"
OpName %20 "color"
OpName %24 "main"
OpName %43 "main"
OpDecorate %5 Decoration(Location) 0
OpDecorate %11 Decoration(Location) 1
OpDecorate %14 Decoration(Location) 2
@@ -35,51 +36,246 @@ Schema: 0
%7 = OpConstant %6 i32(0)
%8 = OpTypePointer StorageClass(Function) %3
%9 = OpTypeVector %3 2
%10 = OpTypePointer StorageClass(Input) %9
%12 = OpConstant %6 i32(1)
%13 = OpTypePointer StorageClass(Function) %9
%15 = OpConstant %6 i32(2)
%16 = OpTypeStruct %3 %9 %9
%17 = OpTypePointer StorageClass(Function) %16
%18 = OpTypeVector %3 4
%19 = OpTypePointer StorageClass(Output) %18
%21 = OpTypeStruct %18
%22 = OpTypePointer StorageClass(Function) %21
%23 = OpConstant %3 f32(1)
%47 = OpTypePointer StorageClass(Function) %18
%10 = OpTypePointer StorageClass(Input) %9
%12 = OpConstant %6 i32(1)
%13 = OpTypePointer StorageClass(Function) %9
%15 = OpConstant %6 i32(2)
%16 = OpTypeStruct %3 %9 %9
%17 = OpTypePointer StorageClass(Function) %16
%18 = OpTypeVector %3 4
%19 = OpTypePointer StorageClass(Output) %18
%21 = OpTypeStruct %18
%22 = OpConstant %3 f32(2)
%23 = OpConstant %3 f32(1)
%24 = OpConstant %3 f32(0)
%25 = OpTypeVector %3 3
%26 = OpTypePointer StorageClass(Function) %25
%27 = OpConstant %3 f32(-2)
%28 = OpConstant %3 f32(-1)
%29 = OpConstant %3 f32(0.2)
%30 = OpConstant %3 f32(4)
%31 = OpTypePointer StorageClass(Function) %6
%32 = OpConstant %6 i32(32)
%33 = OpTypeBool
%34 = OpConstant %3 f32(0.001)
%35 = OpConstant %3 f32(0.35)
%36 = OpConstant %3 f32(20)
%37 = OpConstant %3 f32(0.15)
%38 = OpConstant %3 f32(0.05)
%39 = OpConstant %3 f32(1.15)
%40 = OpConstant %3 f32(7.5)
%41 = OpTypePointer StorageClass(Function) %21
%200 = OpTypePointer StorageClass(Function) %18
%5 = OpVariable %4 StorageClass(Input)
%11 = OpVariable %10 StorageClass(Input)
%14 = OpVariable %10 StorageClass(Input)
%20 = OpVariable %19 StorageClass(Output)
%24 = OpFunction %1 FunctionControl(0) %2
%25 = OpLabel
%26 = OpVariable %22 StorageClass(Function)
%27 = OpVariable %17 StorageClass(Function)
%28 = OpAccessChain %8 %27 %7
OpCopyMemory %28 %5
%29 = OpAccessChain %13 %27 %12
OpCopyMemory %29 %11
%30 = OpAccessChain %13 %27 %15
OpCopyMemory %30 %14
%31 = OpAccessChain %13 %27 %15
%32 = OpLoad %9 %31
%33 = OpCompositeExtract %3 %32 0
%34 = OpAccessChain %13 %27 %12
%35 = OpLoad %9 %34
%36 = OpCompositeExtract %3 %35 0
%37 = OpFDiv %3 %33 %36
%38 = OpAccessChain %13 %27 %15
%39 = OpLoad %9 %38
%40 = OpCompositeExtract %3 %39 1
%41 = OpAccessChain %13 %27 %12
%42 = OpLoad %9 %41
%43 = OpCompositeExtract %3 %42 1
%44 = OpFDiv %3 %40 %43
%45 = OpCompositeConstruct %18 %37 %44 %23 %23
%46 = OpAccessChain %47 %26 %7
OpStore %46 %45
%48 = OpLoad %21 %26
%49 = OpCompositeExtract %18 %48 0
OpStore %20 %49
%11 = OpVariable %10 StorageClass(Input)
%14 = OpVariable %10 StorageClass(Input)
%20 = OpVariable %19 StorageClass(Output)
%43 = OpFunction %1 FunctionControl(0) %2
%44 = OpLabel
%45 = OpVariable %13 StorageClass(Function)
%46 = OpVariable %13 StorageClass(Function)
%47 = OpVariable %26 StorageClass(Function)
%48 = OpVariable %26 StorageClass(Function)
%49 = OpVariable %26 StorageClass(Function)
%50 = OpVariable %8 StorageClass(Function)
%51 = OpVariable %8 StorageClass(Function)
%52 = OpVariable %8 StorageClass(Function)
%53 = OpVariable %26 StorageClass(Function)
%54 = OpVariable %26 StorageClass(Function)
%55 = OpVariable %26 StorageClass(Function)
%56 = OpVariable %31 StorageClass(Function)
%57 = OpVariable %31 StorageClass(Function)
%58 = OpVariable %8 StorageClass(Function)
%59 = OpVariable %8 StorageClass(Function)
%60 = OpVariable %41 StorageClass(Function)
%61 = OpVariable %17 StorageClass(Function)
%62 = OpAccessChain %8 %61 %7
OpCopyMemory %62 %5
%63 = OpAccessChain %13 %61 %12
OpCopyMemory %63 %11
%64 = OpAccessChain %13 %61 %15
OpCopyMemory %64 %14
%65 = OpAccessChain %13 %61 %15
%66 = OpLoad %9 %65
%67 = OpAccessChain %13 %61 %12
%68 = OpLoad %9 %67
%69 = OpFDiv %9 %66 %68
%70 = OpVectorTimesScalar %9 %69 %22
%71 = OpCompositeConstruct %9 %23 %23
%72 = OpFSub %9 %70 %71
OpStore %45 %72
%73 = OpLoad %9 %45
%74 = OpCompositeExtract %3 %73 0
%75 = OpAccessChain %13 %61 %12
%76 = OpLoad %9 %75
%77 = OpCompositeExtract %3 %76 0
%78 = OpAccessChain %13 %61 %12
%79 = OpLoad %9 %78
%80 = OpCompositeExtract %3 %79 1
%81 = OpFDiv %3 %77 %80
%82 = OpFMul %3 %74 %81
%83 = OpLoad %9 %45
%84 = OpCompositeExtract %3 %83 1
%85 = OpCompositeConstruct %9 %82 %84
OpStore %46 %85
%86 = OpCompositeConstruct %25 %24 %24 %24
OpStore %47 %86
%87 = OpCompositeConstruct %25 %24 %24 %27
OpStore %48 %87
%88 = OpLoad %9 %46
%89 = OpCompositeExtract %3 %88 0
%90 = OpLoad %9 %46
%91 = OpCompositeExtract %3 %90 1
%92 = OpCompositeConstruct %25 %89 %91 %23
OpStore %49 %92
OpStore %50 %24
OpStore %51 %24
OpStore %52 %28
%93 = OpLoad %25 %48
OpStore %53 %93
%94 = OpCompositeConstruct %25 %24 %24 %24
OpStore %54 %94
%95 = OpAccessChain %8 %61 %7
%96 = OpLoad %3 %95
%97 = OpFMul %3 %96 %29
%98 = OpExtInst %3 GLSLstd450 Sin %97
%99 = OpFMul %3 %98 %30
%100 = OpAccessChain %8 %61 %7
%101 = OpLoad %3 %100
%102 = OpFMul %3 %101 %29
%103 = OpExtInst %3 GLSLstd450 Cos %102
%104 = OpFMul %3 %103 %30
%105 = OpCompositeConstruct %25 %24 %99 %104
OpStore %55 %105
OpStore %56 %7
OpStore %57 %32
OpBranch %106
%106 = OpLabel
%110 = OpLoad %6 %56
%111 = OpLoad %6 %57
%112 = OpSLessThan %33 %110 %111
OpLoopMerge %108 %109 LoopControl(0)
OpBranchConditional %112 %107 %108
%107 = OpLabel
%113 = OpLoad %25 %48
%114 = OpLoad %25 %49
%115 = OpLoad %3 %50
%116 = OpVectorTimesScalar %25 %114 %115
%117 = OpFAdd %25 %113 %116
OpStore %53 %117
%118 = OpLoad %25 %54
%119 = OpLoad %25 %53
%120 = OpFSub %25 %118 %119
%121 = OpExtInst %3 GLSLstd450 Length %120
%122 = OpFSub %3 %121 %23
OpStore %51 %122
%123 = OpLoad %3 %50
%124 = OpLoad %3 %51
%125 = OpFAdd %3 %123 %124
OpStore %50 %125
%129 = OpLoad %3 %52
%130 = OpFOrdEqual %33 %129 %28
%131 = OpLoad %3 %51
%132 = OpLoad %3 %52
%133 = OpFOrdLessThan %33 %131 %132
%134 = OpLogicalOr %33 %130 %133
OpSelectionMerge %126 SelectionControl(0)
OpBranchConditional %134 %127 %128
%127 = OpLabel
%135 = OpLoad %3 %51
OpStore %52 %135
OpBranch %126
%128 = OpLabel
OpBranch %126
%126 = OpLabel
%139 = OpLoad %3 %51
%140 = OpFOrdLessThanEqual %33 %139 %34
OpSelectionMerge %136 SelectionControl(0)
OpBranchConditional %140 %137 %138
%137 = OpLabel
%141 = OpLoad %25 %54
%142 = OpLoad %25 %53
%143 = OpFSub %25 %141 %142
%144 = OpExtInst %25 GLSLstd450 Normalize %143
%145 = OpLoad %25 %53
%146 = OpLoad %25 %55
%147 = OpFSub %25 %145 %146
%148 = OpExtInst %25 GLSLstd450 Normalize %147
%149 = OpDot %3 %144 %148
%150 = OpFSub %3 %149 %35
%151 = OpExtInst %3 GLSLstd450 FMax %150 %24
OpStore %58 %151
%152 = OpLoad %3 %58
%153 = OpLoad %3 %58
%154 = OpLoad %3 %58
%155 = OpCompositeConstruct %25 %152 %153 %154
OpStore %47 %155
OpBranch %108
%138 = OpLabel
OpBranch %136
%136 = OpLabel
%159 = OpLoad %3 %51
%160 = OpFOrdGreaterThanEqual %33 %159 %36
OpSelectionMerge %156 SelectionControl(0)
OpBranchConditional %160 %157 %158
%157 = OpLabel
%164 = OpLoad %25 %49
%165 = OpExtInst %25 GLSLstd450 Normalize %164
%166 = OpLoad %25 %55
%167 = OpLoad %25 %48
%168 = OpFSub %25 %166 %167
%169 = OpExtInst %25 GLSLstd450 Normalize %168
%170 = OpDot %3 %165 %169
%171 = OpFOrdLessThanEqual %33 %170 %23
OpSelectionMerge %161 SelectionControl(0)
OpBranchConditional %171 %162 %163
%162 = OpLabel
%172 = OpLoad %25 %49
%173 = OpExtInst %25 GLSLstd450 Normalize %172
%174 = OpLoad %25 %55
%175 = OpLoad %25 %48
%176 = OpFSub %25 %174 %175
%177 = OpExtInst %25 GLSLstd450 Normalize %176
%178 = OpDot %3 %173 %177
%179 = OpFAdd %3 %178 %37
%180 = OpExtInst %3 GLSLstd450 FMax %179 %38
%181 = OpFDiv %3 %180 %39
%182 = OpLoad %3 %52
%183 = OpFMul %3 %182 %40
%184 = OpFSub %3 %23 %183
%185 = OpFMul %3 %181 %184
OpStore %59 %185
%186 = OpLoad %3 %59
%187 = OpLoad %3 %59
%188 = OpLoad %3 %59
%189 = OpCompositeConstruct %25 %186 %187 %188
OpStore %47 %189
OpBranch %161
%163 = OpLabel
OpBranch %161
%161 = OpLabel
OpBranch %108
%158 = OpLabel
OpBranch %156
%156 = OpLabel
%190 = OpLoad %6 %56
%191 = OpIAdd %6 %190 %12
OpStore %56 %191
OpBranch %109
%109 = OpLabel
OpBranch %106
%108 = OpLabel
%192 = OpLoad %25 %47
%193 = OpCompositeExtract %3 %192 0
%194 = OpLoad %25 %47
%195 = OpCompositeExtract %3 %194 1
%196 = OpLoad %25 %47
%197 = OpCompositeExtract %3 %196 2
%198 = OpCompositeConstruct %18 %193 %195 %197 %23
%199 = OpAccessChain %200 %60 %7
OpStore %199 %198
%201 = OpLoad %21 %60
%202 = OpCompositeExtract %18 %201 0
OpStore %20 %202
OpReturn
OpFunctionEnd

View File

@@ -62,8 +62,8 @@ fn main(input: FragIn) -> FragOut
}
}
if (col == vec3[f32](0.0, 0.0, 0.0))
discard;
//if (col == vec3[f32](0.0, 0.0, 0.0))
// discard;
let output: FragOut;
output.color = vec4[f32](col.x, col.y, col.z, 1.0);

Binary file not shown.

View File

@@ -1,12 +1,12 @@
Version 1.0
Generator: 2560130
Bound: 210
Bound: 203
Schema: 0
OpCapability Capability(Shader)
%43 = OpExtInstImport "GLSL.std.450"
%42 = OpExtInstImport "GLSL.std.450"
OpMemoryModel AddressingModel(Logical) MemoryModel(GLSL450)
OpEntryPoint ExecutionModel(Fragment) %44 "main" %5 %11 %14 %20
OpExecutionMode %44 ExecutionMode(OriginUpperLeft)
OpEntryPoint ExecutionModel(Fragment) %43 "main" %5 %11 %14 %20
OpExecutionMode %43 ExecutionMode(OriginUpperLeft)
OpSource SourceLanguage(NZSL) 4198400
OpSourceExtension "Version: 1.1"
OpName %16 "FragIn"
@@ -19,7 +19,7 @@ Schema: 0
OpName %11 "res"
OpName %14 "pos"
OpName %20 "color"
OpName %44 "main"
OpName %43 "main"
OpDecorate %5 Decoration(Location) 0
OpDecorate %11 Decoration(Location) 1
OpDecorate %14 Decoration(Location) 2
@@ -64,229 +64,218 @@ Schema: 0
%38 = OpConstant %3 f32(0.05)
%39 = OpConstant %3 f32(1.15)
%40 = OpConstant %3 f32(7.5)
%41 = OpTypeVector %33 3
%42 = OpTypePointer StorageClass(Function) %21
%207 = OpTypePointer StorageClass(Function) %18
%41 = OpTypePointer StorageClass(Function) %21
%200 = OpTypePointer StorageClass(Function) %18
%5 = OpVariable %4 StorageClass(Input)
%11 = OpVariable %10 StorageClass(Input)
%14 = OpVariable %10 StorageClass(Input)
%20 = OpVariable %19 StorageClass(Output)
%44 = OpFunction %1 FunctionControl(0) %2
%45 = OpLabel
%43 = OpFunction %1 FunctionControl(0) %2
%44 = OpLabel
%45 = OpVariable %13 StorageClass(Function)
%46 = OpVariable %13 StorageClass(Function)
%47 = OpVariable %13 StorageClass(Function)
%47 = OpVariable %26 StorageClass(Function)
%48 = OpVariable %26 StorageClass(Function)
%49 = OpVariable %26 StorageClass(Function)
%50 = OpVariable %26 StorageClass(Function)
%50 = OpVariable %8 StorageClass(Function)
%51 = OpVariable %8 StorageClass(Function)
%52 = OpVariable %8 StorageClass(Function)
%53 = OpVariable %8 StorageClass(Function)
%53 = OpVariable %26 StorageClass(Function)
%54 = OpVariable %26 StorageClass(Function)
%55 = OpVariable %26 StorageClass(Function)
%56 = OpVariable %26 StorageClass(Function)
%56 = OpVariable %31 StorageClass(Function)
%57 = OpVariable %31 StorageClass(Function)
%58 = OpVariable %31 StorageClass(Function)
%58 = OpVariable %8 StorageClass(Function)
%59 = OpVariable %8 StorageClass(Function)
%60 = OpVariable %8 StorageClass(Function)
%61 = OpVariable %42 StorageClass(Function)
%62 = OpVariable %17 StorageClass(Function)
%63 = OpAccessChain %8 %62 %7
OpCopyMemory %63 %5
%64 = OpAccessChain %13 %62 %12
OpCopyMemory %64 %11
%65 = OpAccessChain %13 %62 %15
OpCopyMemory %65 %14
%66 = OpAccessChain %13 %62 %15
%67 = OpLoad %9 %66
%68 = OpAccessChain %13 %62 %12
%69 = OpLoad %9 %68
%70 = OpFDiv %9 %67 %69
%71 = OpVectorTimesScalar %9 %70 %22
%72 = OpCompositeConstruct %9 %23 %23
%73 = OpFSub %9 %71 %72
OpStore %46 %73
%74 = OpLoad %9 %46
%75 = OpCompositeExtract %3 %74 0
%76 = OpAccessChain %13 %62 %12
%77 = OpLoad %9 %76
%78 = OpCompositeExtract %3 %77 0
%79 = OpAccessChain %13 %62 %12
%80 = OpLoad %9 %79
%81 = OpCompositeExtract %3 %80 1
%82 = OpFDiv %3 %78 %81
%83 = OpFMul %3 %75 %82
%84 = OpLoad %9 %46
%85 = OpCompositeExtract %3 %84 1
%86 = OpCompositeConstruct %9 %83 %85
%60 = OpVariable %41 StorageClass(Function)
%61 = OpVariable %17 StorageClass(Function)
%62 = OpAccessChain %8 %61 %7
OpCopyMemory %62 %5
%63 = OpAccessChain %13 %61 %12
OpCopyMemory %63 %11
%64 = OpAccessChain %13 %61 %15
OpCopyMemory %64 %14
%65 = OpAccessChain %13 %61 %15
%66 = OpLoad %9 %65
%67 = OpAccessChain %13 %61 %12
%68 = OpLoad %9 %67
%69 = OpFDiv %9 %66 %68
%70 = OpVectorTimesScalar %9 %69 %22
%71 = OpCompositeConstruct %9 %23 %23
%72 = OpFSub %9 %70 %71
OpStore %45 %72
%73 = OpLoad %9 %45
%74 = OpCompositeExtract %3 %73 0
%75 = OpAccessChain %13 %61 %12
%76 = OpLoad %9 %75
%77 = OpCompositeExtract %3 %76 0
%78 = OpAccessChain %13 %61 %12
%79 = OpLoad %9 %78
%80 = OpCompositeExtract %3 %79 1
%81 = OpFDiv %3 %77 %80
%82 = OpFMul %3 %74 %81
%83 = OpLoad %9 %45
%84 = OpCompositeExtract %3 %83 1
%85 = OpCompositeConstruct %9 %82 %84
OpStore %46 %85
%86 = OpCompositeConstruct %25 %24 %24 %24
OpStore %47 %86
%87 = OpCompositeConstruct %25 %24 %24 %24
%87 = OpCompositeConstruct %25 %24 %24 %27
OpStore %48 %87
%88 = OpCompositeConstruct %25 %24 %24 %27
OpStore %49 %88
%89 = OpLoad %9 %47
%90 = OpCompositeExtract %3 %89 0
%91 = OpLoad %9 %47
%92 = OpCompositeExtract %3 %91 1
%93 = OpCompositeConstruct %25 %90 %92 %23
OpStore %50 %93
%88 = OpLoad %9 %46
%89 = OpCompositeExtract %3 %88 0
%90 = OpLoad %9 %46
%91 = OpCompositeExtract %3 %90 1
%92 = OpCompositeConstruct %25 %89 %91 %23
OpStore %49 %92
OpStore %50 %24
OpStore %51 %24
OpStore %52 %24
OpStore %53 %28
%94 = OpLoad %25 %49
OpStore %52 %28
%93 = OpLoad %25 %48
OpStore %53 %93
%94 = OpCompositeConstruct %25 %24 %24 %24
OpStore %54 %94
%95 = OpCompositeConstruct %25 %24 %24 %24
OpStore %55 %95
%96 = OpAccessChain %8 %62 %7
%97 = OpLoad %3 %96
%98 = OpFMul %3 %97 %29
%99 = OpExtInst %3 GLSLstd450 Sin %98
%100 = OpFMul %3 %99 %30
%101 = OpAccessChain %8 %62 %7
%102 = OpLoad %3 %101
%103 = OpFMul %3 %102 %29
%104 = OpExtInst %3 GLSLstd450 Cos %103
%105 = OpFMul %3 %104 %30
%106 = OpCompositeConstruct %25 %24 %100 %105
OpStore %56 %106
OpStore %57 %7
OpStore %58 %32
OpBranch %107
%107 = OpLabel
%95 = OpAccessChain %8 %61 %7
%96 = OpLoad %3 %95
%97 = OpFMul %3 %96 %29
%98 = OpExtInst %3 GLSLstd450 Sin %97
%99 = OpFMul %3 %98 %30
%100 = OpAccessChain %8 %61 %7
%101 = OpLoad %3 %100
%102 = OpFMul %3 %101 %29
%103 = OpExtInst %3 GLSLstd450 Cos %102
%104 = OpFMul %3 %103 %30
%105 = OpCompositeConstruct %25 %24 %99 %104
OpStore %55 %105
OpStore %56 %7
OpStore %57 %32
OpBranch %106
%106 = OpLabel
%110 = OpLoad %6 %56
%111 = OpLoad %6 %57
%112 = OpLoad %6 %58
%113 = OpSLessThan %33 %111 %112
OpLoopMerge %109 %110 LoopControl(0)
OpBranchConditional %113 %108 %109
%108 = OpLabel
%112 = OpSLessThan %33 %110 %111
OpLoopMerge %108 %109 LoopControl(0)
OpBranchConditional %112 %107 %108
%107 = OpLabel
%113 = OpLoad %25 %48
%114 = OpLoad %25 %49
%115 = OpLoad %25 %50
%116 = OpLoad %3 %51
%117 = OpVectorTimesScalar %25 %115 %116
%118 = OpFAdd %25 %114 %117
OpStore %54 %118
%119 = OpLoad %25 %55
%120 = OpLoad %25 %54
%121 = OpFSub %25 %119 %120
%122 = OpExtInst %3 GLSLstd450 Length %121
%123 = OpFSub %3 %122 %23
OpStore %52 %123
%115 = OpLoad %3 %50
%116 = OpVectorTimesScalar %25 %114 %115
%117 = OpFAdd %25 %113 %116
OpStore %53 %117
%118 = OpLoad %25 %54
%119 = OpLoad %25 %53
%120 = OpFSub %25 %118 %119
%121 = OpExtInst %3 GLSLstd450 Length %120
%122 = OpFSub %3 %121 %23
OpStore %51 %122
%123 = OpLoad %3 %50
%124 = OpLoad %3 %51
%125 = OpLoad %3 %52
%126 = OpFAdd %3 %124 %125
OpStore %51 %126
%130 = OpLoad %3 %53
%131 = OpFOrdEqual %33 %130 %28
%125 = OpFAdd %3 %123 %124
OpStore %50 %125
%129 = OpLoad %3 %52
%130 = OpFOrdEqual %33 %129 %28
%131 = OpLoad %3 %51
%132 = OpLoad %3 %52
%133 = OpLoad %3 %53
%134 = OpFOrdLessThan %33 %132 %133
%135 = OpLogicalOr %33 %131 %134
OpSelectionMerge %127 SelectionControl(0)
OpBranchConditional %135 %128 %129
%128 = OpLabel
%136 = OpLoad %3 %52
OpStore %53 %136
OpBranch %127
%129 = OpLabel
OpBranch %127
%133 = OpFOrdLessThan %33 %131 %132
%134 = OpLogicalOr %33 %130 %133
OpSelectionMerge %126 SelectionControl(0)
OpBranchConditional %134 %127 %128
%127 = OpLabel
%140 = OpLoad %3 %52
%141 = OpFOrdLessThanEqual %33 %140 %34
OpSelectionMerge %137 SelectionControl(0)
OpBranchConditional %141 %138 %139
%138 = OpLabel
%142 = OpLoad %25 %55
%143 = OpLoad %25 %54
%144 = OpFSub %25 %142 %143
%145 = OpExtInst %25 GLSLstd450 Normalize %144
%146 = OpLoad %25 %54
%147 = OpLoad %25 %56
%148 = OpFSub %25 %146 %147
%149 = OpExtInst %25 GLSLstd450 Normalize %148
%150 = OpDot %3 %145 %149
%151 = OpFSub %3 %150 %35
%152 = OpExtInst %3 GLSLstd450 FMax %151 %24
OpStore %59 %152
%153 = OpLoad %3 %59
%154 = OpLoad %3 %59
%155 = OpLoad %3 %59
%156 = OpCompositeConstruct %25 %153 %154 %155
OpStore %48 %156
OpBranch %109
%139 = OpLabel
OpBranch %137
%135 = OpLoad %3 %51
OpStore %52 %135
OpBranch %126
%128 = OpLabel
OpBranch %126
%126 = OpLabel
%139 = OpLoad %3 %51
%140 = OpFOrdLessThanEqual %33 %139 %34
OpSelectionMerge %136 SelectionControl(0)
OpBranchConditional %140 %137 %138
%137 = OpLabel
%160 = OpLoad %3 %52
%161 = OpFOrdGreaterThanEqual %33 %160 %36
OpSelectionMerge %157 SelectionControl(0)
OpBranchConditional %161 %158 %159
%158 = OpLabel
%165 = OpLoad %25 %50
%166 = OpExtInst %25 GLSLstd450 Normalize %165
%167 = OpLoad %25 %56
%168 = OpLoad %25 %49
%169 = OpFSub %25 %167 %168
%170 = OpExtInst %25 GLSLstd450 Normalize %169
%171 = OpDot %3 %166 %170
%172 = OpFOrdLessThanEqual %33 %171 %23
OpSelectionMerge %162 SelectionControl(0)
OpBranchConditional %172 %163 %164
%163 = OpLabel
%173 = OpLoad %25 %50
%174 = OpExtInst %25 GLSLstd450 Normalize %173
%175 = OpLoad %25 %56
%176 = OpLoad %25 %49
%177 = OpFSub %25 %175 %176
%178 = OpExtInst %25 GLSLstd450 Normalize %177
%179 = OpDot %3 %174 %178
%180 = OpFAdd %3 %179 %37
%181 = OpExtInst %3 GLSLstd450 FMax %180 %38
%182 = OpFDiv %3 %181 %39
%183 = OpLoad %3 %53
%184 = OpFMul %3 %183 %40
%185 = OpFSub %3 %23 %184
%186 = OpFMul %3 %182 %185
OpStore %60 %186
%187 = OpLoad %3 %60
%188 = OpLoad %3 %60
%189 = OpLoad %3 %60
%190 = OpCompositeConstruct %25 %187 %188 %189
OpStore %48 %190
OpBranch %162
%164 = OpLabel
OpBranch %162
%162 = OpLabel
OpBranch %109
%159 = OpLabel
OpBranch %157
%141 = OpLoad %25 %54
%142 = OpLoad %25 %53
%143 = OpFSub %25 %141 %142
%144 = OpExtInst %25 GLSLstd450 Normalize %143
%145 = OpLoad %25 %53
%146 = OpLoad %25 %55
%147 = OpFSub %25 %145 %146
%148 = OpExtInst %25 GLSLstd450 Normalize %147
%149 = OpDot %3 %144 %148
%150 = OpFSub %3 %149 %35
%151 = OpExtInst %3 GLSLstd450 FMax %150 %24
OpStore %58 %151
%152 = OpLoad %3 %58
%153 = OpLoad %3 %58
%154 = OpLoad %3 %58
%155 = OpCompositeConstruct %25 %152 %153 %154
OpStore %47 %155
OpBranch %108
%138 = OpLabel
OpBranch %136
%136 = OpLabel
%159 = OpLoad %3 %51
%160 = OpFOrdGreaterThanEqual %33 %159 %36
OpSelectionMerge %156 SelectionControl(0)
OpBranchConditional %160 %157 %158
%157 = OpLabel
%191 = OpLoad %6 %57
%192 = OpIAdd %6 %191 %12
OpStore %57 %192
OpBranch %110
%110 = OpLabel
OpBranch %107
%164 = OpLoad %25 %49
%165 = OpExtInst %25 GLSLstd450 Normalize %164
%166 = OpLoad %25 %55
%167 = OpLoad %25 %48
%168 = OpFSub %25 %166 %167
%169 = OpExtInst %25 GLSLstd450 Normalize %168
%170 = OpDot %3 %165 %169
%171 = OpFOrdLessThanEqual %33 %170 %23
OpSelectionMerge %161 SelectionControl(0)
OpBranchConditional %171 %162 %163
%162 = OpLabel
%172 = OpLoad %25 %49
%173 = OpExtInst %25 GLSLstd450 Normalize %172
%174 = OpLoad %25 %55
%175 = OpLoad %25 %48
%176 = OpFSub %25 %174 %175
%177 = OpExtInst %25 GLSLstd450 Normalize %176
%178 = OpDot %3 %173 %177
%179 = OpFAdd %3 %178 %37
%180 = OpExtInst %3 GLSLstd450 FMax %179 %38
%181 = OpFDiv %3 %180 %39
%182 = OpLoad %3 %52
%183 = OpFMul %3 %182 %40
%184 = OpFSub %3 %23 %183
%185 = OpFMul %3 %181 %184
OpStore %59 %185
%186 = OpLoad %3 %59
%187 = OpLoad %3 %59
%188 = OpLoad %3 %59
%189 = OpCompositeConstruct %25 %186 %187 %188
OpStore %47 %189
OpBranch %161
%163 = OpLabel
OpBranch %161
%161 = OpLabel
OpBranch %108
%158 = OpLabel
OpBranch %156
%156 = OpLabel
%190 = OpLoad %6 %56
%191 = OpIAdd %6 %190 %12
OpStore %56 %191
OpBranch %109
%109 = OpLabel
%196 = OpLoad %25 %48
%197 = OpCompositeConstruct %25 %24 %24 %24
%198 = OpFOrdEqual %41 %196 %197
OpSelectionMerge %193 SelectionControl(0)
OpBranchConditional %198 %194 %195
%194 = OpLabel
OpKill
%195 = OpLabel
OpBranch %193
%193 = OpLabel
%199 = OpLoad %25 %48
%200 = OpCompositeExtract %3 %199 0
%201 = OpLoad %25 %48
%202 = OpCompositeExtract %3 %201 1
%203 = OpLoad %25 %48
%204 = OpCompositeExtract %3 %203 2
%205 = OpCompositeConstruct %18 %200 %202 %204 %23
%206 = OpAccessChain %207 %61 %7
OpStore %206 %205
%208 = OpLoad %21 %61
%209 = OpCompositeExtract %18 %208 0
OpStore %20 %209
OpBranch %106
%108 = OpLabel
%192 = OpLoad %25 %47
%193 = OpCompositeExtract %3 %192 0
%194 = OpLoad %25 %47
%195 = OpCompositeExtract %3 %194 1
%196 = OpLoad %25 %47
%197 = OpCompositeExtract %3 %196 2
%198 = OpCompositeConstruct %18 %193 %195 %197 %23
%199 = OpAccessChain %200 %60 %7
OpStore %199 %198
%201 = OpLoad %21 %60
%202 = OpCompositeExtract %18 %201 0
OpStore %20 %202
OpReturn
OpFunctionEnd

91
src/GLSL_std_450/GLSL_std_450.zig git.filemode.normal_file
View File

@@ -0,0 +1,91 @@
//! A jam file of translated GLSL std450 header's enums and utils
pub const GLSLstd450Version: u32 = 100;
pub const GLSLstd450Revision: u32 = 3;
pub const GLSLOp = enum(u32) {
Bad = 0,
Round = 1,
RoundEven = 2,
Trunc = 3,
FAbs = 4,
SAbs = 5,
FSign = 6,
SSign = 7,
Floor = 8,
Ceil = 9,
Fract = 10,
Radians = 11,
Degrees = 12,
Sin = 13,
Cos = 14,
Tan = 15,
Asin = 16,
Acos = 17,
Atan = 18,
Sinh = 19,
Cosh = 20,
Tanh = 21,
Asinh = 22,
Acosh = 23,
Atanh = 24,
Atan2 = 25,
Pow = 26,
Exp = 27,
Log = 28,
Exp2 = 29,
Log2 = 30,
Sqrt = 31,
InverseSqrt = 32,
Determinant = 33,
MatrixInverse = 34,
Modf = 35,
ModfStruct = 36,
FMin = 37,
UMin = 38,
SMin = 39,
FMax = 40,
UMax = 41,
SMax = 42,
FClamp = 43,
UClamp = 44,
SClamp = 45,
FMix = 46,
IMix = 47,
Step = 48,
SmoothStep = 49,
Fma = 50,
Frexp = 51,
FrexpStruct = 52,
Ldexp = 53,
PackSnorm4x8 = 54,
PackUnorm4x8 = 55,
PackSnorm2x16 = 56,
PackUnorm2x16 = 57,
PackHalf2x16 = 58,
PackDouble2x32 = 59,
UnpackSnorm2x16 = 60,
UnpackUnorm2x16 = 61,
UnpackHalf2x16 = 62,
UnpackSnorm4x8 = 63,
UnpackUnorm4x8 = 64,
UnpackDouble2x32 = 65,
Length = 66,
Distance = 67,
Cross = 68,
Normalize = 69,
FaceForward = 70,
Reflect = 71,
Refract = 72,
FindILsb = 73,
FindSMsb = 74,
FindUMsb = 75,
InterpolateAtCentroid = 76,
InterpolateAtSample = 77,
InterpolateAtOffset = 78,
NMin = 79,
NMax = 80,
NClamp = 81,
};
pub const GLSLOpMaxValue: usize = 82;

312
src/GLSL_std_450/opcodes.zig git.filemode.normal_file
View File

@@ -0,0 +1,312 @@
const std = @import("std");
const spv = @import("../spv.zig");
const ext = @import("GLSL_std_450.zig");
const opc = @import("../opcodes.zig");
const Module = @import("../Module.zig");
const Runtime = @import("../Runtime.zig");
const Result = @import("../Result.zig");
const WordIterator = @import("../WordIterator.zig");
const RuntimeError = Runtime.RuntimeError;
const ValueType = opc.ValueType;
const getValuePrimitiveField = opc.getValuePrimitiveField;
const getValuePrimitiveFieldType = opc.getValuePrimitiveFieldType;
const SpvVoid = spv.SpvVoid;
const SpvByte = spv.SpvByte;
const SpvWord = spv.SpvWord;
const SpvBool = spv.SpvBool;
const MathOp = enum {
Acos,
Acosh,
Asin,
Asinh,
Atan,
Atan2,
Atanh,
Ceil,
Cos,
Cosh,
Determinant,
Exp,
Exp2,
FAbs,
FClamp,
FMax,
FMin,
FMix,
FSign,
Floor,
Fract,
IMix,
InverseSqrt,
Log,
Log2,
Modf,
Pow,
Round,
RoundEven,
SAbs,
SClamp,
SMax,
SMin,
SSign,
Sin,
Sinh,
Sqrt,
Tan,
Tanh,
Trunc,
UClamp,
UMax,
UMin,
};
pub const OpCodeExtFunc = opc.OpCodeExtFunc;
/// Not an EnumMap as it is way too slow for this purpose
pub var runtime_dispatcher = [_]?OpCodeExtFunc{null} ** ext.GLSLOpMaxValue;
pub fn initRuntimeDispatcher() void {
// zig fmt: off
runtime_dispatcher[@intFromEnum(ext.GLSLOp.Cos)] = MathEngine(.Float, .Cos).opSingleOperator;
runtime_dispatcher[@intFromEnum(ext.GLSLOp.FMax)] = MathEngine(.Float, .FMax).opDoubleOperators;
runtime_dispatcher[@intFromEnum(ext.GLSLOp.Length)] = opLength;
runtime_dispatcher[@intFromEnum(ext.GLSLOp.Normalize)] = opNormalize;
runtime_dispatcher[@intFromEnum(ext.GLSLOp.Sin)] = MathEngine(.Float, .Sin).opSingleOperator;
// zig fmt: on
}
fn MathEngine(comptime T: ValueType, comptime Op: MathOp) type {
return struct {
fn opSingleOperator(_: std.mem.Allocator, target_type_id: SpvWord, id: SpvWord, _: SpvWord, rt: *Runtime) RuntimeError!void {
const target_type = (try rt.results[target_type_id].getVariant()).Type;
const dst = try rt.results[id].getValue();
const src = try rt.results[try rt.it.next()].getValue();
const lane_bits = try Result.resolveLaneBitWidth(target_type, rt);
const operator = struct {
fn operation(comptime TT: type, x: TT) RuntimeError!TT {
return switch (Op) {
.Sin => @sin(x),
.Cos => @cos(x),
else => RuntimeError.InvalidSpirV,
};
}
fn applyScalar(bit_count: SpvWord, d: *Result.Value, s: *const Result.Value) RuntimeError!void {
switch (bit_count) {
inline 8, 16, 32, 64 => |bits| {
if (bits == 8 and T == .Float) return RuntimeError.InvalidSpirV;
const ScalarT = getValuePrimitiveFieldType(T, bits);
const d_field = try getValuePrimitiveField(T, bits, d);
const s_field = try getValuePrimitiveField(T, bits, @constCast(s));
d_field.* = try operation(ScalarT, s_field.*);
},
else => return RuntimeError.InvalidSpirV,
}
}
};
switch (dst.*) {
.Int, .Float => try operator.applyScalar(lane_bits, dst, src),
.Vector => |dst_vec| for (dst_vec, src.Vector) |*d_lane, s_lane| {
try operator.applyScalar(lane_bits, d_lane, &s_lane);
},
.Vector4f32 => |*d| d.* = try operator.operation(@Vector(4, f32), src.Vector4f32),
.Vector3f32 => |*d| d.* = try operator.operation(@Vector(3, f32), src.Vector3f32),
.Vector2f32 => |*d| d.* = try operator.operation(@Vector(2, f32), src.Vector2f32),
//.Vector4i32 => |*d| d.* = try operator.operation(@Vector(4, i32), src.Vector4i32),
//.Vector3i32 => |*d| d.* = try operator.operation(@Vector(3, i32), src.Vector3i32),
//.Vector2i32 => |*d| d.* = try operator.operation(@Vector(2, i32), src.Vector2i32),
//.Vector4u32 => |*d| d.* = try operator.operation(@Vector(4, u32), src.Vector4u32),
//.Vector3u32 => |*d| d.* = try operator.operation(@Vector(3, u32), src.Vector3u32),
//.Vector2u32 => |*d| d.* = try operator.operation(@Vector(2, u32), src.Vector2u32),
else => return RuntimeError.InvalidSpirV,
}
}
fn opDoubleOperators(_: std.mem.Allocator, target_type_id: SpvWord, id: SpvWord, _: SpvWord, rt: *Runtime) RuntimeError!void {
const target_type = (try rt.results[target_type_id].getVariant()).Type;
const dst = try rt.results[id].getValue();
const lhs = try rt.results[try rt.it.next()].getValue();
const rhs = try rt.results[try rt.it.next()].getValue();
const lane_bits = try Result.resolveLaneBitWidth(target_type, rt);
const operator = struct {
fn operation(comptime TT: type, l: TT, r: TT) RuntimeError!TT {
return switch (Op) {
.FMax => @max(l, r),
else => RuntimeError.InvalidSpirV,
};
}
fn applyScalar(bit_count: SpvWord, d: *Result.Value, l: *const Result.Value, r: *const Result.Value) RuntimeError!void {
switch (bit_count) {
inline 8, 16, 32, 64 => |bits| {
if (bits == 8 and T == .Float) return RuntimeError.InvalidSpirV;
const ScalarT = getValuePrimitiveFieldType(T, bits);
const d_field = try getValuePrimitiveField(T, bits, d);
const l_field = try getValuePrimitiveField(T, bits, @constCast(l));
const r_field = try getValuePrimitiveField(T, bits, @constCast(r));
d_field.* = try operation(ScalarT, l_field.*, r_field.*);
},
else => return RuntimeError.InvalidSpirV,
}
}
inline fn applySIMDVector(comptime ElemT: type, comptime N: usize, d: *@Vector(N, ElemT), l: *const @Vector(N, ElemT), r: *const @Vector(N, ElemT)) RuntimeError!void {
inline for (0..N) |i| {
d[i] = try operation(ElemT, l[i], r[i]);
}
}
};
switch (dst.*) {
.Int, .Float => try operator.applyScalar(lane_bits, dst, lhs, rhs),
.Vector => |dst_vec| for (dst_vec, lhs.Vector, rhs.Vector) |*d_lane, l_lane, r_lane| {
try operator.applyScalar(lane_bits, d_lane, &l_lane, &r_lane);
},
.Vector4f32 => |*d| try operator.applySIMDVector(f32, 4, d, &lhs.Vector4f32, &rhs.Vector4f32),
.Vector3f32 => |*d| try operator.applySIMDVector(f32, 3, d, &lhs.Vector3f32, &rhs.Vector3f32),
.Vector2f32 => |*d| try operator.applySIMDVector(f32, 2, d, &lhs.Vector2f32, &rhs.Vector2f32),
.Vector4i32 => |*d| try operator.applySIMDVector(i32, 4, d, &lhs.Vector4i32, &rhs.Vector4i32),
.Vector3i32 => |*d| try operator.applySIMDVector(i32, 3, d, &lhs.Vector3i32, &rhs.Vector3i32),
.Vector2i32 => |*d| try operator.applySIMDVector(i32, 2, d, &lhs.Vector2i32, &rhs.Vector2i32),
.Vector4u32 => |*d| try operator.applySIMDVector(u32, 4, d, &lhs.Vector4u32, &rhs.Vector4u32),
.Vector3u32 => |*d| try operator.applySIMDVector(u32, 3, d, &lhs.Vector3u32, &rhs.Vector3u32),
.Vector2u32 => |*d| try operator.applySIMDVector(u32, 2, d, &lhs.Vector2u32, &rhs.Vector2u32),
else => return RuntimeError.InvalidSpirV,
}
}
};
}
inline fn sumSIMDVector(comptime ElemT: type, comptime N: usize, d: *ElemT, v: *const @Vector(N, ElemT)) void {
inline for (0..N) |i| {
d.* += v[i];
}
}
fn opLength(_: std.mem.Allocator, target_type_id: SpvWord, id: SpvWord, _: SpvWord, rt: *Runtime) RuntimeError!void {
const target_type = (try rt.results[target_type_id].getVariant()).Type;
const dst = try rt.results[id].getValue();
const src = try rt.results[try rt.it.next()].getValue();
const lane_bits = try Result.resolveLaneBitWidth(target_type, rt);
switch (lane_bits) {
inline 16, 32, 64 => |bits| {
var sum: std.meta.Float(bits) = 0.0;
const d_field = try getValuePrimitiveField(.Float, bits, dst);
if (bits == 32) { // More likely to be SIMD if f32
switch (src.*) {
.Vector4f32 => |src_vec| sumSIMDVector(f32, 4, &sum, &src_vec),
.Vector3f32 => |src_vec| sumSIMDVector(f32, 3, &sum, &src_vec),
.Vector2f32 => |src_vec| sumSIMDVector(f32, 2, &sum, &src_vec),
else => {},
}
}
switch (src.*) {
.Float => {
// Fast path
const s_field = try getValuePrimitiveField(.Float, bits, src);
d_field.* = s_field.*;
return;
},
.Vector => |src_vec| for (src_vec) |*s_lane| {
const s_field = try getValuePrimitiveField(.Float, bits, s_lane);
sum += s_field.*;
},
.Vector4f32, .Vector3f32, .Vector2f32 => {},
else => return RuntimeError.InvalidSpirV,
}
d_field.* = @sqrt(sum);
},
else => return RuntimeError.InvalidSpirV,
}
}
fn opNormalize(_: std.mem.Allocator, target_type_id: SpvWord, id: SpvWord, _: SpvWord, rt: *Runtime) RuntimeError!void {
const target_type = (try rt.results[target_type_id].getVariant()).Type;
const dst = try rt.results[id].getValue();
const src = try rt.results[try rt.it.next()].getValue();
const lane_bits = try Result.resolveLaneBitWidth(target_type, rt);
switch (lane_bits) {
inline 16, 32, 64 => |bits| {
var sum: std.meta.Float(bits) = 0.0;
if (bits == 32) { // More likely to be SIMD if f32
switch (src.*) {
.Vector4f32 => |src_vec| sumSIMDVector(f32, 4, &sum, &src_vec),
.Vector3f32 => |src_vec| sumSIMDVector(f32, 3, &sum, &src_vec),
.Vector2f32 => |src_vec| sumSIMDVector(f32, 2, &sum, &src_vec),
else => {},
}
}
switch (src.*) {
.Float => {
const s_field = try getValuePrimitiveField(.Float, bits, src);
sum = s_field.*;
},
.Vector => |src_vec| for (src_vec) |*s_lane| {
const s_field = try getValuePrimitiveField(.Float, bits, s_lane);
sum += s_field.*;
},
.Vector4f32, .Vector3f32, .Vector2f32 => {},
else => return RuntimeError.InvalidSpirV,
}
sum = @sqrt(sum);
if (bits == 32) {
switch (dst.*) {
.Vector4f32 => |*dst_vec| inline for (0..4) |i| {
dst_vec[i] = src.Vector4f32[i] / sum;
},
.Vector3f32 => |*dst_vec| inline for (0..3) |i| {
dst_vec[i] = src.Vector3f32[i] / sum;
},
.Vector2f32 => |*dst_vec| inline for (0..2) |i| {
dst_vec[i] = src.Vector2f32[i] / sum;
},
else => {},
}
}
switch (dst.*) {
.Vector => |dst_vec| for (dst_vec, src.Vector) |*d_lane, *s_lane| {
const d_field = try getValuePrimitiveField(.Float, bits, d_lane);
const s_field = try getValuePrimitiveField(.Float, bits, s_lane);
d_field.* = s_field.* / sum;
},
.Vector4f32, .Vector3f32, .Vector2f32 => {},
else => return RuntimeError.InvalidSpirV,
}
},
else => return RuntimeError.InvalidSpirV,
}
}

View File

@@ -2,7 +2,8 @@ const std = @import("std");
const spv = @import("spv.zig");
const op = @import("opcodes.zig");
const RuntimeError = @import("Runtime.zig").RuntimeError;
const Runtime = @import("Runtime.zig");
const RuntimeError = Runtime.RuntimeError;
const SpvVoid = spv.SpvVoid;
const SpvByte = spv.SpvByte;
@@ -227,13 +228,7 @@ pub const Value = union(Type) {
}
};
pub const VariantData = union(Variant) {
String: []const u8,
Extension: struct {
/// Should not be allocated but rather a pointer to a static array
dispatcher: []op.OpCodeExtFunc,
},
Type: union(Type) {
pub const TypeData = union(Type) {
Void: struct {},
Bool: struct {},
Int: struct {
@@ -285,7 +280,15 @@ pub const VariantData = union(Variant) {
storage_class: spv.SpvStorageClass,
target: SpvWord,
},
};
pub const VariantData = union(Variant) {
String: []const u8,
Extension: struct {
/// Should not be allocated but rather a pointer to a static array
dispatcher: []?op.OpCodeExtFunc,
},
Type: TypeData,
Variable: struct {
storage_class: spv.SpvStorageClass,
type_word: SpvWord,
@@ -364,7 +367,7 @@ pub fn deinit(self: *Self, allocator: std.mem.Allocator) void {
self.decorations.deinit(allocator);
}
pub fn getValueTypeWord(self: *Self) RuntimeError!SpvWord {
pub inline fn getValueTypeWord(self: *Self) RuntimeError!SpvWord {
return switch ((try self.getVariant()).*) {
.Variable => |v| v.type_word,
.Constant => |c| c.type_word,
@@ -374,7 +377,7 @@ pub fn getValueTypeWord(self: *Self) RuntimeError!SpvWord {
};
}
pub fn getValueType(self: *Self) RuntimeError!Type {
pub inline fn getValueType(self: *Self) RuntimeError!Type {
return switch ((try self.getVariant()).*) {
.Variable => |v| v.type,
.Constant => |c| c.type,
@@ -383,7 +386,7 @@ pub fn getValueType(self: *Self) RuntimeError!Type {
};
}
pub fn getValue(self: *Self) RuntimeError!*Value {
pub inline fn getValue(self: *Self) RuntimeError!*Value {
return switch ((try self.getVariant()).*) {
.Variable => |*v| &v.value,
.Constant => |*c| &c.value,
@@ -471,6 +474,26 @@ pub fn dupe(self: *const Self, allocator: std.mem.Allocator) RuntimeError!Self {
};
}
pub fn resolveLaneBitWidth(target_type: TypeData, rt: *const Runtime) RuntimeError!SpvWord {
return sw: switch (target_type) {
.Bool => 8,
.Float => |f| f.bit_length,
.Int => |i| i.bit_length,
.Vector => |v| continue :sw (try rt.results[v.components_type_word].getVariant()).Type,
.Vector4f32,
.Vector3f32,
.Vector2f32,
.Vector4i32,
.Vector3i32,
.Vector2i32,
.Vector4u32,
.Vector3u32,
.Vector2u32,
=> return 32,
else => return RuntimeError.InvalidSpirV,
};
}
pub fn resolveType(self: *const Self, results: []const Self) *const Self {
return if (self.variant) |variant|
switch (variant) {

View File

@@ -131,10 +131,7 @@ pub fn callEntryPoint(self: *Self, allocator: std.mem.Allocator, entry_point_ind
var it_tmp = self.it; // Save because operations may iter on this iterator
if (op.runtime_dispatcher[opcode]) |pfn| {
pfn(allocator, word_count, self) catch |err| switch (err) {
RuntimeError.Killed => return,
else => return err,
};
try pfn(allocator, word_count, self);
}
if (!self.it.did_jump) {
_ = it_tmp.skipN(word_count);

View File

@@ -36,3 +36,10 @@ pub const Runtime = @import("Runtime.zig");
const opcodes = @import("opcodes.zig");
const spv = @import("spv.zig");
pub const SpvVoid = spv.SpvVoid;
pub const SpvByte = spv.SpvByte;
pub const SpvWord = spv.SpvWord;
pub const SpvBool = spv.SpvBool;
pub const GLSL_std_450 = @import("GLSL_std_450/opcodes.zig");

View File

@@ -1,6 +1,8 @@
const std = @import("std");
const spv = @import("spv.zig");
const GLSL_std_450 = @import("GLSL_std_450/opcodes.zig");
const Module = @import("Module.zig");
const Runtime = @import("Runtime.zig");
const Result = @import("Result.zig");
@@ -13,13 +15,7 @@ const SpvByte = spv.SpvByte;
const SpvWord = spv.SpvWord;
const SpvBool = spv.SpvBool;
// OpExtInst Sin
// OpExtInst Cos
// OpExtInst Length
// OpExtInst Normalize
// OpExtInst FMax
const ValueType = enum {
pub const ValueType = enum {
Bool,
Float,
SInt,
@@ -230,11 +226,11 @@ pub fn initRuntimeDispatcher() void {
runtime_dispatcher[@intFromEnum(spv.SpvOp.ISub)] = MathEngine(.SInt, .Sub).op;
runtime_dispatcher[@intFromEnum(spv.SpvOp.Kill)] = opKill;
runtime_dispatcher[@intFromEnum(spv.SpvOp.Load)] = opLoad;
runtime_dispatcher[@intFromEnum(spv.SpvOp.LogicalAnd)] = CondEngine(.Float, .LogicalAnd).op;
runtime_dispatcher[@intFromEnum(spv.SpvOp.LogicalEqual)] = CondEngine(.Float, .LogicalEqual).op;
runtime_dispatcher[@intFromEnum(spv.SpvOp.LogicalNot)] = CondEngine(.Float, .LogicalNot).op;
runtime_dispatcher[@intFromEnum(spv.SpvOp.LogicalNotEqual)] = CondEngine(.Float, .LogicalNotEqual).op;
runtime_dispatcher[@intFromEnum(spv.SpvOp.LogicalOr)] = CondEngine(.Float, .LogicalOr).op;
runtime_dispatcher[@intFromEnum(spv.SpvOp.LogicalAnd)] = CondEngine(.Bool, .LogicalAnd).op;
runtime_dispatcher[@intFromEnum(spv.SpvOp.LogicalEqual)] = CondEngine(.Bool, .LogicalEqual).op;
runtime_dispatcher[@intFromEnum(spv.SpvOp.LogicalNot)] = CondEngine(.Bool, .LogicalNot).op;
runtime_dispatcher[@intFromEnum(spv.SpvOp.LogicalNotEqual)] = CondEngine(.Bool, .LogicalNotEqual).op;
runtime_dispatcher[@intFromEnum(spv.SpvOp.LogicalOr)] = CondEngine(.Bool, .LogicalOr).op;
runtime_dispatcher[@intFromEnum(spv.SpvOp.MatrixTimesMatrix)] = MathEngine(.Float, .MatrixTimesMatrix).op; // TODO
runtime_dispatcher[@intFromEnum(spv.SpvOp.MatrixTimesScalar)] = MathEngine(.Float, .MatrixTimesScalar).op; // TODO
runtime_dispatcher[@intFromEnum(spv.SpvOp.MatrixTimesVector)] = MathEngine(.Float, .MatrixTimesVector).op; // TODO
@@ -261,38 +257,37 @@ pub fn initRuntimeDispatcher() void {
runtime_dispatcher[@intFromEnum(spv.SpvOp.UMod)] = MathEngine(.UInt, .Mod).op;
runtime_dispatcher[@intFromEnum(spv.SpvOp.VectorTimesMatrix)] = MathEngine(.Float, .VectorTimesMatrix).op; // TODO
runtime_dispatcher[@intFromEnum(spv.SpvOp.VectorTimesScalar)] = MathEngine(.Float, .VectorTimesScalar).op;
runtime_dispatcher[@intFromEnum(spv.SpvOp.ExtInst)] = opExtInst;
// zig fmt: on
// Extensions init
GLSL_std_450.initRuntimeDispatcher();
}
fn BitEngine(comptime T: ValueType, comptime Op: BitOp) type {
if (T == .Float) @compileError("Invalid value type");
fn extEqlName(a: []const u8, b: []const u8) bool {
for (0..@min(a.len, b.len)) |i| {
if (a[i] != b[i]) return false;
}
return true;
}
const extensions_map = std.StaticStringMapWithEql([]?OpCodeExtFunc, extEqlName).initComptime(.{
.{ "GLSL.std.450", GLSL_std_450.runtime_dispatcher[0..] },
});
fn BitOperator(comptime T: ValueType, comptime Op: BitOp) type {
return struct {
fn op(_: std.mem.Allocator, _: SpvWord, rt: *Runtime) RuntimeError!void {
const target_type = (try rt.results[try rt.it.next()].getVariant()).Type;
const value = try rt.results[try rt.it.next()].getValue();
const op1_value = try rt.results[try rt.it.next()].getValue();
const op2_value: ?*Result.Value = switch (Op) {
.Not, .BitCount, .BitReverse => null,
else => try rt.results[try rt.it.next()].getValue(),
};
comptime {
if (T == .Float) @compileError("Invalid value type");
}
const size = sw: switch (target_type) {
.Vector => |v| continue :sw (try rt.results[v.components_type_word].getVariant()).Type,
.Vector4f32,
.Vector3f32,
.Vector2f32,
.Vector4i32,
.Vector3i32,
.Vector2i32,
.Vector4u32,
.Vector3u32,
.Vector2u32,
=> 32,
.Int => |i| i.bit_length,
else => return RuntimeError.InvalidSpirV,
inline fn isUnaryOp() bool {
return comptime switch (Op) {
.Not, .BitCount, .BitReverse => true,
else => false,
};
}
const operator = struct {
inline fn bitMask(bits: u64) u64 {
return if (bits >= 32) ~@as(u64, 0) else (@as(u64, 0x1) << @intCast(bits)) - 1;
}
@@ -306,85 +301,227 @@ fn BitEngine(comptime T: ValueType, comptime Op: BitOp) type {
return (v >> @intCast(offset)) & @as(TT, @intCast(bitMask(count)));
}
fn operation(comptime TT: type, rt2: *Runtime, op1: TT, op2: ?TT) RuntimeError!TT {
switch (Op) {
.BitCount => return @bitSizeOf(TT),
.BitReverse => return @bitReverse(op1),
.Not => return ~op1,
else => {},
fn operationUnary(comptime TT: type, op1: TT) RuntimeError!TT {
return switch (Op) {
.BitCount => @as(TT, @intCast(@bitSizeOf(TT))), // keep return type TT
.BitReverse => @bitReverse(op1),
.Not => ~op1,
else => RuntimeError.InvalidSpirV,
};
}
return if (op2) |v2|
switch (Op) {
fn operationBinary(comptime TT: type, rt: *Runtime, op1: TT, op2: TT) RuntimeError!TT {
return switch (Op) {
.BitFieldInsert => blk: {
const offset = try rt2.results[try rt2.it.next()].getValue();
const count = try rt2.results[try rt2.it.next()].getValue();
break :blk bitInsert(TT, op1, v2, offset.Int.uint64, count.Int.uint64);
const offset = try rt.results[try rt.it.next()].getValue();
const count = try rt.results[try rt.it.next()].getValue();
break :blk bitInsert(TT, op1, op2, offset.Int.uint64, count.Int.uint64);
},
.BitFieldSExtract => blk: {
if (T == .UInt) return RuntimeError.InvalidSpirV;
const count = try rt2.results[try rt2.it.next()].getValue();
break :blk bitExtract(TT, op1, v2, count.Int.uint64);
const count = try rt.results[try rt.it.next()].getValue();
break :blk bitExtract(TT, op1, op2, count.Int.uint64);
},
.BitFieldUExtract => blk: {
if (T == .SInt) return RuntimeError.InvalidSpirV;
const count = try rt2.results[try rt2.it.next()].getValue();
break :blk bitExtract(TT, op1, v2, count.Int.uint64);
const count = try rt.results[try rt.it.next()].getValue();
break :blk bitExtract(TT, op1, op2, count.Int.uint64);
},
.BitwiseAnd => op1 & v2,
.BitwiseOr => op1 | v2,
.BitwiseXor => op1 ^ v2,
.ShiftLeft => op1 << @intCast(v2),
.ShiftRight, .ShiftRightArithmetic => op1 >> @intCast(v2),
else => return RuntimeError.InvalidSpirV,
}
else
RuntimeError.InvalidSpirV;
.BitwiseAnd => op1 & op2,
.BitwiseOr => op1 | op2,
.BitwiseXor => op1 ^ op2,
.ShiftLeft => op1 << @intCast(op2),
.ShiftRight, .ShiftRightArithmetic => op1 >> @intCast(op2),
else => RuntimeError.InvalidSpirV,
};
}
fn process(rt2: *Runtime, bit_count: SpvWord, v: *Result.Value, op1_v: *const Result.Value, op2_v: ?*const Result.Value) RuntimeError!void {
fn applyScalarBits(rt: *Runtime, bit_count: SpvWord, dst: *Result.Value, op1_v: *const Result.Value, op2_v: ?*const Result.Value) RuntimeError!void {
switch (bit_count) {
inline 8, 16, 32, 64 => |i| {
(try getValuePrimitiveField(T, i, v)).* = try operation(
getValuePrimitiveFieldType(T, i),
rt2,
(try getValuePrimitiveField(T, i, @constCast(op1_v))).*,
if (op2_v) |v2|
(try getValuePrimitiveField(T, i, @constCast(v2))).*
else
null,
);
inline 8, 16, 32, 64 => |bits| {
const TT = getValuePrimitiveFieldType(T, bits);
const a = (try getValuePrimitiveField(T, bits, @constCast(op1_v))).*;
const out = if (comptime isUnaryOp()) blk: {
break :blk try operationUnary(TT, a);
} else blk: {
const b_ptr = op2_v orelse return RuntimeError.InvalidSpirV;
const b = (try getValuePrimitiveField(T, bits, @constCast(b_ptr))).*;
break :blk try operationBinary(TT, rt, a, b);
};
(try getValuePrimitiveField(T, bits, dst)).* = out;
},
else => return RuntimeError.InvalidSpirV,
}
}
fn laneRhsPtr(op2_value: ?*Result.Value, index: usize) ?*const Result.Value {
if (comptime isUnaryOp()) return null;
const v = op2_value orelse return null;
return &v.Vector[index];
}
fn applyFixedVector(comptime ElemT: type, comptime N: usize, dst: *[N]ElemT, op1: *[N]ElemT, op2_value: ?*Result.Value) RuntimeError!void {
if (comptime isUnaryOp()) {
inline for (0..N) |i| dst[i] = try operationUnary(ElemT, op1[i]);
} else {
const op2 = op2_value orelse return RuntimeError.InvalidSpirV;
const b: *const [N]ElemT = switch (N) {
2 => &op2.*.Vector2u32, // will be overridden by call sites per ElemT/tag
3 => &op2.*.Vector3u32,
4 => &op2.*.Vector4u32,
else => unreachable,
};
// NOTE: the above dummy mapping isnt type-correct for i32; call sites below pass correct rhs pointer.
_ = b;
return RuntimeError.InvalidSpirV;
}
}
fn applyFixedVectorBinary(
comptime ElemT: type,
comptime N: usize,
rt: *Runtime,
dst: *[N]ElemT,
op1: *[N]ElemT,
op2: *[N]ElemT,
) RuntimeError!void {
inline for (0..N) |i| dst[i] = try operationBinary(ElemT, rt, op1[i], op2[i]);
}
fn applyFixedVectorUnary(
comptime ElemT: type,
comptime N: usize,
dst: *[N]ElemT,
op1: *[N]ElemT,
) RuntimeError!void {
inline for (0..N) |i| dst[i] = try operationUnary(ElemT, op1[i]);
}
};
}
fn BitEngine(comptime T: ValueType, comptime Op: BitOp) type {
return struct {
fn op(_: std.mem.Allocator, _: SpvWord, rt: *Runtime) RuntimeError!void {
const target_type = (try rt.results[try rt.it.next()].getVariant()).Type;
const dst = try rt.results[try rt.it.next()].getValue();
const op1 = try rt.results[try rt.it.next()].getValue();
const operator = BitOperator(T, Op);
const op2_value: ?*Result.Value = if (comptime operator.isUnaryOp()) null else try rt.results[try rt.it.next()].getValue();
const lane_bits = try Result.resolveLaneBitWidth(target_type, rt);
switch (dst.*) {
.Int => try operator.applyScalarBits(rt, lane_bits, dst, op1, if (comptime operator.isUnaryOp()) null else op2_value),
.Vector => |dst_vec| {
const op1_vec = op1.Vector;
if (dst_vec.len != op1_vec.len) return RuntimeError.InvalidSpirV;
for (dst_vec, op1_vec, 0..) |*d_lane, a_lane, i| {
var tmp_a = a_lane;
const b_ptr = operator.laneRhsPtr(op2_value, i);
try operator.applyScalarBits(rt, lane_bits, d_lane, &tmp_a, b_ptr);
}
},
.Vector4i32 => |*d| {
if (comptime operator.isUnaryOp())
try operator.applyFixedVectorUnary(i32, 4, d, &op1.Vector4i32)
else
try operator.applyFixedVectorBinary(i32, 4, rt, d, &op1.Vector4i32, &op2_value.?.Vector4i32);
},
.Vector3i32 => |*d| {
if (comptime operator.isUnaryOp())
try operator.applyFixedVectorUnary(i32, 3, d, &op1.Vector3i32)
else
try operator.applyFixedVectorBinary(i32, 3, rt, d, &op1.Vector3i32, &op2_value.?.Vector3i32);
},
.Vector2i32 => |*d| {
if (comptime operator.isUnaryOp())
try operator.applyFixedVectorUnary(i32, 2, d, &op1.Vector2i32)
else
try operator.applyFixedVectorBinary(i32, 2, rt, d, &op1.Vector2i32, &op2_value.?.Vector2i32);
},
.Vector4u32 => |*d| {
if (comptime operator.isUnaryOp())
try operator.applyFixedVectorUnary(u32, 4, d, &op1.Vector4u32)
else
try operator.applyFixedVectorBinary(u32, 4, rt, d, &op1.Vector4u32, &op2_value.?.Vector4u32);
},
.Vector3u32 => |*d| {
if (comptime operator.isUnaryOp())
try operator.applyFixedVectorUnary(u32, 3, d, &op1.Vector3u32)
else
try operator.applyFixedVectorBinary(u32, 3, rt, d, &op1.Vector3u32, &op2_value.?.Vector3u32);
},
.Vector2u32 => |*d| {
if (comptime operator.isUnaryOp())
try operator.applyFixedVectorUnary(u32, 2, d, &op1.Vector2u32)
else
try operator.applyFixedVectorBinary(u32, 2, rt, d, &op1.Vector2u32, &op2_value.?.Vector2u32);
},
else => return RuntimeError.InvalidSpirV,
}
}
};
}
switch (value.*) {
.Int => try operator.process(rt, size, value, op1_value, op2_value),
.Vector => |vec| for (vec, op1_value.Vector, 0..) |*val, op1_v, i|
try operator.process(rt, size, val, &op1_v, if (op2_value) |op2_v| &op2_v.Vector[i] else null),
// No bit manipulation on VectorXf32
.Vector4i32 => |*vec| inline for (0..4) |i| {
vec[i] = try operator.operation(i32, rt, op1_value.Vector4i32[i], if (op2_value) |op2_v| op2_v.Vector4i32[i] else null);
},
.Vector3i32 => |*vec| inline for (0..3) |i| {
vec[i] = try operator.operation(i32, rt, op1_value.Vector3i32[i], if (op2_value) |op2_v| op2_v.Vector3i32[i] else null);
},
.Vector2i32 => |*vec| inline for (0..2) |i| {
vec[i] = try operator.operation(i32, rt, op1_value.Vector2i32[i], if (op2_value) |op2_v| op2_v.Vector2i32[i] else null);
},
.Vector4u32 => |*vec| inline for (0..4) |i| {
vec[i] = try operator.operation(u32, rt, op1_value.Vector4u32[i], if (op2_value) |op2_v| op2_v.Vector4u32[i] else null);
},
.Vector3u32 => |*vec| inline for (0..3) |i| {
vec[i] = try operator.operation(u32, rt, op1_value.Vector3u32[i], if (op2_value) |op2_v| op2_v.Vector3u32[i] else null);
},
.Vector2u32 => |*vec| inline for (0..2) |i| {
vec[i] = try operator.operation(u32, rt, op1_value.Vector2u32[i], if (op2_value) |op2_v| op2_v.Vector2u32[i] else null);
fn CondOperator(comptime T: ValueType, comptime Op: CondOp) type {
return struct {
fn operation(comptime TT: type, a: TT, b: TT) RuntimeError!bool {
return switch (Op) {
.Equal, .LogicalEqual => a == b,
.NotEqual, .LogicalNotEqual => a != b,
.Greater => a > b,
.GreaterEqual => a >= b,
.Less => a < b,
.LessEqual => a <= b,
.LogicalAnd => a and b,
.LogicalOr => a or b,
else => RuntimeError.InvalidSpirV,
};
}
fn operationUnary(comptime TT: type, a: TT) RuntimeError!bool {
return switch (Op) {
.LogicalNot => !a,
else => RuntimeError.InvalidSpirV,
};
}
fn applyLane(bit_count: SpvWord, dst_bool: *Result.Value, a_v: *const Result.Value, b_v: ?*const Result.Value) RuntimeError!void {
switch (bit_count) {
inline 8, 16, 32, 64 => |bits| {
if (bits == 8 and T == .Float) return RuntimeError.InvalidSpirV;
const TT = getValuePrimitiveFieldType(T, bits);
const a = (try getValuePrimitiveField(T, bits, @constCast(a_v))).*;
if (comptime Op == .LogicalNot) {
dst_bool.Bool = try operationUnary(TT, a);
} else {
const b_ptr = b_v orelse return RuntimeError.InvalidSpirV;
const b = (try getValuePrimitiveField(T, bits, @constCast(b_ptr))).*;
dst_bool.Bool = try operation(TT, a, b);
}
},
else => return RuntimeError.InvalidSpirV,
}
}
fn laneRhsPtr(op2_value: ?*Result.Value, index: usize) ?*const Result.Value {
if (comptime Op == .LogicalNot) return null;
const v = op2_value orelse return null;
return &v.Vector[index];
}
};
}
@@ -397,199 +534,151 @@ fn CondEngine(comptime T: ValueType, comptime Op: CondOp) type {
else => return RuntimeError.InvalidSpirV,
}
const value = try rt.results[try rt.it.next()].getValue();
const dst = try rt.results[try rt.it.next()].getValue();
const op1_result = &rt.results[try rt.it.next()];
const op1_type = try op1_result.getValueTypeWord();
const op1_value = try op1_result.getValue();
const op2_value: ?*Result.Value = switch (Op) {
.LogicalNot => null,
else => try rt.results[try rt.it.next()].getValue(),
};
const size = sw: switch ((try rt.results[op1_type].getVariant()).Type) {
.Vector => |v| continue :sw (try rt.results[v.components_type_word].getVariant()).Type,
.Vector4f32,
.Vector3f32,
.Vector2f32,
.Vector4i32,
.Vector3i32,
.Vector2i32,
.Vector4u32,
.Vector3u32,
.Vector2u32,
=> 32,
.Float => |f| if (T == .Float) f.bit_length else return RuntimeError.InvalidSpirV,
.Int => |i| if (T == .SInt or T == .UInt) i.bit_length else return RuntimeError.InvalidSpirV,
else => return RuntimeError.InvalidSpirV,
};
const op2_value: ?*Result.Value = if (comptime Op == .LogicalNot) null else try rt.results[try rt.it.next()].getValue();
const operator = struct {
fn operation(comptime TT: type, op1: TT, op2: ?TT) RuntimeError!bool {
return switch (Op) {
.Equal, .LogicalEqual => op1 == op2 orelse return RuntimeError.InvalidSpirV,
.NotEqual, .LogicalNotEqual => op1 != op2 orelse return RuntimeError.InvalidSpirV,
.Greater => op1 > op2 orelse return RuntimeError.InvalidSpirV,
.GreaterEqual => op1 >= op2 orelse return RuntimeError.InvalidSpirV,
.Less => op1 < op2 orelse return RuntimeError.InvalidSpirV,
.LessEqual => op1 <= op2 orelse return RuntimeError.InvalidSpirV,
.LogicalAnd => (op1 != @as(TT, 0)) and ((op2 orelse return RuntimeError.InvalidSpirV) != @as(TT, 0)),
.LogicalOr => (op1 != @as(TT, 0)) or ((op2 orelse return RuntimeError.InvalidSpirV) != @as(TT, 0)),
.LogicalNot => (op1 == @as(TT, 0)),
};
}
const lane_bits = try Result.resolveLaneBitWidth((try rt.results[op1_type].getVariant()).Type, rt);
fn process(bit_count: SpvWord, v: *Result.Value, op1_v: *const Result.Value, op2_v: ?*const Result.Value) RuntimeError!void {
switch (bit_count) {
inline 8, 16, 32, 64 => |i| {
if (i == 8 and T == .Float) { // No f8
return RuntimeError.InvalidSpirV;
}
v.Bool = try operation(
getValuePrimitiveFieldType(T, i),
(try getValuePrimitiveField(T, i, @constCast(op1_v))).*,
if (op2_v) |val| (try getValuePrimitiveField(T, i, @constCast(val))).* else null,
);
const operator = CondOperator(T, Op);
switch (dst.*) {
.Bool => try operator.applyLane(lane_bits, dst, op1_value, op2_value),
.Vector => |dst_vec| for (dst_vec, op1_value.Vector, 0..) |*d_lane, a_lane, i| {
const b_ptr = operator.laneRhsPtr(op2_value, i);
try operator.applyLane(lane_bits, d_lane, &a_lane, b_ptr);
},
else => return RuntimeError.InvalidSpirV,
}
}
};
switch (value.*) {
.Bool => try operator.process(size, value, op1_value, op2_value),
.Vector => |vec| for (vec, op1_value.Vector, 0..) |*val, op1_v, i| {
try operator.process(size, val, &op1_v, if (op2_value) |op2_v| &op2_v.Vector[i] else null);
},
// No Vector specializations for booleans
else => return RuntimeError.InvalidSpirV,
}
}
};
}
fn ConversionEngine(comptime From: ValueType, comptime To: ValueType) type {
fn ConversionEngine(comptime from_kind: ValueType, comptime to_kind: ValueType) type {
return struct {
fn op(_: std.mem.Allocator, _: SpvWord, rt: *Runtime) RuntimeError!void {
const target_type = (try rt.results[try rt.it.next()].getVariant()).Type;
const value = try rt.results[try rt.it.next()].getValue();
const op_result = &rt.results[try rt.it.next()];
const op_type = try op_result.getValueTypeWord();
const op_value = try op_result.getValue();
const dst_value = try rt.results[try rt.it.next()].getValue();
const from_size = sw: switch ((try rt.results[op_type].getVariant()).Type) {
.Vector => |v| continue :sw (try rt.results[v.components_type_word].getVariant()).Type,
.Vector4f32,
.Vector3f32,
.Vector2f32,
.Vector4i32,
.Vector3i32,
.Vector2i32,
.Vector4u32,
.Vector3u32,
.Vector2u32,
=> 32,
.Float => |f| if (From == .Float) f.bit_length else return RuntimeError.InvalidSpirV,
.Int => |i| if (From == .SInt or From == .UInt) i.bit_length else return RuntimeError.InvalidSpirV,
const src_result = &rt.results[try rt.it.next()];
const src_type_word = try src_result.getValueTypeWord();
const src_value = try src_result.getValue();
const from_bits = try Result.resolveLaneBitWidth((try rt.results[src_type_word].getVariant()).Type, rt);
const to_bits = try Result.resolveLaneBitWidth(target_type, rt);
const caster = struct {
fn castLane(comptime ToT: type, from_bit_count: SpvWord, from: *Result.Value) RuntimeError!ToT {
return switch (from_bit_count) {
inline 8, 16, 32, 64 => |bits| blk: {
if (bits == 8 and from_kind == .Float) return RuntimeError.InvalidSpirV; // No f8
const v = (try getValuePrimitiveField(from_kind, bits, from)).*;
break :blk std.math.lossyCast(ToT, v);
},
else => return RuntimeError.InvalidSpirV,
};
}
const to_size = sw: switch (target_type) {
.Vector => |v| continue :sw (try rt.results[v.components_type_word].getVariant()).Type,
.Vector4f32,
.Vector3f32,
.Vector2f32,
.Vector4i32,
.Vector3i32,
.Vector2i32,
.Vector4u32,
.Vector3u32,
.Vector2u32,
=> 32,
.Float => |f| if (To == .Float) f.bit_length else return RuntimeError.InvalidSpirV,
.Int => |i| if (To == .SInt or To == .UInt) i.bit_length else return RuntimeError.InvalidSpirV,
else => return RuntimeError.InvalidSpirV,
};
const operator = struct {
fn process(from_bit_count: SpvWord, to_bit_count: SpvWord, to: *Result.Value, from: *Result.Value) RuntimeError!void {
fn applyScalar(from_bit_count: SpvWord, to_bit_count: SpvWord, dst: *Result.Value, from: *Result.Value) RuntimeError!void {
switch (to_bit_count) {
inline 8, 16, 32, 64 => |i| {
if (i == 8 and To == .Float) {
return RuntimeError.InvalidSpirV; // No f8
}
const ToType = getValuePrimitiveFieldType(To, i);
(try getValuePrimitiveField(To, i, to)).* = std.math.lossyCast(
ToType,
switch (from_bit_count) {
inline 8, 16, 32, 64 => |j| blk: {
if (j == 8 and From == .Float) {
return RuntimeError.InvalidSpirV; // Same
}
break :blk (try getValuePrimitiveField(From, j, from)).*;
},
else => return RuntimeError.InvalidSpirV,
},
);
inline 8, 16, 32, 64 => |bits| {
if (bits == 8 and to_kind == .Float) return RuntimeError.InvalidSpirV; // No f8
const ToT = getValuePrimitiveFieldType(to_kind, bits);
(try getValuePrimitiveField(to_kind, bits, dst)).* = try castLane(ToT, from_bit_count, from);
},
else => return RuntimeError.InvalidSpirV,
}
}
fn processVecSpe(comptime T: type, from_bit_count: SpvWord, from: *Result.Value, index: usize) RuntimeError!T {
return switch (from.*) {
.Vector3f32 => |vec| std.math.lossyCast(T, vec[index]),
.Vector2f32 => |vec| std.math.lossyCast(T, vec[index]),
.Vector4i32 => |vec| std.math.lossyCast(T, vec[index]),
.Vector3i32 => |vec| std.math.lossyCast(T, vec[index]),
.Vector2i32 => |vec| std.math.lossyCast(T, vec[index]),
.Vector4u32 => |vec| std.math.lossyCast(T, vec[index]),
.Vector3u32 => |vec| std.math.lossyCast(T, vec[index]),
.Vector2u32 => |vec| std.math.lossyCast(T, vec[index]),
inline else => switch (from_bit_count) {
inline 8, 16, 32, 64 => |i| std.math.lossyCast(T, blk: {
if (i == 8 and From == .Float) {
return RuntimeError.InvalidSpirV;
fn castSIMDVector(comptime ToT: type, comptime N: usize, dst_arr: *[N]ToT, src_arr: *const [N]ToT) void {
inline for (0..N) |i| dst_arr[i] = std.math.lossyCast(ToT, src_arr[i]);
}
break :blk (try getValuePrimitiveField(From, i, from)).*;
}),
else => return RuntimeError.InvalidSpirV,
},
};
fn castSIMDVectorFromOther(comptime ToT: type, comptime FromT: type, comptime N: usize, dst_arr: *[N]ToT, src_arr: *const [N]FromT) void {
inline for (0..N) |i| dst_arr[i] = std.math.lossyCast(ToT, src_arr[i]);
}
};
switch (value.*) {
.Float => if (To == .Float) try operator.process(from_size, to_size, value, op_value) else return RuntimeError.InvalidSpirV,
.Int => if (To == .SInt or To == .UInt) try operator.process(from_size, to_size, value, op_value) else return RuntimeError.InvalidSpirV,
.Vector => |vec| for (vec, op_value.Vector) |*val, *op_v| try operator.process(from_size, to_size, val, op_v),
.Vector4f32 => |*vec| inline for (0..4) |i| {
vec[i] = try operator.processVecSpe(f32, from_size, op_value, i);
switch (dst_value.*) {
.Float => {
if (to_kind != .Float) return RuntimeError.InvalidSpirV;
try caster.applyScalar(from_bits, to_bits, dst_value, src_value);
},
.Vector3f32 => |*vec| inline for (0..3) |i| {
vec[i] = try operator.processVecSpe(f32, from_size, op_value, i);
.Int => {
if (to_kind != .SInt and to_kind != .UInt) return RuntimeError.InvalidSpirV;
try caster.applyScalar(from_bits, to_bits, dst_value, src_value);
},
.Vector2f32 => |*vec| inline for (0..2) |i| {
vec[i] = try operator.processVecSpe(f32, from_size, op_value, i);
.Vector => |dst_vec| {
const src_vec = src_value.Vector;
if (dst_vec.len != src_vec.len) return RuntimeError.InvalidSpirV;
for (dst_vec, src_vec) |*d_lane, *s_lane| {
try caster.applyScalar(from_bits, to_bits, d_lane, s_lane);
}
},
.Vector4i32 => |*vec| inline for (0..4) |i| {
vec[i] = try operator.processVecSpe(i32, from_size, op_value, i);
.Vector4f32 => |*dst| switch (src_value.*) {
.Vector4f32 => caster.castSIMDVector(f32, 4, dst, &src_value.Vector4f32),
.Vector4i32 => caster.castSIMDVectorFromOther(f32, i32, 4, dst, &src_value.Vector4i32),
.Vector4u32 => caster.castSIMDVectorFromOther(f32, u32, 4, dst, &src_value.Vector4u32),
else => return RuntimeError.InvalidSpirV,
},
.Vector3i32 => |*vec| inline for (0..3) |i| {
vec[i] = try operator.processVecSpe(i32, from_size, op_value, i);
.Vector3f32 => |*dst| switch (src_value.*) {
.Vector3f32 => caster.castSIMDVector(f32, 3, dst, &src_value.Vector3f32),
.Vector3i32 => caster.castSIMDVectorFromOther(f32, i32, 3, dst, &src_value.Vector3i32),
.Vector3u32 => caster.castSIMDVectorFromOther(f32, u32, 3, dst, &src_value.Vector3u32),
else => return RuntimeError.InvalidSpirV,
},
.Vector2i32 => |*vec| inline for (0..2) |i| {
vec[i] = try operator.processVecSpe(i32, from_size, op_value, i);
.Vector2f32 => |*dst| switch (src_value.*) {
.Vector2f32 => caster.castSIMDVector(f32, 2, dst, &src_value.Vector2f32),
.Vector2i32 => caster.castSIMDVectorFromOther(f32, i32, 2, dst, &src_value.Vector2i32),
.Vector2u32 => caster.castSIMDVectorFromOther(f32, u32, 2, dst, &src_value.Vector2u32),
else => return RuntimeError.InvalidSpirV,
},
.Vector4u32 => |*vec| inline for (0..4) |i| {
vec[i] = try operator.processVecSpe(u32, from_size, op_value, i);
.Vector4i32 => |*dst| switch (src_value.*) {
.Vector4f32 => caster.castSIMDVectorFromOther(i32, f32, 4, dst, &src_value.Vector4f32),
.Vector4i32 => caster.castSIMDVector(i32, 4, dst, &src_value.Vector4i32),
.Vector4u32 => caster.castSIMDVectorFromOther(i32, u32, 4, dst, &src_value.Vector4u32),
else => return RuntimeError.InvalidSpirV,
},
.Vector3u32 => |*vec| inline for (0..3) |i| {
vec[i] = try operator.processVecSpe(u32, from_size, op_value, i);
.Vector3i32 => |*dst| switch (src_value.*) {
.Vector3f32 => caster.castSIMDVectorFromOther(i32, f32, 3, dst, &src_value.Vector3f32),
.Vector3i32 => caster.castSIMDVector(i32, 3, dst, &src_value.Vector3i32),
.Vector3u32 => caster.castSIMDVectorFromOther(i32, u32, 3, dst, &src_value.Vector3u32),
else => return RuntimeError.InvalidSpirV,
},
.Vector2u32 => |*vec| inline for (0..2) |i| {
vec[i] = try operator.processVecSpe(u32, from_size, op_value, i);
.Vector2i32 => |*dst| switch (src_value.*) {
.Vector2f32 => caster.castSIMDVectorFromOther(i32, f32, 2, dst, &src_value.Vector2f32),
.Vector2i32 => caster.castSIMDVector(i32, 2, dst, &src_value.Vector2i32),
.Vector2u32 => caster.castSIMDVectorFromOther(i32, u32, 2, dst, &src_value.Vector2u32),
else => return RuntimeError.InvalidSpirV,
},
.Vector4u32 => |*dst| switch (src_value.*) {
.Vector4f32 => caster.castSIMDVectorFromOther(u32, f32, 4, dst, &src_value.Vector4f32),
.Vector4i32 => caster.castSIMDVectorFromOther(u32, i32, 4, dst, &src_value.Vector4i32),
.Vector4u32 => caster.castSIMDVector(u32, 4, dst, &src_value.Vector4u32),
else => return RuntimeError.InvalidSpirV,
},
.Vector3u32 => |*dst| switch (src_value.*) {
.Vector3f32 => caster.castSIMDVectorFromOther(u32, f32, 3, dst, &src_value.Vector3f32),
.Vector3i32 => caster.castSIMDVectorFromOther(u32, i32, 3, dst, &src_value.Vector3i32),
.Vector3u32 => caster.castSIMDVector(u32, 3, dst, &src_value.Vector3u32),
else => return RuntimeError.InvalidSpirV,
},
.Vector2u32 => |*dst| switch (src_value.*) {
.Vector2f32 => caster.castSIMDVectorFromOther(u32, f32, 2, dst, &src_value.Vector2f32),
.Vector2i32 => caster.castSIMDVectorFromOther(u32, i32, 2, dst, &src_value.Vector2i32),
.Vector2u32 => caster.castSIMDVector(u32, 2, dst, &src_value.Vector2u32),
else => return RuntimeError.InvalidSpirV,
},
else => return RuntimeError.InvalidSpirV,
}
}
@@ -600,26 +689,11 @@ fn MathEngine(comptime T: ValueType, comptime Op: MathOp) type {
return struct {
fn op(_: std.mem.Allocator, _: SpvWord, rt: *Runtime) RuntimeError!void {
const target_type = (try rt.results[try rt.it.next()].getVariant()).Type;
const value = try rt.results[try rt.it.next()].getValue();
const op1_value = try rt.results[try rt.it.next()].getValue();
const op2_value = try rt.results[try rt.it.next()].getValue();
const dst = try rt.results[try rt.it.next()].getValue();
const lhs = try rt.results[try rt.it.next()].getValue();
const rhs = try rt.results[try rt.it.next()].getValue();
const size = sw: switch (target_type) {
.Vector => |v| continue :sw (try rt.results[v.components_type_word].getVariant()).Type,
.Vector4f32,
.Vector3f32,
.Vector2f32,
.Vector4i32,
.Vector3i32,
.Vector2i32,
.Vector4u32,
.Vector3u32,
.Vector2u32,
=> 32,
.Float => |f| if (T == .Float) f.bit_length else return RuntimeError.InvalidSpirV,
.Int => |i| if (T == .SInt or T == .UInt) i.bit_length else return RuntimeError.InvalidSpirV,
else => return RuntimeError.InvalidSpirV,
};
const lane_bits = try Result.resolveLaneBitWidth(target_type, rt);
const operator = struct {
fn operation(comptime TT: type, op1: TT, op2: TT) RuntimeError!TT {
@@ -637,68 +711,77 @@ fn MathEngine(comptime T: ValueType, comptime Op: MathOp) type {
};
}
fn process(bit_count: SpvWord, v: *Result.Value, op1_v: *const Result.Value, op2_v: *const Result.Value) RuntimeError!void {
fn applyScalar(bit_count: SpvWord, d: *Result.Value, l: *Result.Value, r: *Result.Value) RuntimeError!void {
switch (bit_count) {
inline 8, 16, 32, 64 => |i| {
if (i == 8 and T == .Float) { // No f8
return RuntimeError.InvalidSpirV;
}
(try getValuePrimitiveField(T, i, v)).* = try operation(
getValuePrimitiveFieldType(T, i),
(try getValuePrimitiveField(T, i, @constCast(op1_v))).*,
(try getValuePrimitiveField(T, i, @constCast(op2_v))).*,
);
inline 8, 16, 32, 64 => |bits| {
if (bits == 8 and T == .Float) return RuntimeError.InvalidSpirV;
const ScalarT = getValuePrimitiveFieldType(T, bits);
const d_field = try getValuePrimitiveField(T, bits, d);
const l_field = try getValuePrimitiveField(T, bits, l);
const r_field = try getValuePrimitiveField(T, bits, r);
d_field.* = try operation(ScalarT, l_field.*, r_field.*);
},
else => return RuntimeError.InvalidSpirV,
}
}
inline fn applyVectorTimesScalarF32(d: []Result.Value, l: []const Result.Value, r: f32) void {
for (d, l) |*d_v, l_v| {
d_v.Float.float32 = l_v.Float.float32 * r;
}
}
inline fn applySIMDVector(comptime ElemT: type, comptime N: usize, d: *@Vector(N, ElemT), l: *const @Vector(N, ElemT), r: *const @Vector(N, ElemT)) RuntimeError!void {
inline for (0..N) |i| {
d[i] = try operation(ElemT, l[i], r[i]);
}
}
inline fn applyVectorSIMDTimesScalarF32(comptime N: usize, d: *@Vector(N, f32), l: *const @Vector(N, f32), r: f32) void {
inline for (0..N) |i| {
d[i] = l[i] * r;
}
}
inline fn applySIMDVectorf32(comptime N: usize, d: *@Vector(N, f32), l: *const @Vector(N, f32), r: *const Result.Value) RuntimeError!void {
switch (Op) {
.VectorTimesScalar => applyVectorSIMDTimesScalarF32(N, d, l, r.Float.float32),
else => {
const rh: *const @Vector(N, f32) = switch (N) {
2 => &r.Vector2f32,
3 => &r.Vector3f32,
4 => &r.Vector4f32,
else => unreachable,
};
try applySIMDVector(f32, N, d, l, rh);
},
}
}
};
switch (value.*) {
.Float => if (T == .Float) try operator.process(size, value, op1_value, op2_value) else return RuntimeError.InvalidSpirV,
.Int => if (T == .SInt or T == .UInt) try operator.process(size, value, op1_value, op2_value) else return RuntimeError.InvalidSpirV,
.Vector => |vec| for (vec, op1_value.Vector, 0..) |*val, op1_v, i| {
switch (Op) {
.VectorTimesScalar => try operator.process(size, val, &op1_v, op2_value),
else => try operator.process(size, val, &op1_v, &op2_value.Vector[i]),
}
switch (dst.*) {
.Int, .Float => try operator.applyScalar(lane_bits, dst, lhs, rhs),
.Vector => |dst_vec| switch (Op) {
.VectorTimesScalar => operator.applyVectorTimesScalarF32(dst_vec, lhs.Vector, rhs.Float.float32),
else => for (dst_vec, lhs.Vector, rhs.Vector) |*d_lane, *l_lane, *r_lane| {
try operator.applyScalar(lane_bits, d_lane, l_lane, r_lane);
},
.Vector4f32 => |*vec| inline for (0..4) |i| {
switch (Op) {
.VectorTimesScalar => vec[i] = op1_value.Vector4f32[i] * op2_value.Float.float32,
else => vec[i] = try operator.operation(f32, op1_value.Vector4f32[i], op2_value.Vector4f32[i]),
}
},
.Vector3f32 => |*vec| inline for (0..3) |i| {
switch (Op) {
.VectorTimesScalar => vec[i] = op1_value.Vector3f32[i] * op2_value.Float.float32,
else => vec[i] = try operator.operation(f32, op1_value.Vector3f32[i], op2_value.Vector3f32[i]),
}
},
.Vector2f32 => |*vec| inline for (0..2) |i| {
switch (Op) {
.VectorTimesScalar => vec[i] = op1_value.Vector2f32[i] * op2_value.Float.float32,
else => vec[i] = try operator.operation(f32, op1_value.Vector2f32[i], op2_value.Vector2f32[i]),
}
},
.Vector4i32 => |*vec| inline for (0..4) |i| {
vec[i] = try operator.operation(i32, op1_value.Vector4i32[i], op2_value.Vector4i32[i]);
},
.Vector3i32 => |*vec| inline for (0..3) |i| {
vec[i] = try operator.operation(i32, op1_value.Vector3i32[i], op2_value.Vector3i32[i]);
},
.Vector2i32 => |*vec| inline for (0..2) |i| {
vec[i] = try operator.operation(i32, op1_value.Vector2i32[i], op2_value.Vector2i32[i]);
},
.Vector4u32 => |*vec| inline for (0..4) |i| {
vec[i] = try operator.operation(u32, op1_value.Vector4u32[i], op2_value.Vector4u32[i]);
},
.Vector3u32 => |*vec| inline for (0..3) |i| {
vec[i] = try operator.operation(u32, op1_value.Vector3u32[i], op2_value.Vector3u32[i]);
},
.Vector2u32 => |*vec| inline for (0..2) |i| {
vec[i] = try operator.operation(u32, op1_value.Vector2u32[i], op2_value.Vector2u32[i]);
},
.Vector4f32 => |*d| try operator.applySIMDVectorf32(4, d, &lhs.Vector4f32, rhs),
.Vector3f32 => |*d| try operator.applySIMDVectorf32(3, d, &lhs.Vector3f32, rhs),
.Vector2f32 => |*d| try operator.applySIMDVectorf32(2, d, &lhs.Vector2f32, rhs),
.Vector4i32 => |*d| try operator.applySIMDVector(i32, 4, d, &lhs.Vector4i32, &rhs.Vector4i32),
.Vector3i32 => |*d| try operator.applySIMDVector(i32, 3, d, &lhs.Vector3i32, &rhs.Vector3i32),
.Vector2i32 => |*d| try operator.applySIMDVector(i32, 2, d, &lhs.Vector2i32, &rhs.Vector2i32),
.Vector4u32 => |*d| try operator.applySIMDVector(u32, 4, d, &lhs.Vector4u32, &rhs.Vector4u32),
.Vector3u32 => |*d| try operator.applySIMDVector(u32, 3, d, &lhs.Vector3u32, &rhs.Vector3u32),
.Vector2u32 => |*d| try operator.applySIMDVector(u32, 2, d, &lhs.Vector2u32, &rhs.Vector2u32),
else => return RuntimeError.InvalidSpirV,
}
}
@@ -784,20 +867,21 @@ fn opBitcast(_: std.mem.Allocator, _: SpvWord, rt: *Runtime) RuntimeError!void {
}
fn copyValue(dst: *Result.Value, src: *const Result.Value) void {
if (src.getCompositeDataOrNull()) |src_slice| {
if (dst.getCompositeDataOrNull()) |dst_slice| {
switch (src.*) {
.Vector, .Matrix, .Array, .Structure => |src_slice| {
const dst_slice = switch (dst.*) {
.Vector, .Matrix, .Array, .Structure => |d| d,
else => unreachable,
};
for (0..@min(dst_slice.len, src_slice.len)) |i| {
copyValue(&dst_slice[i], &src_slice[i]);
}
} else {
unreachable;
}
} else {
dst.* = src.*;
},
else => dst.* = src.*,
}
}
fn getValuePrimitiveField(comptime T: ValueType, comptime BitCount: SpvWord, v: *Result.Value) RuntimeError!*getValuePrimitiveFieldType(T, BitCount) {
pub fn getValuePrimitiveField(comptime T: ValueType, comptime BitCount: SpvWord, v: *Result.Value) RuntimeError!*getValuePrimitiveFieldType(T, BitCount) {
return switch (T) {
.Bool => &v.Bool,
.Float => switch (BitCount) {
@@ -815,7 +899,7 @@ fn getValuePrimitiveField(comptime T: ValueType, comptime BitCount: SpvWord, v:
};
}
fn getValuePrimitiveFieldType(comptime T: ValueType, comptime BitCount: SpvWord) type {
pub fn getValuePrimitiveFieldType(comptime T: ValueType, comptime BitCount: SpvWord) type {
return switch (T) {
.Bool => bool,
.Float => std.meta.Float(BitCount),
@@ -1112,7 +1196,7 @@ fn opExtInst(allocator: std.mem.Allocator, word_count: SpvWord, rt: *Runtime) Ru
const set = try rt.it.next();
const inst = try rt.it.next();
switch (try rt.results[set].getVariant()) {
switch ((try rt.results[set].getVariant()).*) {
.Extension => |ext| if (ext.dispatcher[inst]) |pfn| {
try pfn(allocator, target_type, id, word_count, rt);
},
@@ -1122,10 +1206,11 @@ fn opExtInst(allocator: std.mem.Allocator, word_count: SpvWord, rt: *Runtime) Ru
fn opExtInstImport(allocator: std.mem.Allocator, word_count: SpvWord, rt: *Runtime) RuntimeError!void {
const id = try rt.it.next();
rt.mod.results[id].name = try readStringN(allocator, &rt.it, word_count - 1);
const name = try readStringN(allocator, &rt.it, word_count - 1);
rt.mod.results[id].name = name;
rt.mod.results[id].variant = .{
.Extension = .{
.dispatcher = undefined,
.dispatcher = if (extensions_map.get(name)) |map| map else return RuntimeError.UnsupportedExtension,
},
};
}