rework byval ZIR instructions; forbid runtime vector indexes by andrewrk · Pull Request #25154 · ziglang/zig

andrewrk · 2025-09-05T02:15:08Z

fixes #22906
fixes #13938
fixes #25111
partially addresses #24061 by making vector indexes comptime-known
partially addresses #5973

Upgrade Guide

for (0..vector_len) |i| {
   _ = vector[i];
}

⬇️

// coerce the vector to an array
const vector_type = @typeInfo(@TypeOf(vector)).vector;
const array: [vector_type.len]vector_type.child = vector;
for (&array) |elem| {
    _ = elem;
}

Performance Data Points

llvm backend ReleaseFast compiler building Debug Zig compiler, master vs branch

Benchmark 1 (3 runs): master/bin/zig build-exe ...
  measurement          mean ± σ            min … max           outliers         delta
  wall_time          12.3s  ± 16.2ms    12.3s  … 12.3s           0 ( 0%)        0%
  peak_rss           1.36GB ± 22.5MB    1.34GB … 1.39GB          0 ( 0%)        0%
  cpu_cycles          105G  ±  291M      105G  …  105G           0 ( 0%)        0%
  instructions        212G  ± 3.89M      212G  …  212G           0 ( 0%)        0%
  cache_references   6.67G  ± 19.6M     6.66G  … 6.69G           0 ( 0%)        0%
  cache_misses        466M  ± 5.03M      461M  …  471M           0 ( 0%)        0%
  branch_misses       404M  ± 1.57M      403M  …  406M           0 ( 0%)        0%
Benchmark 2 (3 runs): branch/bin/zig build-exe ...
  measurement          mean ± σ            min … max           outliers         delta
  wall_time          12.9s  ± 36.4ms    12.9s  … 13.0s           0 ( 0%)        💩+  5.1% ±  0.5%
  peak_rss           1.40GB ± 8.90MB    1.40GB … 1.41GB          0 ( 0%)          +  3.0% ±  2.8%
  cpu_cycles          109G  ±  170M      109G  …  109G           0 ( 0%)        💩+  4.2% ±  0.5%
  instructions        221G  ± 5.81M      221G  …  221G           0 ( 0%)        💩+  4.0% ±  0.0%
  cache_references   7.08G  ± 8.88M     7.07G  … 7.09G           0 ( 0%)        💩+  6.1% ±  0.5%
  cache_misses        486M  ± 3.12M      483M  …  489M           0 ( 0%)        💩+  4.4% ±  2.0%
  branch_misses       413M  ± 1.25M      412M  …  414M           0 ( 0%)        💩+  2.1% ±  0.8%

Zig Compiler ReleaseSmall Size

💩 13.6 -> 14.1M (+4%)

The test case from #13938

Debug ⚡ 29677949 -> 236675ns (-99%)
ReleaseFast ⚡ 28415587 -> 491ns (-100%)

llvm backend ReleaseFast building Debug Zig compiler, using llvm backend vs using x86 backend, with the branch

Benchmark 1 (3 runs): stage3/bin/zig build-exe -fllvm -flld ...
  measurement          mean ± σ            min … max           outliers         delta
  wall_time          60.0s  ± 63.6ms    59.9s  … 60.0s           0 ( 0%)        0%
  peak_rss           4.01GB ±  447KB    4.01GB … 4.01GB          0 ( 0%)        0%
  cpu_cycles          316G  ±  385M      315G  …  316G           0 ( 0%)        0%
  instructions        583G  ±  221M      583G  …  583G           0 ( 0%)        0%
  cache_references   22.4G  ± 49.5M     22.3G  … 22.4G           0 ( 0%)        0%
  cache_misses       4.54G  ± 9.15M     4.53G  … 4.55G           0 ( 0%)        0%
  branch_misses      2.27G  ± 1.79M     2.27G  … 2.27G           0 ( 0%)        0%
Benchmark 2 (3 runs): stage3/bin/zig build-exe -fno-llvm -fno-lld ...
  measurement          mean ± σ            min … max           outliers         delta
  wall_time          13.0s  ± 29.7ms    13.0s  … 13.1s           0 ( 0%)        ⚡- 78.3% ±  0.2%
  peak_rss           1.38GB ± 31.0MB    1.34GB … 1.41GB          0 ( 0%)        ⚡- 65.7% ±  1.2%
  cpu_cycles          111G  ±  136M      110G  …  111G           0 ( 0%)        ⚡- 65.0% ±  0.2%
  instructions        222G  ± 10.5M      222G  …  222G           0 ( 0%)        ⚡- 62.0% ±  0.1%
  cache_references   7.17G  ± 10.5M     7.16G  … 7.18G           0 ( 0%)        ⚡- 67.9% ±  0.4%
  cache_misses        482M  ± 3.52M      478M  …  485M           0 ( 0%)        ⚡- 89.4% ±  0.3%
  branch_misses       424M  ± 1.23M      423M  …  425M           0 ( 0%)        ⚡- 81.3% ±  0.2%

x86 backend Debug compiler building Debug Zig compiler, master vs branch

Benchmark 1 (3 runs): master/stage4/bin/zig build-exe ...
  measurement          mean ± σ            min … max           outliers         delta
  wall_time           423s  ± 4.75s      420s  …  428s           0 ( 0%)        0%
  peak_rss           3.59GB ±  166MB    3.50GB … 3.79GB          0 ( 0%)        0%
  cpu_cycles         3.52T  ±  106G     3.46T  … 3.65T           0 ( 0%)        0%
  instructions       2.67T  ± 84.4G     2.62T  … 2.77T           0 ( 0%)        0%
  cache_references    161G  ± 3.52G      159G  …  165G           0 ( 0%)        0%
  cache_misses       20.0G  ±  336M     19.8G  … 20.4G           0 ( 0%)        0%
  branch_misses      3.88G  ±  116M     3.81G  … 4.01G           0 ( 0%)        0%
Benchmark 2 (3 runs): branch/stage4/bin/zig build-exe ...
  measurement          mean ± σ            min … max           outliers         delta
  wall_time           469s  ± 3.71s      466s  …  473s           0 ( 0%)        💩+ 10.9% ±  2.3%
  peak_rss           3.56GB ± 36.3MB    3.54GB … 3.60GB          0 ( 0%)          -  1.0% ±  7.6%
  cpu_cycles         3.78T  ± 21.7G     3.76T  … 3.80T           0 ( 0%)        💩+  7.2% ±  4.9%
  instructions       2.78T  ± 17.3G     2.77T  … 2.80T           0 ( 0%)          +  4.1% ±  5.2%
  cache_references    170G  ±  602M      169G  …  170G           0 ( 0%)        💩+  5.2% ±  3.6%
  cache_misses       20.4G  ± 50.1M     20.3G  … 20.4G           0 ( 0%)          +  1.6% ±  2.7%
  branch_misses      4.01G  ± 24.3M     3.99G  … 4.04G           0 ( 0%)          +  3.4% ±  4.9%

llvm backend Debug vs x86_64 backend Debug, building a subset of the Debug Zig compiler from this branch

Benchmark 1 (3 runs): master-llvm/bin/zig build-exe ...
  measurement          mean ± σ            min … max           outliers         delta
  wall_time          19.6s  ± 85.8ms    19.5s  … 19.7s           0 ( 0%)        0%
  peak_rss            798MB ± 4.74MB     795MB …  803MB          0 ( 0%)        0%
  cpu_cycles          220G  ±  285M      219G  …  220G           0 ( 0%)        0%
  instructions        485G  ± 5.71M      485G  …  485G           0 ( 0%)        0%
  cache_references   20.2G  ± 12.0M     20.2G  … 20.2G           0 ( 0%)        0%
  cache_misses       1.05G  ± 8.68M     1.04G  … 1.06G           0 ( 0%)        0%
  branch_misses       416M  ±  843K      415M  …  416M           0 ( 0%)        0%
Benchmark 2 (3 runs): master-x86_64/bin/zig build-exe ...
  measurement          mean ± σ            min … max           outliers         delta
  wall_time          58.5s  ±  190ms    58.3s  … 58.7s           0 ( 0%)        💩+198.8% ±  1.7%
  peak_rss            897MB ± 10.6MB     886MB …  907MB          0 ( 0%)        💩+ 12.5% ±  2.3%
  cpu_cycles          645G  ±  921M      644G  …  646G           0 ( 0%)        💩+193.5% ±  0.7%
  instructions        559G  ± 3.89M      559G  …  559G           0 ( 0%)        💩+ 15.2% ±  0.0%
  cache_references   59.8G  ± 28.4M     59.8G  … 59.8G           0 ( 0%)        💩+196.1% ±  0.2%
  cache_misses       3.63G  ± 36.6M     3.60G  … 3.67G           0 ( 0%)        💩+245.3% ±  5.7%
  branch_misses       665M  ± 5.26M      659M  …  669M           0 ( 0%)        💩+ 60.1% ±  2.1%

Benchmark 1 (3 runs): branch-llvm/llvm/bin/zig build-exe ...
  measurement          mean ± σ            min … max           outliers         delta
  wall_time          28.2s  ± 1.17s     27.5s  … 29.6s           0 ( 0%)        0%
  peak_rss            960MB ±  199MB     841MB … 1.19GB          0 ( 0%)        0%
  cpu_cycles          314G  ± 36.3G      293G  …  356G           0 ( 0%)        0%
  instructions        611G  ± 74.2G      568G  …  696G           0 ( 0%)        0%
  cache_references   26.9G  ± 2.43G     25.5G  … 29.8G           0 ( 0%)        0%
  cache_misses       1.60G  ±  127M     1.52G  … 1.75G           0 ( 0%)        0%
  branch_misses       518M  ±  103M      458M  …  637M           0 ( 0%)        0%
Benchmark 2 (3 runs): branch-x86_64/debug/bin/zig build-exe ...
  measurement          mean ± σ            min … max           outliers         delta
  wall_time          70.9s  ± 2.95s     69.0s  … 74.3s           0 ( 0%)        💩+151.1% ± 18.0%
  peak_rss            992MB ±  210MB     866MB … 1.23GB          0 ( 0%)          +  3.3% ± 48.3%
  cpu_cycles          775G  ± 91.2G      721G  …  880G           0 ( 0%)        💩+146.6% ± 50.1%
  instructions        655G  ± 94.2G      601G  …  764G           0 ( 0%)          +  7.3% ± 31.5%
  cache_references   69.1G  ± 5.49G     65.6G  … 75.4G           0 ( 0%)        💩+156.4% ± 35.7%
  cache_misses       3.85G  ±  195M     3.74G  … 4.08G           0 ( 0%)        💩+141.0% ± 23.4%
  branch_misses       991M  ±  144M      903M  … 1.16G           0 ( 0%)        💩+ 91.4% ± 55.0%

andrewrk · 2025-09-05T22:20:07Z

How about on this branch @jacobly0 can you repro the x86 backend failures the CI is running into?

jacobly0 · 2025-09-05T23:16:57Z

--- a/src/arch/x86_64/CodeGen.zig
+++ b/src/arch/x86_64/CodeGen.zig
@@ -92414,7 +92414,7 @@ fn genBody(cg: *CodeGen, body: []const Air.Inst.Index) InnerError!void {
                     .clobbers = .{ .eflags = true },
                     .each = .{ .once = &.{
                         .{ ._, ._, .mov, .tmp0p, .sa(.src0, .sub_unaligned_size), ._, ._ },
-                        .{ .@"0:", .vp_d, .movzxb, .tmp1y, .memia(.src0x, .tmp0, .add_unaligned_size), ._, ._ },
+                        .{ .@"0:", .vp_d, .movzxb, .tmp1y, .memia(.src0q, .tmp0, .add_unaligned_size), ._, ._ },
                         .{ ._, .v_dqa, .mov, .memsia(.dst0y, .@"4", .tmp0, .add_unaligned_size), .tmp1y, ._, ._ },
                         .{ ._, ._, .add, .tmp0p, .si(8), ._, ._ },
                         .{ ._, ._nc, .j, .@"0b", ._, ._, ._ },

andrewrk · 2025-09-06T05:16:05Z

lots of this triggered by test-cases:

/home/andy/dev/zig/lib/std/Io/Writer.zig:342:5: error: unimplemented ptr_slice_len_ptr
pub fn defaultRebase(w: *Writer, preserve: usize, minimum_len: usize) Error!void {
~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

from src/codegen/aarch64/Select.zig line 964

mlugg

See comments.

This change also regresses unused variable errors in some cases, such as:

var x: [1]u8 = .{0};
_ = x[0];

and:

const x: S = .{ .y = 0 };
_ = x.y;

I definitely mentioned this to you months ago, but I can forgive you forgetting that 😛
This requires some bigger AstGen changes to fix; I recommend opening a follow-up issue after merge and I'll get to it when I can.

src/Sema.zig

src/Value.zig

test/cases/compile_errors/comptime_var_referenced_by_type.zig

test/cases/compile_errors/enum_field_value_references_enum.zig

test/cases/compile_errors/packed_union_fields_mismatch.zig

andrewrk · 2025-09-06T23:46:22Z

Blocking on #25172, which also affects master branch.

andrewrk · 2025-09-08T06:05:45Z

Some aarch64 backend failures in the x86_64-linux-debug-llvm run

jacobly0 · 2025-09-08T06:14:58Z

Some aarch64 backend failures in the x86_64-linux-debug-llvm run

Those are all ptr_add/ptr_sub of pointers to zero-bit types which nobody has convinced me should be valid Air (although ironically the aarch64 backend design is the best equipped to handle that without harming codegen quality, but that isn't really a good enough reason to assume it should be allowed).

andrewrk · 2025-09-08T06:22:28Z

Understood; will remedy that as a prereq to making progress here then.

jacobly0 · 2025-09-08T06:29:11Z

I guess an argument in favor of not allowing them is that the index operand is dead in a way that is not reflected in Air.Liveness, which affects the output size of the llvm and c backends which are not likely to be rewritten to not use Air.Liveness anytime soon.

andrewrk · 2025-09-09T00:49:14Z

Have a reduction handy? It's not one of the examples from #22419 and from looking at Sema.zig I can see only these code paths emit those instructions:

arithmetic, but that's protected by error: pointer arithmetic requires element type 'void' to have runtime bits
array concatenation
memcpy safety
slice syntax

jacobly0 · 2025-09-09T01:04:09Z

idk just

--- a/src/Sema.zig
+++ b/src/Sema.zig
@@ -809,6 +809,10 @@ pub const Block = struct {
     }
 
     pub fn addInst(block: *Block, inst: Air.Inst) error{OutOfMemory}!Air.Inst.Ref {
+        switch (inst.tag) {
+            else => {},
+            .ptr_add, .ptr_sub => assert(block.sema.typeOf(@enumFromInt(block.sema.air_extra.items[inst.data.ty_pl.payload])).childType(block.sema.pt.zcu).hasRuntimeBitsSema(block.sema.pt) catch unreachable),
+        }
         return (try block.addInstAsIndex(inst)).toRef();
     }

and run the behavior tests, looks like it is just calls to analyzePtrArithmetic that are not from analyzeArithmetic which is what checks for that error.

andrewrk · 2025-09-09T01:46:41Z

Good idea. Yeah it was a pretty trivial fix actually. Seems to be just one missing check inside analyzePtrArithmetic.

andrewrk · 2025-09-09T01:49:08Z

Disabling spirv64-vulkan coverage due to no active maintainer:

test
+- test-modules
   +- test-behavior
      +- compile test Debug spirv64-vulkan 29 errors
/home/ci/actions-runner4/_work/zig/zig/lib/std/testing.zig:362:38: error: cannot access element of logical pointer '[]const u64'
            if (!std.meta.eql(actual[index], expected[index])) break :diff_index index;
                                     ^~~~~
referenced by:
    expectEqualInner__anon_4105: /home/ci/actions-runner4/_work/zig/zig/lib/std/testing.zig:140:34
    expectEqual [inlined]: /home/ci/actions-runner4/_work/zig/zig/lib/std/testing.zig:75:28
    test.peer type resolution: array with smaller child type and vector with larger child type: /home/ci/actions-runner4/_work/zig/zig/test/behavior/cast.zig:1779:20
/home/ci/actions-runner4/_work/zig/zig/lib/std/testing.zig:362:38: error: cannot access element of logical pointer '[]const i32'
            if (!std.meta.eql(actual[index], expected[index])) break :diff_index index;
                                     ^~~~~
/home/ci/actions-runner4/_work/zig/zig/lib/std/testing.zig:362:38: error: cannot access element of logical pointer '[]const f32'
            if (!std.meta.eql(actual[index], expected[index])) break :diff_index index;
                                     ^~~~~
/home/ci/actions-runner4/_work/zig/zig/lib/std/testing.zig:362:38: error: cannot access element of logical pointer '[]const u32'
            if (!std.meta.eql(actual[index], expected[index])) break :diff_index index;
                                     ^~~~~
/home/ci/actions-runner4/_work/zig/zig/lib/std/testing.zig:362:38: error: cannot access element of logical pointer '[]const u1'
            if (!std.meta.eql(actual[index], expected[index])) break :diff_index index;
                                     ^~~~~
/home/ci/actions-runner4/_work/zig/zig/lib/std/testing.zig:362:38: error: cannot access element of logical pointer '[]const u8'
            if (!std.meta.eql(actual[index], expected[index])) break :diff_index index;
                                     ^~~~~
/home/ci/actions-runner4/_work/zig/zig/lib/std/testing.zig:362:38: error: cannot access element of logical pointer '[]const bool'
            if (!std.meta.eql(actual[index], expected[index])) break :diff_index index;
                                     ^~~~~
/home/ci/actions-runner4/_work/zig/zig/test/behavior/cast.zig:462:19: error: value with non-mergable pointer type '*const ?behavior.cast.A' depends on runtime control flow
    try expect((b catch unreachable).?.a == 1);
                ~~^~~~~~~~~~~~~~~~~
/home/ci/actions-runner4/_work/zig/zig/test/behavior/cast.zig:462:19: note: runtime control flow here
/home/ci/actions-runner4/_work/zig/zig/test/behavior/cast.zig:462:19: note: pointers with address space 'generic' cannot be returned from a branch on target spirv-vulkan by compiler backend stage2_spirv
/home/ci/actions-runner4/_work/zig/zig/test/behavior/cast.zig:473:35: error: value with non-mergable pointer type '*const []u8' depends on runtime control flow
    try expect((gimmeErrOrSlice() catch unreachable).len == 0);
                ~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~
/home/ci/actions-runner4/_work/zig/zig/test/behavior/cast.zig:473:35: note: runtime control flow here
/home/ci/actions-runner4/_work/zig/zig/test/behavior/cast.zig:473:35: note: pointers with address space 'generic' cannot be returned from a branch on target spirv-vulkan by compiler backend stage2_spirv
/home/ci/actions-runner4/_work/zig/zig/test/behavior/cast.zig:1270:19: error: value with non-mergable pointer type '*[]const u8' depends on runtime control flow
    try expect((x catch unreachable).len == 0);
                ~~^~~~~~~~~~~~~~~~~
/home/ci/actions-runner4/_work/zig/zig/test/behavior/cast.zig:1270:19: note: runtime control flow here
/home/ci/actions-runner4/_work/zig/zig/test/behavior/cast.zig:1270:19: note: pointers with address space 'generic' cannot be returned from a branch on target spirv-vulkan by compiler backend stage2_spirv
/home/ci/actions-runner4/_work/zig/zig/test/behavior/array.zig:44:1: error: TODO (SPIR-V): implement AIR tag ptr_slice_len_ptr
fn getArrayLen(a: []const u32) usize {
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/home/ci/actions-runner4/_work/zig/zig/test/behavior/array.zig:260:1: error: TODO (SPIR-V): implement AIR tag ptr_slice_len_ptr
test "implicit cast zero sized array ptr to slice" {
^~~~
/home/ci/actions-runner4/_work/zig/zig/test/behavior/bitcast.zig:177:9: error: TODO (SPIR-V): implement field access for packed structs
        fn doTheTest() !void {
        ^~~~~~~~~~~~~~~~~~~~
/home/ci/actions-runner4/_work/zig/zig/test/behavior/bitcast.zig:233:9: error: TODO (SPIR-V): implement field access for packed structs
        fn doTheTest() !void {
        ^~~~~~~~~~~~~~~~~~~~
/home/ci/actions-runner4/_work/zig/zig/test/behavior/cast.zig:1444:1: error: TODO (SPIR-V): implement AIR tag ptr_slice_len_ptr
test "coerce undefined single-item pointer of array to error union of slice" {
^~~~
/home/ci/actions-runner4/_work/zig/zig/test/behavior/cast.zig:1454:1: error: TODO (SPIR-V): implement AIR tag ptr_slice_len_ptr
test "pointer to empty struct literal to mutable slice" {
^~~~
/home/ci/actions-runner4/_work/zig/zig/test/behavior/fn.zig:406:1: error: TODO (SPIR-V): implement AIR tag ptr_slice_len_ptr
test "import passed byref to function in return type" {
^~~~
/home/ci/actions-runner4/_work/zig/zig/test/behavior/fn.zig:450:9: error: TODO (SPIR-V): implement AIR tag try_ptr
        fn errUnion(s: anyerror!@This()) !void {
        ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/home/ci/actions-runner4/_work/zig/zig/test/behavior/packed-struct.zig:251:1: error: cannot perform pointer cast: '*behavior.packed-struct.test.nested packed struct unaligned.S2' to '*align(1:0:2) u4'
test "nested packed struct unaligned" {
^~~~
/home/ci/actions-runner4/_work/zig/zig/test/behavior/packed-struct.zig:898:9: error: TODO (SPIR-V): implement field access for packed structs
        fn foo(p: Packed, a1: u64, a2: u64, a3: u64, a4: u64, a5: u64) callconv(.c) bool {
        ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/home/ci/actions-runner4/_work/zig/zig/test/behavior/packed-struct.zig:970:1: error: TODO (SPIR-V): implement field access for packed structs
test "store undefined to packed result location" {
^~~~
/home/ci/actions-runner4/_work/zig/zig/test/behavior/slice.zig:75:1: error: TODO (SPIR-V): implement AIR tag ptr_slice_len_ptr
fn assertLenIsZero(msg: []const u8) !void {
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/home/ci/actions-runner4/_work/zig/zig/test/behavior/slice.zig:265:1: error: TODO (SPIR-V): implement AIR tag ptr_slice_len_ptr
test "result location zero sized array inside struct field implicit cast to slice" {
^~~~
/home/ci/actions-runner4/_work/zig/zig/test/behavior/struct.zig:1009:9: error: TODO (SPIR-V): implement field access for packed structs
        fn doTheTest() !void {
        ^~~~~~~~~~~~~~~~~~~~
/home/ci/actions-runner4/_work/zig/zig/lib/std/testing.zig:78:1: error: TODO (SPIR-V): implement field access for packed structs
fn expectEqualInner(comptime T: type, expected: T, actual: T) !void {
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/home/ci/actions-runner4/_work/zig/zig/test/behavior/union.zig:1348:9: error: TODO (SPIR-V): implement field access for packed structs
        fn unpack(self: @This()) usize {
        ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/home/ci/actions-runner4/_work/zig/zig/test/behavior/union.zig:1904:1: error: TODO (SPIR-V): implement field access for packed structs
test "packed union initialized via reintepreted struct field initializer" {
^~~~
/home/ci/actions-runner4/_work/zig/zig/test/behavior/union.zig:1947:1: error: TODO (SPIR-V): implement field access for packed structs
test "store of comptime reinterpreted memory to packed union" {
^~~~
/home/ci/actions-runner4/_work/zig/zig/test/behavior/vector.zig:1266:1: error: TODO (SPIR-V): implement AIR tag mul_sat
test "zero multiplicand" {
^~~~
error: the following command failed with 29 compilation errors:
/home/ci/actions-runner4/_work/zig/zig/build-debug/stage3-debug/bin/zig test -fno-llvm -fno-lld -ODebug -target spirv64-vulkan -mcpu vulkan_v1_2+float16+float64 -Mroot=/home/ci/actions-runner4/_work/zig/zig/test/behavior.zig --cache-dir /home/ci/actions-runner4/_work/zig/zig/zig-local-cache --global-cache-dir /home/ci/actions-runner4/_work/zig/zig/zig-global-cache --name test -L /home/ci/deps/zig+llvm+lld+clang-x86_64-linux-musl-0.16.0-dev.104+689461e31/lib -I /home/ci/deps/zig+llvm+lld+clang-x86_64-linux-musl-0.16.0-dev.104+689461e31/include --zig-lib-dir /home/ci/actions-runner4/_work/zig/zig/lib --listen=-

As soon as someone steps up and fixes it, coverage can be reinstated.

andrewrk · 2025-09-09T02:03:26Z

Disabling wasm backend coverage due to no active maintainer:

test
+- test-modules
   +- test-behavior
      +- run test behavior-wasm32-wasi-none-lime1-Debug-selfhosted-no-lld failure
Error: failed to run main module `/home/ci/actions-runner7/_work/zig/zig/zig-local-cache/o/2b0ea36eeaad87895fcb86b02fb65fc2/test.wasm`

Caused by:
    0: failed to instantiate "/home/ci/actions-runner7/_work/zig/zig/zig-local-cache/o/2b0ea36eeaad87895fcb86b02fb65fc2/test.wasm"
    1: error while executing at wasm backtrace:
           0: 0x14a659 - <unknown>!behavior.union.test.signed enum tag with negative value
           1: 0x5fd2 - <unknown>!test_runner.mainServer
           2: 0x6599 - <unknown>!test_runner.main
           3: 0x29cd - <unknown>!start.wasi_start
    2: wasm trap: wasm `unreachable` instruction executed
error: while executing test 'behavior.union.test.signed enum tag with negative value', the following command exited with code 134 (expected exited with code 0):
wasmtime --dir=. -- /home/ci/actions-runner7/_work/zig/zig/zig-local-cache/o/2b0ea36eeaad87895fcb86b02fb65fc2/test.wasm --cache-dir=/home/ci/actions-runner7/_work/zig/zig/zig-local-cache --seed=0x904666fb --listen=-

As soon as someone steps up and fixes it, coverage can be reinstated. Meanwhile, I'm personally prioritizing stabilizing the language over completing the wasm backend.

jacobly0 · 2025-09-09T21:44:44Z

current partial progress:

--- a/src/arch/x86_64/CodeGen.zig
+++ b/src/arch/x86_64/CodeGen.zig
@@ -186938,7 +186938,7 @@ const Temp = struct {
                 },
                 .struct_type => {
                     assert(src_regs.len - part_index == std.math.divCeil(u32, src_abi_size, 8) catch unreachable);
-                    break :part_ty .u64;
+                    break :part_ty if (src_abi_size <= 4) .u32 else .u64;
                 },
                 .tuple_type => |tuple_type| {
                     assert(tuple_type.types.len == src_regs.len);
@@ -186947,10 +186947,6 @@ const Temp = struct {
             };
             const part_size: u31 = @intCast(part_ty.abiSize(zcu));
             const src_rc = src_reg.class();
-            const part_bit_size = switch (src_rc) {
-                else => 8 * part_size,
-                .x87 => part_ty.bitSize(zcu),
-            };
             if (src_rc == .x87 or std.math.isPowerOfTwo(part_size)) {
                 // hack around linker relocation bugs
                 switch (ptr.tracking(cg).short) {
@@ -186959,7 +186955,10 @@ const Temp = struct {
                 }
                 const strat = try cg.moveStrategy(part_ty, src_rc, false);
                 try strat.write(cg, try ptr.tracking(cg).short.deref().mem(cg, .{
-                    .size = .fromBitSize(part_bit_size),
+                    .size = switch (src_rc) {
+                        else => .fromBitSize(8 * part_size),
+                        .x87 => .tbyte,
+                    },
                     .disp = part_disp,
                 }), registerAlias(src_reg, part_size));
             } else {

Edit: Oh, it works fine, I was just testing wrong.

It wasn't checking bit pointer data.

If the `nav_ty` is resolved by the `nav_val`, then we need to also mark the `nav_ty` as in progress when we begin resolving the `nav_val`.

This reverts commit ba4d460.

The latest bugfix reverted this case to its old behavior (which is a reasonable behavior to have).

it was calculating host integer size in a wrong way. just use integer abi size

This reverts commit dedccec.

tracked by #24061 - these should be re-enabled once that is solved.

...of array passed as arg closes #22906

…in GpuGrid.Manager * See this issue for details ziglang/zig#25154

andrewrk mentioned this pull request Sep 5, 2025

remove elem_val and field_val ZIR instructions; forbid runtime vector indexes #25138

Closed

3 tasks

andrewrk force-pushed the no-decl-val-3 branch from df99660 to eca0bed Compare September 5, 2025 02:39

andrewrk added breaking Implementing this issue could cause existing code to no longer compile or have different behavior. release notes This PR should be mentioned in the release notes. labels Sep 5, 2025

andrewrk force-pushed the no-decl-val-3 branch from cf84728 to 434b9cb Compare September 5, 2025 22:36

andrewrk force-pushed the no-decl-val-3 branch 2 times, most recently from b7e95f3 to d8dfa7b Compare September 6, 2025 02:59

andrewrk force-pushed the no-decl-val-3 branch 2 times, most recently from e0bdebd to 6741734 Compare September 6, 2025 06:07

mlugg requested changes Sep 6, 2025

View reviewed changes

andrewrk force-pushed the no-decl-val-3 branch 3 times, most recently from 33cc67d to f3a3dc6 Compare September 8, 2025 03:24

andrewrk force-pushed the no-decl-val-3 branch from f3a3dc6 to 94129f5 Compare September 9, 2025 02:04

andrewrk force-pushed the no-decl-val-3 branch from 94129f5 to bd8cccb Compare September 10, 2025 07:08

jacobly0 force-pushed the no-decl-val-3 branch from 9794951 to ba4d460 Compare September 10, 2025 07:18

andrewrk and others added 16 commits September 20, 2025 18:33

Sema: add missed logic to checkPtrAttributes

725dec6

It wasn't checking bit pointer data.

langref: don't assume too much about pointer to packed struct field

d7d5049

Zcu: fix analysis of type of decl with inferred type

2f9d8d5

If the `nav_ty` is resolved by the `nav_val`, then we need to also mark the `nav_ty` as in progress when we begin resolving the `nav_val`.

Revert "delete failing test case"

fcc7e37

This reverts commit ba4d460.

cases: update for new error

baaf715

The latest bugfix reverted this case to its old behavior (which is a reasonable behavior to have).

frontend: another packedStructFieldPtrInfo fix

5d8c1fb

it was calculating host integer size in a wrong way. just use integer abi size

disable failing stage2_aarch64 behavior tests

2ba03e9

x86_64: rewrite vector element pointer access

60cdaca

aarch64: fix behavior failures

5144f10

standalone: fix misaligned stack crash

f81a721

Revert "frontend: another packedStructFieldPtrInfo fix"

20925f2

This reverts commit dedccec.

fix rebase error

0e16d93

allow some test cases to regress

ed7ff0b

tracked by #24061 - these should be re-enabled once that is solved.

add behavior test: resist alias of explicit copy...

500afbf

...of array passed as arg closes #22906

add behavior test: avoid unused field function body compile error

1c9ac9d

add behavior test: comptime C pointer to optional pointer

633162e

andrewrk force-pushed the no-decl-val-3 branch from 4dd288f to 633162e Compare September 21, 2025 02:00

andrewrk enabled auto-merge September 21, 2025 02:02

andrewrk merged commit 010d9a6 into master Sep 21, 2025
12 of 14 checks passed

andrewrk deleted the no-decl-val-3 branch September 21, 2025 08:49

mlugg mentioned this pull request Sep 22, 2025

Proposal: Load lhs after rhs is executed for assignment operators like += #25315

Open

linusg mentioned this pull request Sep 23, 2025

frontend: vectors and arrays no longer support in-memory coercion #25186

Merged

MatthiasPortzel mentioned this pull request Oct 5, 2025

Volatile not respected in some edge cases #21033

Open

squeek502 mentioned this pull request Oct 11, 2025

local variable is never mutated regression for field and array accesses #25540

Closed

xdBronch mentioned this pull request Oct 14, 2025

Debug mode - Referencing field copies the object (self hosted backend) #25184

Open

Daggerfall-is-the-best-TES-game mentioned this pull request Oct 15, 2025

Update Zig and use the x86_64 backend and incremental compilation PixelGuys/Cubyz#1583

Closed

7 tasks

noxabellus added a commit to noxabellus/zgpu that referenced this pull request Oct 29, 2025

migrate to zig master to avoid compiler bug and stop stack overflows …

70b54ec

…in GpuGrid.Manager * See this issue for details ziglang/zig#25154

Uh oh!

Comments

Conversation

andrewrk commented Sep 5, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Upgrade Guide

Performance Data Points

llvm backend ReleaseFast compiler building Debug Zig compiler, master vs branch

Zig Compiler ReleaseSmall Size

The test case from #13938

llvm backend ReleaseFast building Debug Zig compiler, using llvm backend vs using x86 backend, with the branch

x86 backend Debug compiler building Debug Zig compiler, master vs branch

llvm backend Debug vs x86_64 backend Debug, building a subset of the Debug Zig compiler from this branch

Uh oh!

andrewrk commented Sep 5, 2025

Uh oh!

jacobly0 commented Sep 5, 2025

Uh oh!

andrewrk commented Sep 6, 2025

Uh oh!

mlugg left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

andrewrk commented Sep 6, 2025

Uh oh!

andrewrk commented Sep 8, 2025

Uh oh!

jacobly0 commented Sep 8, 2025

Uh oh!

andrewrk commented Sep 8, 2025

Uh oh!

jacobly0 commented Sep 8, 2025

Uh oh!

andrewrk commented Sep 9, 2025

Uh oh!

jacobly0 commented Sep 9, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

andrewrk commented Sep 9, 2025

Uh oh!

andrewrk commented Sep 9, 2025

Uh oh!

andrewrk commented Sep 9, 2025

Uh oh!

jacobly0 commented Sep 9, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

andrewrk commented Sep 5, 2025 •

edited

Loading

mlugg left a comment •

edited

Loading

jacobly0 commented Sep 9, 2025 •

edited

Loading

jacobly0 commented Sep 9, 2025 •

edited

Loading