I'm writing a small numeric library whose kernels run on the CPU, on CUDA and on Vulkan through the self-hosted SPIR-V backend, and I was making them generic over the float type. `f64` needs its own `exp` and `log` on the GPU, so I wrote them as polynomial series, and the `f64` Vulkan module stopped validating. This fixes the three backend bugs underneath that, and a crash I ran into after merging master.
The series take floats apart with `@bitCast` on vectors. Between vectors of the same length this reaches the Vulkan/OpenGL fallback in `bitCast`, which stores the source into a function variable of the destination type:
```zig
export fn main() callconv(.{ .spirv_kernel = .{ .x = 1, .y = 1, .z = 1 } }) void {
var x: @Vector(2, f32) = .{ 1, 2 };
_ = &x;
const bits: @Vector(2, i32) = @bitCast(x);
_ = bits;
}
```
```
$ zig build-exe -fno-llvm -target spirv64-vulkan -mcpu vulkan_v1_2 bitcast.zig -femit-bin=bitcast.spv
$ spirv-val --target-env vulkan1.2 bitcast.spv
error: line 45: OpStore Pointer <id> '36[%36]'s type does not match Object <id> '29[%29]'s type.
```
An elementwise vector bitcast is now a scalar `bitCast` per lane, the way `buildConvert` and friends already unroll vectors. Going through the scalar path also normalizes lanes of strange integers: a single `OpBitcast` on the whole vector validates for `@Vector(n, u3)` to `@Vector(n, i3)`, but leaves lanes that don't compare equal to the negative values they should hold.
The series also convert between integer and float lanes. A vector of one lane, or of more than four without `vector16`, is an array in SPIR-V, and `airFloatFromInt` and `airIntFromFloat` emitted a single `OpConvert*` with that array as its result type:
```zig
export fn main() callconv(.{ .spirv_kernel = .{ .x = 1, .y = 1, .z = 1 } }) void {
var x: @Vector(1, i32) = .{1};
_ = &x;
const y: @Vector(1, f32) = @floatFromInt(x);
_ = y;
}
```
```
$ spirv-val --target-env vulkan1.2 convert.spv
error: line 40: Expected float scalar or vector type as Result Type: ConvertSToF
%y = OpConvertSToF %_arr_f32_u32_1 %26
```
`@floatCast` and `@intCast` already go through `buildConvert`, which unrolls the vector; these two now do as well, with `buildConvert` picking the int/float conversion opcodes.
The coefficients sit in a comptime array that the series take as a comptime slice, and that alone fails to compile, although nothing reads the slice at runtime:
```zig
const weights = [_]f32{ 1, 2, 3 };
inline fn first(comptime w: []const f32) f32 {
return w[0];
}
export fn main() callconv(.{ .spirv_kernel = .{ .x = 1, .y = 1, .z = 1 } }) void {
var x: f32 = first(&weights);
_ = &x;
}
```
```
$ zig build-exe -fno-llvm -target spirv64-vulkan -mcpu vulkan_v1_2 dbgvar.zig -femit-bin=dbgvar.spv
dbgvar.zig:7:8: error: TODO (SPIR-V): pointer to constant 'dbgvar.weights'
```
It appears `airDbgVar` lowers the operand only to attach an `OpName` to it, and a pointer to a non-extern constant is still a TODO. A comptime-known value has no runtime variable of its own to name, so `airDbgVar` now leaves interned operands alone. The same thing held back 58 of the behavior tests skipped on SPIR-V, and the test commit un-skips them.
After merging master, four of those tests crashed the compiler in `structFieldPtr`, which has `unreachable` for packed containers since ebb8683ef1 ([#36996](https://codeberg.org/ziglang/zig/pulls/36996)):
```zig
const Flags = packed struct { a: bool, b: bool };
export fn main() callconv(.{ .spirv_kernel = .{ .x = 1, .y = 1, .z = 1 } }) void {
var flags: Flags = .{ .a = true, .b = false };
const b = &flags.b;
b.* = true;
}
```
```
thread 3961624 panic: reached unreachable code
src/codegen/spirv/CodeGen.zig:6906:27: 0x4767157 in structFieldPtr (main.zig)
.@"packed" => unreachable,
```
My reading is that the `unreachable` holds for the pointer derivations it was written for, since `constantPtr` now turns packed pointers into pointers to their host integer first, but `airStructFieldPtr` and `airStructFieldPtrIndex` still pass packed structs through. A pointer to a packed field is a bit-pointer into the struct's own host integer, so it is the struct pointer: that is what the old code computed, and what the C backend now does with `.begin`. Packed unions get back the TODO error they had before. I may be missing a plan for packed field pointers here, and I'm happy to change it if so. The commit also un-skips three packed tests that crash on master the same way. I suppose nothing caught this because those tests are skipped on SPIR-V and the `spirv64-vulkan` behavior target is commented out in `test/tests.zig`.
I ran the behavior tests through `tools/unskip_spirv_tests.zig` and [zig-spirv-test-executor](https://github.com/Snektron/zig-spirv-test-executor) on a Radeon 680M, and the new and un-skipped tests on an RTX 4060 as well. Every fix has tests that fail on master and pass with it; the vector ones compare lane by lane, since `expectEqual` on vectors doesn't compile for SPIR-V yet. Running each behavior file whole, nothing that passes on master fails. Five more skips already pass on master without these changes, so I left them out, but I can add them. The new vector tests copy the aarch64, arm, sparc64 and riscv64 skips of the tests around them, which I couldn't check.
Reviewed-on: https://codeberg.org/ziglang/zig/pulls/37040
Reviewed-by: Ali Cheraghi <alichraghi@noreply.codeberg.org>
* `tools/generate_c_size_and_align_checks.zig`: superceded by any cbe
output artifact
* `cTypePreferredAlignment`: no uses, if brought back, cbe should be
updated to verify `__alignof`
Closes#19991Closes#20881Closes#35523
and remove "release" from the enum tag names.
no functional change, however, despite the addition of
backwards-compatibile declarations in this patch, it is breaking because
expressions that use `==` or `!=` operators will not able to use the
deprecated names.
I predict these names to survive until the 1.0 tag without any more
breakage.
composite integers (wider than usize) are represented in as `[N]u32` arrays.
however, no operations were implemented before.
following operations are implemented:
- bitwise (and, or, xor, not)
- comparison (eq/neq)
- add/sub/mul/shl/shr
- `@abs`, `@intCast`
this removes 36 SPIR-V behavior test skips.
also fixed `derivePtr` to use `OpAccessChain` instead of `OpBitcast`
for array pointer casts (e.g. `*[100]u8` to `*u8`) in logical addressing
mode, where pointer bitcasts are not valid.
Our implementation did the classic add-sub rounding trick `(y = x +/- C =+ C - x)`
with `C = 1 / eps(T) = 2^(mantissa - 1)`. This approach only works for values whose
magnitude is below the rounding capacity of the constant. For a 64-bit mantissa
(like f80 has), `C = 2^63` only rounds for `|x| < 2^63`. Before we allowed this to
be ran on `e < bias + 64` aka `|x| < 2^64`. And because it isn't large enough,
we lose a bit to rounding.
For reference, the musl implementation does the same thing, using `mantissa - 1`:
https://git.musl-libc.org/cgit/musl/tree/src/math/ceill.c#n18
where `LDBL_MANT_DIG` is 64 for `long double` on x86.
This commit also combines the floor and ceil implementations into one generic one.
This commit reworks how Sema handles arithmetic on comptime-known
values, fixing many bugs in the process.
The general pattern is that arithmetic on comptime-known values is now
handled by the new namespace `Sema.arith`. Functions handling comptime
arithmetic no longer live on `Value`; this is because some of them can
emit compile errors, so some *can't* go on `Value`. Only semantic
analysis should really be doing arithmetic on `Value`s anyway, so it
makes sense for it to integrate more tightly with `Sema`.
This commit also implements more coherent rules surrounding how
`undefined` interacts with comptime and mixed-comptime-runtime
arithmetic. The rules are as follows.
* If an operation cannot trigger Illegal Behavior, and any operand is
`undefined`, the result is `undefined`. This includes operations like
`0 *| undef`, where the LHS logically *could* be used to determine a
defined result. This is partly to simplify the language, but mostly to
permit codegen backends to represent `undefined` values as completely
invalid states.
* If an operation *can* trigger Illegal Behvaior, and any operand is
`undefined`, then Illegal Behavior results. This occurs even if the
operand in question isn't the one that "decides" illegal behavior; for
instance, `undef / 1` is undefined. This is for the same reasons as
described above.
* An operation which would trigger Illegal Behavior, when evaluated at
comptime, instead triggers a compile error. Additionally, if one
operand is comptime-known undef, such that the other (runtime-known)
operand isn't needed to determine that Illegal Behavior would occur,
the compile error is triggered.
* The only situation in which an operation with one comptime-known
operand has a comptime-known result is if that operand is undefined,
in which case the result is either undefined or a compile error per
the above rules. This could potentially be loosened in future (for
instance, `0 * rt` could be comptime-known 0 with a runtime assertion
that `rt` is not undefined), but at least for now, defining it more
conservatively simplifies the language and allows us to easily change
this in future if desired.
This commit fixes many bugs regarding the handling of `undefined`,
particularly in vectors. Along with a collection of smaller tests, two
very large test cases are added to check arithmetic on `undefined`.
The operations which have been rewritten in this PR are:
* `+`, `+%`, `+|`, `@addWithOverflow`
* `-`, `-%`, `-|`, `@subWithOverflow`
* `*`, `*%`, `*|`, `@mulWithOverflow`
* `/`, `@divFloor`, `@divTrunc`, `@divExact`
* `%`, `@rem`, `@mod`
Other arithmetic operations are currently unchanged.
Resolves: #22743Resolves: #22745Resolves: #22748Resolves: #22749Resolves: #22914
Functions like isMinGW() and isGnuLibC() have a good reason to exist: They look
at multiple components of the target. But functions like isWasm(), isDarwin(),
isGnu(), etc only exist to save 4-8 characters. I don't think this is a good
enough reason to keep them, especially given that:
* It's not immediately obvious to a reader whether target.isDarwin() means the
same thing as target.os.tag.isDarwin() precisely because isMinGW() and similar
functions *do* look at multiple components.
* It's not clear where we would draw the line. The logical conclusion before
this commit would be to also wrap Arch.isX86(), Os.Tag.isSolarish(),
Abi.isOpenHarmony(), etc... this obviously quickly gets out of hand.
* It's nice to just have a single correct way of doing something.
The old isARM() function was a portability trap. With the name it had, it seemed
like the obviously correct function to use, but it didn't include Thumb. In the
vast majority of cases where someone wants to ask "is the target Arm?", Thumb
*should* be included.
There are exactly 3 cases in the codebase where we do actually need to exclude
Thumb, although one of those is in Aro and mirrors a check in Clang that is
itself likely a bug. These rare cases can just add an extra isThumb() check.
Alex Rønne Petersen