CUDA PTX inline asm listing fails internal exception for `-e stmt_html`

[antonysigma]

Description

Error log:

Unhandled exception: halide/src/StmtToHTML.cpp:841 Condition failed: current_id != -1

CUDA PTX listing that causes the current_id to reset:

I am trying to create a minimum reproducible case in the meantime. My hunch is that StmtToHTML.cpp does not reference count the nested braces, and instead treating { as a new CUDA kernel.

Related to: #7843 .

Reproducing case

.visible .entry _kernel_total_field_ft_mux_s0___outermost___outermost_v81_block_id_x (
    .param .u64 _kernel_total_field_ft_mux_s0___outermost___outermost_v81_block_id_x_param_0,
    .param .u64 _kernel_total_field_ft_mux_s0___outermost___outermost_v81_block_id_x_param_1,
    .param .u64 _kernel_total_field_ft_mux_s0___outermost___outermost_v81_block_id_x_param_2,
    .param .u64 _kernel_total_field_ft_mux_s0___outermost___outermost_v81_block_id_x_param_3
)                                       // @_kernel_total_field_ft_mux_s0___outermost___outermost_v81_block_id_x
{
...
$L__BB14_14:                            // %.lr.ph43
                                        //   Parent Loop BB14_1 Depth=1
                                        //     Parent Loop BB14_2 Depth=2
                                        //       Parent Loop BB14_3 Depth=3
                                        //         Parent Loop BB14_4 Depth=4
                                        // =>        This Inner Loop Header: Depth=5
    ld.const.u32 	%r131, [%rd37];
    // begin inline asm
{
... [assertion failure]

How did you get Halide?

None

Halide version

18.0.0

Halide commit (if known)

No response

Target

x86-64-linux-cuda-cuda_capability_75

Operating system

No response

Additional context

No response