[Bug] TSTORE_FP _validate_fixpipe_payload 不接受 scaling pointer 作为 f32_f16/f32_bf16 的 vector payload

[pbbb205]

Component

PTO Dialect / ODS (include/PTO/IR)

Description

TSTORE_FP fails: _validate_fixpipe_payload rejects SemanticPtrType(memory_space="scaling") as f32_f16/f32_bf16 pre_quant payload — only accepts scalar

Reproduction (minimal)

xx

Expected behavior

所有使用 pto.tstore_fp（ACC + scaling buffer → GM）的场景。

Actual behavior / error logs

pto.mte_l0c_gm pre_quant payload must be an f16, bf16, or f32 scalar in TileLang DSL v1

Git commit

c038b4b

Host platform

None

Target Ascend arch (if relevant)

None

PTOAS build level (if relevant)

None

[Zhendong404]

根据pre_quant的定义，f32_f16/f32_bf16只能接受标量payload

pre_quant mode families:

Family	Acc source	Result meaning	Payload
f32_f16, f32_bf16	f32	Convert f32 accumulator values to f16 or bf16; rounding is nearest, ties to even	Scalar payload is required by syntax but does not select per-channel scaling
qf322hif8_pre_*, qf322fp8_pre_*	f32	Scale and quantize f32 to hif8/fp8-style destination payloads	Scalar scale or vector scale rows; hybrid modes use the target hybrid rule
qf322f32_pre_*	f32	Apply quant scaling while keeping f32 destination values	Scalar scale or vector scale rows
qf322f16_pre_*, qf322bf16_pre_*	f32	Scale f32, then convert to f16 or bf16 destination values	Scalar scale or vector scale rows
qf322b8_pre_*, qf322s4_pre_*	f32	Scale, offset, round, and narrow f32 to 8-bit or signed 4-bit integer payloads	Scalar or vector scale/offset parameter set
qf162b8_pre_*, qf162s4_pre_*	f32	Convert through an f16-domain pre-stage, then scale/narrow to integer payloads	Scalar or vector scale/offset parameter set
qf162s16_pre_*	i32	Convert through an f16-domain pre-stage, then scale/narrow to signed 16-bit payloads	Scalar or vector scale/offset parameter set
deqs32_int_*, deqs16_*	i32	Rescale integer accumulator values in an integer destination family	Scalar or vector multiplier/offset parameter set
req8_*, req4_*	i32	Requantize i32 accumulator values to 8-bit or 4-bit integer payloads	Scalar or vector multiplier/offset/sign parameter set
deqf16_*	i32	Dequantize i32 accumulator values to f16 destination values	Scalar or vector multiplier/offset parameter set
qs322bf16_pre_*	i32	Scale i32 accumulator values and convert to bf16 destination values	Scalar or vector multiplier/offset parameter set