dynamic distpatch of fp8 kernels (#14245)

Signed-off-by: Jeff Daily <jeff.daily@amd.com>
This commit is contained in:
Jeff Daily
2025-03-11 07:54:56 -07:00
committed by GitHub
parent 08a1a1121d
commit a1c8f3796c
25 changed files with 292 additions and 159 deletions

View File

@ -6,6 +6,11 @@
#include <torch/all.h>
// Need a special dispatch case macro since we will nest the FP8 dispatch.
// Instead of the usual 'scalar_t', this names the dispatched type 'fp8_t'.
#define AT_DISPATCH_FP8_CASE(enum_type, ...) \
AT_PRIVATE_CASE_TYPE_USING_HINT(enum_type, fp8_t, __VA_ARGS__)
#define VLLM_DISPATCH_CASE_FLOATING_TYPES(...) \
AT_DISPATCH_CASE(at::ScalarType::Float, __VA_ARGS__) \
AT_DISPATCH_CASE(at::ScalarType::Half, __VA_ARGS__) \
@ -14,17 +19,32 @@
#define VLLM_DISPATCH_FLOATING_TYPES(TYPE, NAME, ...) \
AT_DISPATCH_SWITCH(TYPE, NAME, VLLM_DISPATCH_CASE_FLOATING_TYPES(__VA_ARGS__))
// TODO(luka/varun): use FP8_TYPE macro after refactoring
#ifndef USE_ROCM
// ROCm devices might use either fn or fnuz, so set up dispatch table for both.
// A host-based check at runtime will create a preferred FP8 type for ROCm
// such that the correct kernel is dispatched.
#ifdef USE_ROCM
#define VLLM_DISPATCH_CASE_FP8_TYPES(...) \
AT_DISPATCH_FP8_CASE(at::ScalarType::Float8_e4m3fn, __VA_ARGS__) \
AT_DISPATCH_FP8_CASE(at::ScalarType::Float8_e4m3fnuz, __VA_ARGS__)
#define VLLM_DISPATCH_CASE_QUANT_TYPES(...) \
AT_DISPATCH_CASE(at::ScalarType::Float8_e4m3fn, __VA_ARGS__) \
AT_DISPATCH_CASE(at::ScalarType::Float8_e4m3fnuz, __VA_ARGS__) \
AT_DISPATCH_CASE(at::ScalarType::Char, __VA_ARGS__)
#else
#define VLLM_DISPATCH_CASE_FP8_TYPES(...) \
AT_DISPATCH_FP8_CASE(at::ScalarType::Float8_e4m3fn, __VA_ARGS__)
#define VLLM_DISPATCH_CASE_QUANT_TYPES(...) \
AT_DISPATCH_CASE(at::ScalarType::Float8_e4m3fn, __VA_ARGS__) \
AT_DISPATCH_CASE(at::ScalarType::Char, __VA_ARGS__)
#else
#define VLLM_DISPATCH_CASE_QUANT_TYPES(...) \
AT_DISPATCH_CASE(at::ScalarType::Float8_e4m3fnuz, __VA_ARGS__) \
AT_DISPATCH_CASE(at::ScalarType::Char, __VA_ARGS__)
#endif
// When using this dispatch macro, the type is 'fp8_t' not 'scalar_t'.
// See AT_DISPATCH_FP8_CASE above.
#define VLLM_DISPATCH_FP8_TYPES(TYPE, NAME, ...) \
AT_DISPATCH_SWITCH(TYPE, NAME, VLLM_DISPATCH_CASE_FP8_TYPES(__VA_ARGS__))
#define VLLM_DISPATCH_QUANT_TYPES(TYPE, NAME, ...) \
AT_DISPATCH_SWITCH(TYPE, NAME, VLLM_DISPATCH_CASE_QUANT_TYPES(__VA_ARGS__))