mirror of
https://github.com/ggml-org/llama.cpp.git
synced 2025-11-09 10:17:06 +00:00
CANN: Fix SOC_TYPE compile bug (#10519)
* CANN: Fix the bug build fail on Ascend310P under two cases: 1) Manual specify SOC_TYPE 2) Under some unusual compile environment * Update the cann backend News content: Support F16 and F32 data type model for Ascend 310P NPU. * fix CANN compile fail bug: the assert in ascend kernel function doesn't supportted on some CANN version
This commit is contained in:
@@ -1,6 +1,14 @@
|
||||
#include "kernel_operator.h"
|
||||
|
||||
using namespace AscendC;
|
||||
#ifdef ASCEND_310P
|
||||
extern "C" __global__ __aicore__ void ascendc_quantize_f16_q8_0(
|
||||
GM_ADDR input_gm, GM_ADDR output_gm, GM_ADDR input_ne_gm,
|
||||
GM_ADDR input_nb_gm, GM_ADDR output_ne_gm) {
|
||||
// let following test cases can continue run, here just print error information. Of Cource the test case that call this operator is failed.
|
||||
printf("Ascend310P not support f16->8bit quantization.\n");
|
||||
}
|
||||
#else
|
||||
|
||||
#define BUFFER_NUM 2
|
||||
#define QK8_0 32
|
||||
@@ -206,3 +214,5 @@ extern "C" __global__ __aicore__ void ascendc_quantize_f16_q8_0(
|
||||
op.init(input_gm, output_gm, input_ne_ub, input_nb_ub, output_ne_ub);
|
||||
op.calculate();
|
||||
}
|
||||
|
||||
#endif // #ifdef ASCEND_310P
|
||||
|
||||
Reference in New Issue
Block a user