mrq/bitsandbytes-rocm
1
0
Fork 1
Code Issues 1 Pull Requests Packages Projects Releases Wiki Activity

Increased occupancy.

Browse Source
This commit is contained in:
Tim Dettmers 2023-07-19 16:08:37 -07:00
parent e229fbce66
commit c82f51c0f7
3 changed files with 53 additions and 59 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
bitsandbytes/cuda_setup/main.py
View File

@ -361,4 +361,4 @@ def evaluate_cuda_setup():
"if not has_cublaslt (CC < 7.5), then we have to choose _nocublaslt.so" "if not has_cublaslt (CC < 7.5), then we have to choose _nocublaslt.so"
binary_name = f"libbitsandbytes_cuda{cuda_version_string}_nocublaslt.so" binary_name = f"libbitsandbytes_cuda{cuda_version_string}_nocublaslt.so"
return binary_name, cudart_path, cc, cuda_version_string return binary_name, cudart_path, cc, cuda_version_string

96
csrc/kernels.cu
View File

@ -3540,8 +3540,8 @@ template <typename T, int THREADS, int BITS> __global__ void kgemm_4bit_inferenc
float local_C = 0.0f; float local_C = 0.0f;
unsigned char local_B_4bit[num_values_8bit]; unsigned char local_B_4bit[num_values_8bit];
T local_B[num_values_4bit]; T local_B[num_values_4bit/4];
T local_A[num_values_4bit]; T local_A[num_values_4bit/4];
__shared__ T quant_map[16]; __shared__ T quant_map[16];
T local_absmax = T(0.0f); T local_absmax = T(0.0f);
@ -3582,61 +3582,55 @@ template <typename T, int THREADS, int BITS> __global__ void kgemm_4bit_inferenc
local_B_4bit[j] = 0b01110111; local_B_4bit[j] = 0b01110111;
} }
#pragma unroll for(int i = 0; i < 4; i++)
for(int k = 0; k < num_values_8bit; k++)
{ {
#if __CUDA_ARCH__ >= 800 #pragma unroll
local_B[k*2] = quant_map[local_B_4bit[k] >> 4]*local_absmax; for(int k = 0; k < num_values_8bit/4; k++)
local_B[k*2 + 1] = quant_map[local_B_4bit[k] & 0x0F]*local_absmax;
#else
// bf16 multipliation not supported
local_B[k*2] = T((float)quant_map[local_B_4bit[k] >> 4]*(float)local_absmax);
local_B[k*2 + 1] = T((float)quant_map[local_B_4bit[k] & 0x0F]*(float)local_absmax);
#endif
}
if(inner_idx+num_values_4bit < K)
{
// this is also relatively important for performance
if(BITS==16)
{ {
reinterpret_cast<int4(&)[num_values_4bit]>(local_A)[0] = reinterpret_cast<int4*>(A)[inner_idx/(num_values_4bit/4) + 0]; #if __CUDA_ARCH__ >= 800
reinterpret_cast<int4(&)[num_values_4bit]>(local_A)[1] = reinterpret_cast<int4*>(A)[inner_idx/(num_values_4bit/4) + 1]; local_B[k*2] = quant_map[local_B_4bit[(i*num_values_8bit/4) + k] >> 4]*local_absmax;
reinterpret_cast<int4(&)[num_values_4bit]>(local_A)[2] = reinterpret_cast<int4*>(A)[inner_idx/(num_values_4bit/4) + 2]; local_B[k*2 + 1] = quant_map[local_B_4bit[(i*num_values_8bit/4) + k] & 0x0F]*local_absmax;
reinterpret_cast<int4(&)[num_values_4bit]>(local_A)[3] = reinterpret_cast<int4*>(A)[inner_idx/(num_values_4bit/4) + 3]; #else
// bf16 multipliation not supported
local_B[k*2] = T((float)quant_map[local_B_4bit[(i*num_values_8bit/4) + k] >> 4]*(float)local_absmax);
local_B[k*2 + 1] = T((float)quant_map[local_B_4bit[(i*num_values_8bit/4) + k] & 0x0F]*(float)local_absmax);
#endif
}
if(inner_idx+(num_values_4bit/4) + (i*num_values_4bit/4) < K)
{
// this is also relatively important for performance
if(BITS==16)
{
reinterpret_cast<int4(&)[num_values_4bit]>(local_A)[0] = reinterpret_cast<int4*>(A)[inner_idx/(num_values_4bit/4) + i];
}
else
{
reinterpret_cast<int4(&)[num_values_4bit]>(local_A)[0] = reinterpret_cast<int4*>(A)[inner_idx/(num_values_4bit/8) + (2*i) + 0];
reinterpret_cast<int4(&)[num_values_4bit]>(local_A)[1] = reinterpret_cast<int4*>(A)[inner_idx/(num_values_4bit/8) + (2*i) + 1];
}
} }
else else
{ #pragma unroll
reinterpret_cast<int4(&)[num_values_4bit]>(local_A)[0] = reinterpret_cast<int4*>(A)[inner_idx/(num_values_4bit/8) + 0]; for(int k = 0; k < num_values_4bit/4; k++)
reinterpret_cast<int4(&)[num_values_4bit]>(local_A)[1] = reinterpret_cast<int4*>(A)[inner_idx/(num_values_4bit/8) + 1]; if(inner_idx + (i*num_values_4bit/4) + k < K)
reinterpret_cast<int4(&)[num_values_4bit]>(local_A)[2] = reinterpret_cast<int4*>(A)[inner_idx/(num_values_4bit/8) + 2]; local_A[k] = A[inner_idx + k + (i*num_values_4bit/4)];
reinterpret_cast<int4(&)[num_values_4bit]>(local_A)[3] = reinterpret_cast<int4*>(A)[inner_idx/(num_values_4bit/8) + 3]; else
reinterpret_cast<int4(&)[num_values_4bit]>(local_A)[4] = reinterpret_cast<int4*>(A)[inner_idx/(num_values_4bit/8) + 4]; local_A[k] = T(0.0f);
reinterpret_cast<int4(&)[num_values_4bit]>(local_A)[5] = reinterpret_cast<int4*>(A)[inner_idx/(num_values_4bit/8) + 5];
reinterpret_cast<int4(&)[num_values_4bit]>(local_A)[6] = reinterpret_cast<int4*>(A)[inner_idx/(num_values_4bit/8) + 6];
reinterpret_cast<int4(&)[num_values_4bit]>(local_A)[7] = reinterpret_cast<int4*>(A)[inner_idx/(num_values_4bit/8) + 7];
}
}
else // accumulate in float; small performance hit for Ampere, but lower error for outputs
#pragma unroll #pragma unroll
for(int k = 0; k < num_values_4bit; k++) for(int k = 0; k < num_values_4bit/4; k++)
if(inner_idx + k < K) {
local_A[k] = A[inner_idx + k]; #if __CUDA_ARCH__ >= 800
else local_C += (float)(local_A[k]*local_B[k]);
local_A[k] = T(0.0f); #else
// bf16 multipliation not supported
local_C += ((float)local_A[k]*(float)local_B[k]);
// accumulate in float; small performance hit for Ampere, but lower error for outputs #endif
#pragma unroll }
for(int k = 0; k < num_values_4bit; k++)
{
#if __CUDA_ARCH__ >= 800
local_C += (float)(local_A[k]*local_B[k]);
#else
// bf16 multipliation not supported
local_C += ((float)local_A[k]*(float)local_B[k]);
#endif
} }
} }

14
tests/test_functional.py
View File

@ -2366,7 +2366,7 @@ def test_normal_map_tree():
def test_gemv_4bit(dtype, storage_type, double_quant, kind): def test_gemv_4bit(dtype, storage_type, double_quant, kind):
for dim in [128, 256, 512, 1024]: for dim in [128, 256, 512, 1024]:
#for dim in [4*1024]: #for dim in [4*1024]:
#for dim in [1*128]: #for dim in [1*16]:
errs1 = [] errs1 = []
errs2 = [] errs2 = []
errs3 = [] errs3 = []
@ -2446,11 +2446,11 @@ def test_gemv_4bit(dtype, storage_type, double_quant, kind):
# #
#print('='*80) #print('='*80)
#print(f'For matmul: {A.shape}, {B.shape}, {kind}, {dtype}, {storage_type}, double_quant={double_quant}:') #print(f'For matmul: {A.shape}, {B.shape}, {kind}, {dtype}, {storage_type}, double_quant={double_quant}:')
#print(C1.flatten()[-20:]) print(C1.flatten()[-20:])
#print(C2.flatten()[-20:]) print(C2.flatten()[-20:])
#print(f'inference vs training abs: {err1}') print(f'inference vs training abs: {err1}')
#print(f'inference vs training rel: {relerr1}') print(f'inference vs training rel: {relerr1}')
#print(f'inference vs training max: {maxerr1}') print(f'inference vs training max: {maxerr1}')
#print(f'inference vs training vs torch err ratio abs: {absratio}') #print(f'inference vs training vs torch err ratio abs: {absratio}')
#print(f'inference vs training vs torch err ratio rel: {relratio}') #print(f'inference vs training vs torch err ratio rel: {relratio}')
#print(f'inference vs training vs torch err ratio max: {maxratio}') #print(f'inference vs training vs torch err ratio max: {maxratio}')
@ -2478,7 +2478,7 @@ def test_gemv_4bit(dtype, storage_type, double_quant, kind):
assert maxratio < 1.005 and maxratio > 0.995 assert maxratio < 1.005 and maxratio > 0.995
elif dtype == torch.bfloat16: elif dtype == torch.bfloat16:
if dim <= 512: if dim <= 512:
assert err1 < 5e-4 assert err1 < 6e-4
assert relerr1 < 0.007 assert relerr1 < 0.007
assert maxerr1 < 0.015 assert maxerr1 < 0.015
else: else: