105 lines
4.4 KiB
Python
105 lines
4.4 KiB
Python
import time
|
|
import numpy as np
|
|
from tinygrad.helpers import dtypes, getenv, prod
|
|
from tinygrad.runtime.ops_hip import RawHIPBuffer, HIPProgram
|
|
|
|
# AMD_LOG_LEVEL=3 ./MIOpenDriver gemm --iter 1000 --time 1 --a_w 2048 --a_h 2048 --b_w 2048
|
|
# 5.5: Cijk_Ailk_Bljk_HHS_BH_MT128x128x16_MI16x16x16x1_SN_1LDSB0_APM1_ABV0_ACED0_AF0EM1_AF1EM1_AMAS3_ASE_ASGT_ASAE01_ASCE01_ASEM1_AAC0_BL1_BS1_DTL0_DTVA0_DVO0_ETSP_EPS1_FL0_GRVW8_GSU1_GSUASB_GLS0_ISA1100_IU1_K1_KLA_LBSPP128_LPA0_LPB8_LDL1_LRVW16_LWPMn1_LDW0_FMA_MIAV1_MDA2_NTA0_NTB0_NTC0_NTD0_NEPBS0_NLCA1_NLCB1_ONLL1_OPLV0_PK0_PAP0_PGR1_PLR1_RK0_SIA1_SS1_SU32_SUM0_SUS128_SCIUI1_SPO0_SRVW0_SSO0_SVW4_SNLL0_TT4_64_TLDS1_USFGROn1_VAW2_VSn1_VW4_WSGRA1_WSGRB1_WS32_WG32_4_1_WGM4
|
|
# 5.6: Cijk_Ailk_Bljk_HHS_BH_MT128x128x16_MI16x16x16x1_SN_1LDSB0_APM1_ABV0_ACED0_AF0EM1_AF1EM1_AMAS3_ASE_ASGT_ASLT_ASAE01_ASCE01_ASEM1_AAC0_BL1_BS1_DTL0_DTVA0_DVO0_ETSP_EPS1_FL0_GRPM1_GRVW8_GSU1_GSUASB_GLS0_ISA1100_IU1_K1_KLA_LBSPP128_LPA0_LPB8_LDL1_LRVW16_LWPMn1_LDW0_FMA_MIAV1_MDA2_MO40_NTA0_NTB0_NTC0_NTD0_NEPBS0_NLCA1_NLCB1_ONLL1_OPLV0_PK0_PAP0_PGR1_PLR1_RK0_SIA1_SS1_SU32_SUM0_SUS128_SCIUI1_SPO0_SRVW0_SSO0_SVW4_SNLL0_TT4_64_TLDS1_USFGROn1_VAW2_VSn1_VW4_WSGRA1_WSGRB1_WS32_WG32_4_1_WGM4
|
|
# gets ~100
|
|
# hipExtModuleLaunchKernel ( 0x0x16ccde0, 2048, 16, 1, 128, 1, 1,
|
|
# 161.60 us = 106.31 TFLOPS
|
|
# with --batch_count 8 / 1.258128 ms / (8*2048*2048*2048*2)/(1.258128)*1e-9 / 109.24 TFLOPS
|
|
|
|
# we only get ~53
|
|
# KY=2 KX=2 N=2048 python3 extra/gemm/hip_matmul.py
|
|
# 4194304 324.76 us, would be 52899.88 GFLOPS matmul, 154.98 GB/s
|
|
|
|
N = getenv("N", 2048)
|
|
KX = getenv("KX", 4)
|
|
KY = getenv("KY", 4)
|
|
assert N%(16*KX) == 0, f"N must be multiple of {16*KX}"
|
|
assert N%(16*KY) == 0, f"N must be multiple of {16*KY}"
|
|
FLOPS = N*N*N*2
|
|
BW = N*N*3*4
|
|
|
|
a = RawHIPBuffer(N*N, dtypes.float32)
|
|
|
|
nb = np.random.default_rng().standard_normal(size=(N,N), dtype=np.float32).astype(np.float16)
|
|
nc = np.random.default_rng().standard_normal(size=(N,N), dtype=np.float32).astype(np.float16)
|
|
b = RawHIPBuffer.fromCPU(nb)
|
|
c = RawHIPBuffer.fromCPU(nc)
|
|
|
|
prog = HIPProgram("test", f"""
|
|
#define F32
|
|
typedef float float8 __attribute__((ext_vector_type(8)));
|
|
typedef _Float16 half16 __attribute__((ext_vector_type(16)));
|
|
extern "C" __global__ void __launch_bounds__ (128, 1) test(float* c, __half* a, __half* b) {{
|
|
const int gx = blockIdx.x*2 + threadIdx.y;
|
|
const int gy = blockIdx.y*2 + threadIdx.z;
|
|
|
|
const int lIdx = threadIdx.x;
|
|
const int lane = lIdx%16;
|
|
|
|
c += gx*{KX*16}*{N} + gy*{KY*16} + (lIdx/16)*{N} + lane;
|
|
a += gx*{KX*16}*{N};
|
|
b += gy*{KY*16};
|
|
|
|
half16 a_frag[{KX}];
|
|
half16 b_frag[{KY}];
|
|
#ifdef F32
|
|
float8 c_frag[{KY}][{KX}] = {{}};
|
|
#else
|
|
half16 c_frag[{KY}][{KX}] = {{}};
|
|
#endif
|
|
|
|
for (int k = 0; k < {N}; k += 16) {{
|
|
__syncthreads();
|
|
for (int ele = 0; ele < 16; ++ele) {{
|
|
for (int x = 0; x < {KX}; x++) {{
|
|
a_frag[x][ele] = a[(k+ele) + x*{16*N} + {N}*lane];
|
|
}}
|
|
}}
|
|
for (int ele = 0; ele < 16; ++ele) {{
|
|
for (int y = 0; y < {KY}; y++) {{
|
|
b_frag[y][ele] = b[(k+ele)*{N} + y*16 + lane];
|
|
}}
|
|
}}
|
|
for (int y = 0; y < {KY}; y++) {{
|
|
for (int x = 0; x < {KX}; x++) {{
|
|
#ifdef F32
|
|
c_frag[y][x] = __builtin_amdgcn_wmma_f32_16x16x16_f16_w32(a_frag[x], b_frag[y], c_frag[y][x]);
|
|
#else
|
|
c_frag[y][x] = __builtin_amdgcn_wmma_f16_16x16x16_f16_w32(a_frag[x], b_frag[y], c_frag[y][x], false);
|
|
#endif
|
|
}}
|
|
}}
|
|
}}
|
|
|
|
for (int ele = 0; ele < 8; ++ele) {{
|
|
for (int y = 0; y < {KY}; y++) {{
|
|
for (int x = 0; x < {KX}; x++) {{
|
|
#ifdef F32
|
|
c[ele*{2*N} + y*16 + x*{16*N}] = c_frag[y][x][ele];
|
|
#else
|
|
c[ele*{2*N} + y*16 + x*{16*N}] = c_frag[y][x][ele*2];
|
|
#endif
|
|
}}
|
|
}}
|
|
}}
|
|
}}""")
|
|
|
|
def timeit(fxn):
|
|
st = time.perf_counter()
|
|
et = fxn()
|
|
ret = time.perf_counter() - st # NOTE: et doesn't contain the launch overhead
|
|
#print(f"{ret*1e6:.2f} us")
|
|
return et
|
|
|
|
global_size, local_size = [N//(KX*16*2), N//(KY*16*2), 1], [32, 2, 2]
|
|
print("global/local size", global_size, local_size, f"local_size:{prod(local_size)} total_size:{prod(global_size+local_size)}")
|
|
tm = min([timeit(lambda: prog(global_size, local_size, a, b, c, wait=True)) for _ in range(1000)])
|
|
na = a.toCPU().reshape(N,N)
|
|
comp = nb.astype(np.float32) @ nc.astype(np.float32)
|
|
print(f"{N*N:10d} {tm*1e6:9.2f} us, would be {FLOPS*1e-9/tm:9.2f} GFLOPS matmul, {BW*1e-9/tm:.2f} GB/s")
|
|
np.testing.assert_allclose(na, comp, atol=1e-2, rtol=1e-2) |