test / src /f32-gemm /4x1-aarch64-neonfma-ld64.S.in
Androidonnxfork's picture
Upload folder using huggingface_hub
8b7c501
// Copyright 2023 Google LLC
//
// This source code is licensed under the BSD-style license found in the
// LICENSE file in the root directory of this source tree.
$assert DATATYPE in ["F32", "QC8", "QC4"]
#include <xnnpack/assembly.h>
$DATATYPE_SPEC = {"F32": "f32", "QC8": "f32_qc8w", "QC4": "f32_qc4w"}[DATATYPE]
# void xnn_${DATATYPE_SPEC}_gemm${"inc" if INC else ""}_minmax_ukernel_4x1__asm_aarch64_neonfma_ld64(
# size_t mr, x0
# size_t nc, x1
# size_t kc, x2 / x0
# const float* a, x3
# size_t a_stride, x4
# const float* w, x5
# float* c, x6
# size_t cm_stride, x7
# size_t cn_stride, [sp] -> x14
$if INC:
# const float* acc, [sp + 8] -> x15
# const xnn_f32_minmax_params* params) [sp + 16] -> (x8)
$elif DATATYPE == "QC4":
# const xnn_f32_qc4w_minmax_params* params) [sp + 8] -> (x8)
$else:
# const xnn_f32_minmax_params* params) [sp + 8] -> (x8)
# d8-d15, x19-x30 need to be preserved if used. x18 is reserved by the OS.
# Register usage
# A0 x3 v0
# A1 x11 v1
# A2 x12 v2
# A3 x4 v3
# B x5 v20
# C0 x6 v24
# C1 x9 v26
# C2 x10 v28
# C3 x7 v30
# Clamp v4 v5
$if DATATYPE == "QC4":
# zerop/mask v6 v7
# temp v21
BEGIN_FUNCTION xnn_${DATATYPE_SPEC}_gemm${"inc" if INC else ""}_minmax_ukernel_4x1__asm_aarch64_neonfma_ld64
$if INC:
# Load cn_stride, acc
LDP x14, x15, [sp]
# Load params pointer
LDR x8, [sp, 16]
$else:
# Load cn_stride, params pointer
LDP x14, x8, [sp]
# Clamp A and C pointers
CMP x0, 2 // if mr < 2
ADD x11, x3, x4 // a1 = a0 + a_stride
ADD x9, x6, x7 // c1 = c0 + cm_stride
CSEL x11, x3, x11, LO // a1 = a0
CSEL x9, x6, x9, LO // c1 = c0
$if DATATYPE == "QC4":
# Load min/max/zerop values
LD3R {v4.2s, v5.2s, v6.2s}, [x8]
MOVI v7.8b, 15
$else:
# Load min/max values
LD2R {v4.2s, v5.2s}, [x8]
ADD x12, x11, x4 // a2 = a1 + a_stride
ADD x10, x9, x7 // c2 = c1 + cm_stride
// if mr <= 2
CSEL x12, x11, x12, LS // a2 = a1
CSEL x10, x9, x10, LS // c2 = c1
CMP x0, 4 // if mr < 4
ADD x4, x12, x4 // a3 = a2 + a_stride
ADD x7, x10, x7 // c3 = c2 + cm_stride
CSEL x4, x12, x4, LO // a3 = a2
CSEL x7, x10, x7, LO // c3 = c2
0:
$if INC:
MOVI v24.2s, 0
MOVI v26.2s, 0
MOVI v28.2s, 0
MOVI v30.2s, 0
# Load initial accumulators
LDR s24, [x15], 4
LDR s26, [x15], 4
LDR s28, [x15], 4
LDR s30, [x15], 4
$else:
# Load initial bias from w into accumulators
MOVI v24.2s, 0
LDR s24, [x5], 4
MOV v26.8b, v24.8b
MOV v28.8b, v24.8b
MOV v30.8b, v24.8b
# Is there at least 2 floats (8 bytes)?
SUBS x0, x2, 8 // k = kc - 8
B.LO 3f
# Main loop - 2 floats of A (8 bytes)
1:
$if DATATYPE == "F32":
LDR d0, [x3], 8
LDR d20, [x5], 8 // 2 F32 weights
LDR d1, [x11], 8
LDR d2, [x12], 8
LDR d3, [x4], 8
$elif DATATYPE == "QC4":
LDR b21, [x5], 1 // 2 QC4 weights
LDR d0, [x3], 8
AND v20.8b, v21.8b, v7.8b // first weight
USHR v21.8b, v21.8b, 4 // second weight
INS v20.b[1], v21.b[0] // both weights
SADDW v20.8h, v6.8h, v20.8b
LDR d1, [x11], 8
SXTL v20.4s, v20.4h
LDR d2, [x12], 8
SCVTF v20.2s, v20.2s
LDR d3, [x4], 8
$else:
LDR h20, [x5], 2 // 8 QC8 weights
LDR d0, [x3], 8
SXTL v20.8h, v20.8b
LDR d1, [x11], 8
SXTL v20.4s, v20.4h
LDR d2, [x12], 8
SCVTF v20.2s, v20.2s
LDR d3, [x4], 8
SUBS x0, x0, 8
FMLA v24.2s, v20.2s, v0.2s
FMLA v26.2s, v20.2s, v1.2s
FMLA v28.2s, v20.2s, v2.2s
FMLA v30.2s, v20.2s, v3.2s
B.HS 1b
FADDP s24, v24.2s
FADDP s26, v26.2s
FADDP s28, v28.2s
FADDP s30, v30.2s
# Is there a remainder?- 1 float of A (4 bytes)
TBNZ x0, 2, 3f
2:
$if DATATYPE in ["QC8", "QC4"]:
# Scale
LDR s20, [x5], 4
FMUL s24, s24, v20.s[0]
FMUL s26, s26, v20.s[0]
FMUL s28, s28, v20.s[0]
FMUL s30, s30, v20.s[0]
# Clamp
FMAX s24, s24, s4
SUBS x1, x1, 1
FMAX s26, s26, s4
FMAX s28, s28, s4
FMAX s30, s30, s4
FMIN s24, s24, s5
FMIN s26, s26, s5
FMIN s28, s28, s5
FMIN s30, s30, s5
$if INC:
ST1 {v30.s}[0], [x7], x14
SUB x3, x3, x2 // a0 -= kc
ST1 {v28.s}[0], [x10], x14
SUB x11, x11, x2 // a1 -= kc
ST1 {v26.s}[0], [x9], x14
SUB x12, x12, x2 // a2 -= kc
ST1 {v24.s}[0], [x6], x14
SUB x4, x4, x2 // a3 -= kc
$else:
ST1 {v24.s}[0], [x6], x14
SUB x3, x3, x2 // a0 -= kc
ST1 {v26.s}[0], [x9], x14
SUB x11, x11, x2 // a1 -= kc
ST1 {v28.s}[0], [x10], x14
SUB x12, x12, x2 // a2 -= kc
ST1 {v30.s}[0], [x7], x14
SUB x4, x4, x2 // a3 -= kc
B.HI 0b
RET
# Remainder- 1 float of A (4 bytes)
3:
LDR s0, [x3], 4
$if DATATYPE == "F32":
LDR s20, [x5], 4
$elif DATATYPE == "QC4":
LDR b20, [x5], 1
SADDW v20.8h, v6.8h, v20.8b
SXTL v20.4s, v20.4h
SCVTF v20.2s, v20.2s
$else:
LDR b20, [x5], 1
SXTL v20.8h, v20.8b
SXTL v20.4s, v20.4h
SCVTF v20.2s, v20.2s
LDR s1, [x11], 4
LDR s2, [x12], 4
LDR s3, [x4], 4
SUBS x0, x0, 4
FMLA s24, s20, v0.s[0]
FMLA s26, s20, v1.s[0]
FMLA s28, s20, v2.s[0]
FMLA s30, s20, v3.s[0]
B 2b
RET
END_FUNCTION xnn_${DATATYPE_SPEC}_gemm${"inc" if INC else ""}_minmax_ukernel_4x1__asm_aarch64_neonfma_ld64
#ifdef __ELF__
.section ".note.GNU-stack","",%progbits
#endif