1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
|
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include "shader_recompiler/frontend/maxwell/translate/impl/common_encoding.h"
#include "shader_recompiler/frontend/maxwell/translate/impl/half_floating_point_helper.h"
namespace Shader::Maxwell {
namespace {
enum class FloatFormat : u64 {
F16 = 1,
F32 = 2,
F64 = 3,
};
enum class RoundingOp : u64 {
None = 0,
Pass = 3,
Round = 8,
Floor = 9,
Ceil = 10,
Trunc = 11,
};
[[nodiscard]] u32 WidthSize(FloatFormat width) {
switch (width) {
case FloatFormat::F16:
return 16;
case FloatFormat::F32:
return 32;
case FloatFormat::F64:
return 64;
default:
throw NotImplementedException("Invalid width {}", width);
}
}
void F2F(TranslatorVisitor& v, u64 insn, const IR::F16F32F64& src_a, bool abs) {
union {
u64 insn;
BitField<0, 8, IR::Reg> dest_reg;
BitField<44, 1, u64> ftz;
BitField<45, 1, u64> neg;
BitField<47, 1, u64> cc;
BitField<50, 1, u64> sat;
BitField<39, 4, u64> rounding_op;
BitField<39, 2, FpRounding> rounding;
BitField<10, 2, FloatFormat> src_size;
BitField<8, 2, FloatFormat> dst_size;
[[nodiscard]] RoundingOp RoundingOperation() const {
constexpr u64 rounding_mask = 0x0B;
return static_cast<RoundingOp>(rounding_op.Value() & rounding_mask);
}
} const f2f{insn};
if (f2f.cc != 0) {
throw NotImplementedException("F2F CC");
}
IR::F16F32F64 input{v.ir.FPAbsNeg(src_a, abs, f2f.neg != 0)};
const bool any_fp64{f2f.src_size == FloatFormat::F64 || f2f.dst_size == FloatFormat::F64};
IR::FpControl fp_control{
.no_contraction = false,
.rounding = IR::FpRounding::DontCare,
.fmz_mode = (f2f.ftz != 0 && !any_fp64 ? IR::FmzMode::FTZ : IR::FmzMode::None),
};
if (f2f.src_size != f2f.dst_size) {
fp_control.rounding = CastFpRounding(f2f.rounding);
input = v.ir.FPConvert(WidthSize(f2f.dst_size), input, fp_control);
} else {
switch (f2f.RoundingOperation()) {
case RoundingOp::None:
case RoundingOp::Pass:
// Make sure NANs are handled properly
switch (f2f.src_size) {
case FloatFormat::F16:
input = v.ir.FPAdd(input, v.ir.FPConvert(16, v.ir.Imm32(0.0f)), fp_control);
break;
case FloatFormat::F32:
input = v.ir.FPAdd(input, v.ir.Imm32(0.0f), fp_control);
break;
case FloatFormat::F64:
input = v.ir.FPAdd(input, v.ir.Imm64(0.0), fp_control);
break;
}
break;
case RoundingOp::Round:
input = v.ir.FPRoundEven(input, fp_control);
break;
case RoundingOp::Floor:
input = v.ir.FPFloor(input, fp_control);
break;
case RoundingOp::Ceil:
input = v.ir.FPCeil(input, fp_control);
break;
case RoundingOp::Trunc:
input = v.ir.FPTrunc(input, fp_control);
break;
default:
throw NotImplementedException("Unimplemented rounding mode {}", f2f.rounding.Value());
}
}
if (f2f.sat != 0 && !any_fp64) {
input = v.ir.FPSaturate(input);
}
switch (f2f.dst_size) {
case FloatFormat::F16: {
const IR::F16 imm{v.ir.FPConvert(16, v.ir.Imm32(0.0f))};
v.X(f2f.dest_reg, v.ir.PackFloat2x16(v.ir.CompositeConstruct(input, imm)));
break;
}
case FloatFormat::F32:
v.F(f2f.dest_reg, input);
break;
case FloatFormat::F64:
v.D(f2f.dest_reg, input);
break;
default:
throw NotImplementedException("Invalid dest format {}", f2f.dst_size.Value());
}
}
} // Anonymous namespace
void TranslatorVisitor::F2F_reg(u64 insn) {
union {
u64 insn;
BitField<49, 1, u64> abs;
BitField<10, 2, FloatFormat> src_size;
BitField<41, 1, u64> selector;
} const f2f{insn};
IR::F16F32F64 src_a;
switch (f2f.src_size) {
case FloatFormat::F16: {
auto [lhs_a, rhs_a]{Extract(ir, GetReg20(insn), Swizzle::H1_H0)};
src_a = f2f.selector != 0 ? rhs_a : lhs_a;
break;
}
case FloatFormat::F32:
src_a = GetFloatReg20(insn);
break;
case FloatFormat::F64:
src_a = GetDoubleReg20(insn);
break;
default:
throw NotImplementedException("Invalid dest format {}", f2f.src_size.Value());
}
F2F(*this, insn, src_a, f2f.abs != 0);
}
void TranslatorVisitor::F2F_cbuf(u64 insn) {
union {
u64 insn;
BitField<49, 1, u64> abs;
BitField<10, 2, FloatFormat> src_size;
BitField<41, 1, u64> selector;
} const f2f{insn};
IR::F16F32F64 src_a;
switch (f2f.src_size) {
case FloatFormat::F16: {
auto [lhs_a, rhs_a]{Extract(ir, GetCbuf(insn), Swizzle::H1_H0)};
src_a = f2f.selector != 0 ? rhs_a : lhs_a;
break;
}
case FloatFormat::F32:
src_a = GetFloatCbuf(insn);
break;
case FloatFormat::F64:
src_a = GetDoubleCbuf(insn);
break;
default:
throw NotImplementedException("Invalid dest format {}", f2f.src_size.Value());
}
F2F(*this, insn, src_a, f2f.abs != 0);
}
void TranslatorVisitor::F2F_imm([[maybe_unused]] u64 insn) {
union {
u64 insn;
BitField<49, 1, u64> abs;
BitField<10, 2, FloatFormat> src_size;
BitField<41, 1, u64> selector;
BitField<20, 19, u64> imm;
BitField<56, 1, u64> imm_neg;
} const f2f{insn};
IR::F16F32F64 src_a;
switch (f2f.src_size) {
case FloatFormat::F16: {
const u32 imm{static_cast<u32>(f2f.imm & 0x0000ffff)};
const IR::Value vector{ir.UnpackFloat2x16(ir.Imm32(imm | (imm << 16)))};
src_a = IR::F16{ir.CompositeExtract(vector, f2f.selector != 0 ? 0 : 1)};
if (f2f.imm_neg != 0) {
throw NotImplementedException("Neg bit on F16");
}
break;
}
case FloatFormat::F32:
src_a = GetFloatImm20(insn);
break;
case FloatFormat::F64:
src_a = GetDoubleImm20(insn);
break;
default:
throw NotImplementedException("Invalid dest format {}", f2f.src_size.Value());
}
F2F(*this, insn, src_a, f2f.abs != 0);
}
} // namespace Shader::Maxwell
|
