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
|
// Copyright 2021 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <string_view>
#include "shader_recompiler/backend/glsl/emit_context.h"
#include "shader_recompiler/backend/glsl/emit_glsl_instructions.h"
#include "shader_recompiler/frontend/ir/value.h"
#include "shader_recompiler/profile.h"
namespace Shader::Backend::GLSL {
namespace {
void SetInBoundsFlag(EmitContext& ctx, IR::Inst& inst) {
IR::Inst* const in_bounds{inst.GetAssociatedPseudoOperation(IR::Opcode::GetInBoundsFromOp)};
if (!in_bounds) {
return;
}
ctx.AddU1("{}=shfl_in_bounds;", *in_bounds);
in_bounds->Invalidate();
}
std::string ComputeMinThreadId(std::string_view thread_id, std::string_view segmentation_mask) {
return fmt::format("({}&{})", thread_id, segmentation_mask);
}
std::string ComputeMaxThreadId(std::string_view min_thread_id, std::string_view clamp,
std::string_view not_seg_mask) {
return fmt::format("({})|({}&{})", min_thread_id, clamp, not_seg_mask);
}
std::string GetMaxThreadId(std::string_view thread_id, std::string_view clamp,
std::string_view segmentation_mask) {
const auto not_seg_mask{fmt::format("(~{})", segmentation_mask)};
const auto min_thread_id{ComputeMinThreadId(thread_id, segmentation_mask)};
return ComputeMaxThreadId(min_thread_id, clamp, not_seg_mask);
}
void UseShuffleNv(EmitContext& ctx, IR::Inst& inst, std::string_view shfl_op,
std::string_view value, std::string_view index,
[[maybe_unused]] std::string_view clamp, std::string_view segmentation_mask) {
const auto width{fmt::format("32u>>(bitCount({}&31u))", segmentation_mask)};
ctx.AddU32("{}={}({},{},{},shfl_in_bounds);", inst, shfl_op, value, index, width);
SetInBoundsFlag(ctx, inst);
}
} // namespace
void EmitLaneId(EmitContext& ctx, IR::Inst& inst) {
ctx.AddU32("{}=gl_SubGroupInvocationARB&31u;", inst);
}
void EmitVoteAll(EmitContext& ctx, IR::Inst& inst, std::string_view pred) {
if (!ctx.profile.warp_size_potentially_larger_than_guest) {
ctx.AddU1("{}=allInvocationsEqualARB({});", inst, pred);
} else {
const auto active_mask{fmt::format("uvec2(ballotARB(true))[gl_SubGroupInvocationARB]")};
const auto ballot{fmt::format("uvec2(ballotARB({}))[gl_SubGroupInvocationARB]", pred)};
ctx.AddU1("{}=({}&{})=={};", inst, ballot, active_mask, active_mask);
}
}
void EmitVoteAny(EmitContext& ctx, IR::Inst& inst, std::string_view pred) {
if (!ctx.profile.warp_size_potentially_larger_than_guest) {
ctx.AddU1("{}=anyInvocationARB({});", inst, pred);
} else {
const auto active_mask{fmt::format("uvec2(ballotARB(true))[gl_SubGroupInvocationARB]")};
const auto ballot{fmt::format("uvec2(ballotARB({}))[gl_SubGroupInvocationARB]", pred)};
ctx.AddU1("{}=({}&{})!=0u;", inst, ballot, active_mask, active_mask);
}
}
void EmitVoteEqual(EmitContext& ctx, IR::Inst& inst, std::string_view pred) {
if (!ctx.profile.warp_size_potentially_larger_than_guest) {
ctx.AddU1("{}=allInvocationsEqualARB({});", inst, pred);
} else {
const auto active_mask{fmt::format("uvec2(ballotARB(true))[gl_SubGroupInvocationARB]")};
const auto ballot{fmt::format("uvec2(ballotARB({}))[gl_SubGroupInvocationARB]", pred)};
const auto value{fmt::format("({}^{})", ballot, active_mask)};
ctx.AddU1("{}=({}==0)||({}=={});", inst, value, value, active_mask);
}
}
void EmitSubgroupBallot(EmitContext& ctx, IR::Inst& inst, std::string_view pred) {
if (!ctx.profile.warp_size_potentially_larger_than_guest) {
ctx.AddU32("{}=uvec2(ballotARB({})).x;", inst, pred);
} else {
ctx.AddU32("{}=uvec2(ballotARB({}))[gl_SubGroupInvocationARB];", inst, pred);
}
}
void EmitSubgroupEqMask(EmitContext& ctx, IR::Inst& inst) {
ctx.AddU32("{}=uint(gl_SubGroupEqMaskARB.x);", inst);
}
void EmitSubgroupLtMask(EmitContext& ctx, IR::Inst& inst) {
ctx.AddU32("{}=uint(gl_SubGroupLtMaskARB.x);", inst);
}
void EmitSubgroupLeMask(EmitContext& ctx, IR::Inst& inst) {
ctx.AddU32("{}=uint(gl_SubGroupLeMaskARB.x);", inst);
}
void EmitSubgroupGtMask(EmitContext& ctx, IR::Inst& inst) {
ctx.AddU32("{}=uint(gl_SubGroupGtMaskARB.x);", inst);
}
void EmitSubgroupGeMask(EmitContext& ctx, IR::Inst& inst) {
ctx.AddU32("{}=uint(gl_SubGroupGeMaskARB.x);", inst);
}
void EmitShuffleIndex(EmitContext& ctx, IR::Inst& inst, std::string_view value,
std::string_view index, std::string_view clamp,
std::string_view segmentation_mask) {
if (ctx.profile.support_gl_warp_intrinsics) {
UseShuffleNv(ctx, inst, "shuffleNV", value, index, clamp, segmentation_mask);
return;
}
const auto not_seg_mask{fmt::format("(~{})", segmentation_mask)};
const auto thread_id{"gl_SubGroupInvocationARB"};
const auto min_thread_id{ComputeMinThreadId(thread_id, segmentation_mask)};
const auto max_thread_id{ComputeMaxThreadId(min_thread_id, clamp, not_seg_mask)};
const auto lhs{fmt::format("({}&{})", index, not_seg_mask)};
const auto src_thread_id{fmt::format("({})|({})", lhs, min_thread_id)};
ctx.Add("shfl_in_bounds=int({})<=int({});", src_thread_id, max_thread_id);
SetInBoundsFlag(ctx, inst);
ctx.AddU32("{}=shfl_in_bounds?readInvocationARB({},{}):{};", inst, value, src_thread_id, value);
}
void EmitShuffleUp(EmitContext& ctx, IR::Inst& inst, std::string_view value, std::string_view index,
std::string_view clamp, std::string_view segmentation_mask) {
if (ctx.profile.support_gl_warp_intrinsics) {
UseShuffleNv(ctx, inst, "shuffleUpNV", value, index, clamp, segmentation_mask);
return;
}
const auto thread_id{"gl_SubGroupInvocationARB"};
const auto max_thread_id{GetMaxThreadId(thread_id, clamp, segmentation_mask)};
const auto src_thread_id{fmt::format("({}-{})", thread_id, index)};
ctx.Add("shfl_in_bounds=int({})>=int({});", src_thread_id, max_thread_id);
SetInBoundsFlag(ctx, inst);
ctx.AddU32("{}=shfl_in_bounds?readInvocationARB({},{}):{};", inst, value, src_thread_id, value);
}
void EmitShuffleDown(EmitContext& ctx, IR::Inst& inst, std::string_view value,
std::string_view index, std::string_view clamp,
std::string_view segmentation_mask) {
if (ctx.profile.support_gl_warp_intrinsics) {
UseShuffleNv(ctx, inst, "shuffleDownNV", value, index, clamp, segmentation_mask);
return;
}
const auto thread_id{"gl_SubGroupInvocationARB"};
const auto max_thread_id{GetMaxThreadId(thread_id, clamp, segmentation_mask)};
const auto src_thread_id{fmt::format("({}+{})", thread_id, index)};
ctx.Add("shfl_in_bounds=int({})<=int({});", src_thread_id, max_thread_id);
SetInBoundsFlag(ctx, inst);
ctx.AddU32("{}=shfl_in_bounds?readInvocationARB({},{}):{};", inst, value, src_thread_id, value);
}
void EmitShuffleButterfly(EmitContext& ctx, IR::Inst& inst, std::string_view value,
std::string_view index, std::string_view clamp,
std::string_view segmentation_mask) {
if (ctx.profile.support_gl_warp_intrinsics) {
UseShuffleNv(ctx, inst, "shuffleXorNV", value, index, clamp, segmentation_mask);
return;
}
const auto thread_id{"gl_SubGroupInvocationARB"};
const auto max_thread_id{GetMaxThreadId(thread_id, clamp, segmentation_mask)};
const auto src_thread_id{fmt::format("({}^{})", thread_id, index)};
ctx.Add("shfl_in_bounds=int({})<=int({});", src_thread_id, max_thread_id);
SetInBoundsFlag(ctx, inst);
ctx.AddU32("{}=shfl_in_bounds?readInvocationARB({},{}):{};", inst, value, src_thread_id, value);
}
void EmitFSwizzleAdd(EmitContext& ctx, IR::Inst& inst, std::string_view op_a, std::string_view op_b,
std::string_view swizzle) {
const auto mask{fmt::format("({}>>((gl_SubGroupInvocationARB&3)<<1))&3", swizzle)};
const std::string modifier_a = fmt::format("FSWZ_A[{}]", mask);
const std::string modifier_b = fmt::format("FSWZ_B[{}]", mask);
ctx.AddF32("{}=({}*{})+({}*{});", inst, op_a, modifier_a, op_b, modifier_b);
}
void EmitDPdxFine(EmitContext& ctx, IR::Inst& inst, std::string_view op_a) {
ctx.AddF32("{}=dFdxFine({});", inst, op_a);
}
void EmitDPdyFine(EmitContext& ctx, IR::Inst& inst, std::string_view op_a) {
ctx.AddF32("{}=dFdyFine({});", inst, op_a);
}
void EmitDPdxCoarse(EmitContext& ctx, IR::Inst& inst, std::string_view op_a) {
ctx.AddF32("{}=dFdxCoarse({});", inst, op_a);
}
void EmitDPdyCoarse(EmitContext& ctx, IR::Inst& inst, std::string_view op_a) {
ctx.AddF32("{}=dFdyCoarse({});", inst, op_a);
}
} // namespace Shader::Backend::GLSL
|