summaryrefslogblamecommitdiffstats
path: root/src/shader_recompiler/backend/spirv/emit_spirv_warp.cpp
blob: cbc5b1c961eb4538e7592344a22b05b067b470a0 (plain) (tree) 
32c5483be






































































































































6c51f4963
















32c5483be

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






































































































































                                                                                                  















                                                                                         
                                     
// Copyright 2021 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.

#include "shader_recompiler/backend/spirv/emit_spirv.h"

namespace Shader::Backend::SPIRV {
namespace {
Id LargeWarpBallot(EmitContext& ctx, Id ballot) {
    const Id shift{ctx.Constant(ctx.U32[1], 5)};
    const Id local_index{ctx.OpLoad(ctx.U32[1], ctx.subgroup_local_invocation_id)};
    return ctx.OpVectorExtractDynamic(ctx.U32[1], ballot, local_index);
}

void SetInBoundsFlag(IR::Inst* inst, Id result) {
    IR::Inst* const in_bounds{inst->GetAssociatedPseudoOperation(IR::Opcode::GetInBoundsFromOp)};
    if (!in_bounds) {
        return;
    }
    in_bounds->SetDefinition(result);
    in_bounds->Invalidate();
}

Id ComputeMinThreadId(EmitContext& ctx, Id thread_id, Id segmentation_mask) {
    return ctx.OpBitwiseAnd(ctx.U32[1], thread_id, segmentation_mask);
}

Id ComputeMaxThreadId(EmitContext& ctx, Id min_thread_id, Id clamp, Id not_seg_mask) {
    return ctx.OpBitwiseOr(ctx.U32[1], min_thread_id,
                           ctx.OpBitwiseAnd(ctx.U32[1], clamp, not_seg_mask));
}

Id GetMaxThreadId(EmitContext& ctx, Id thread_id, Id clamp, Id segmentation_mask) {
    const Id not_seg_mask{ctx.OpNot(ctx.U32[1], segmentation_mask)};
    const Id min_thread_id{ComputeMinThreadId(ctx, thread_id, segmentation_mask)};
    return ComputeMaxThreadId(ctx, min_thread_id, clamp, not_seg_mask);
}

Id SelectValue(EmitContext& ctx, Id in_range, Id value, Id src_thread_id) {
    return ctx.OpSelect(ctx.U32[1], in_range,
                        ctx.OpSubgroupReadInvocationKHR(ctx.U32[1], value, src_thread_id), value);
}
} // Anonymous namespace

Id EmitVoteAll(EmitContext& ctx, Id pred) {
    if (!ctx.profile.warp_size_potentially_larger_than_guest) {
        return ctx.OpSubgroupAllKHR(ctx.U1, pred);
    }
    const Id mask_ballot{ctx.OpSubgroupBallotKHR(ctx.U32[4], ctx.true_value)};
    const Id active_mask{LargeWarpBallot(ctx, mask_ballot)};
    const Id ballot{LargeWarpBallot(ctx, ctx.OpSubgroupBallotKHR(ctx.U32[4], pred))};
    const Id lhs{ctx.OpBitwiseAnd(ctx.U32[1], ballot, active_mask)};
    return ctx.OpIEqual(ctx.U1, lhs, active_mask);
}

Id EmitVoteAny(EmitContext& ctx, Id pred) {
    if (!ctx.profile.warp_size_potentially_larger_than_guest) {
        return ctx.OpSubgroupAnyKHR(ctx.U1, pred);
    }
    const Id mask_ballot{ctx.OpSubgroupBallotKHR(ctx.U32[4], ctx.true_value)};
    const Id active_mask{LargeWarpBallot(ctx, mask_ballot)};
    const Id ballot{LargeWarpBallot(ctx, ctx.OpSubgroupBallotKHR(ctx.U32[4], pred))};
    const Id lhs{ctx.OpBitwiseAnd(ctx.U32[1], ballot, active_mask)};
    return ctx.OpINotEqual(ctx.U1, lhs, ctx.u32_zero_value);
}

Id EmitVoteEqual(EmitContext& ctx, Id pred) {
    if (!ctx.profile.warp_size_potentially_larger_than_guest) {
        return ctx.OpSubgroupAllEqualKHR(ctx.U1, pred);
    }
    const Id mask_ballot{ctx.OpSubgroupBallotKHR(ctx.U32[4], ctx.true_value)};
    const Id active_mask{LargeWarpBallot(ctx, mask_ballot)};
    const Id ballot{LargeWarpBallot(ctx, ctx.OpSubgroupBallotKHR(ctx.U32[4], pred))};
    const Id lhs{ctx.OpBitwiseXor(ctx.U32[1], ballot, active_mask)};
    return ctx.OpLogicalOr(ctx.U1, ctx.OpIEqual(ctx.U1, lhs, ctx.u32_zero_value),
                           ctx.OpIEqual(ctx.U1, lhs, active_mask));
}

Id EmitSubgroupBallot(EmitContext& ctx, Id pred) {
    const Id ballot{ctx.OpSubgroupBallotKHR(ctx.U32[4], pred)};
    if (!ctx.profile.warp_size_potentially_larger_than_guest) {
        return ctx.OpCompositeExtract(ctx.U32[1], ballot, 0U);
    }
    return LargeWarpBallot(ctx, ballot);
}

Id EmitShuffleIndex(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id clamp,
                    Id segmentation_mask) {
    const Id not_seg_mask{ctx.OpNot(ctx.U32[1], segmentation_mask)};
    const Id thread_id{ctx.OpLoad(ctx.U32[1], ctx.subgroup_local_invocation_id)};
    const Id min_thread_id{ComputeMinThreadId(ctx, thread_id, segmentation_mask)};
    const Id max_thread_id{ComputeMaxThreadId(ctx, min_thread_id, clamp, not_seg_mask)};

    const Id lhs{ctx.OpBitwiseAnd(ctx.U32[1], index, not_seg_mask)};
    const Id src_thread_id{ctx.OpBitwiseOr(ctx.U32[1], lhs, min_thread_id)};
    const Id in_range{ctx.OpSLessThanEqual(ctx.U1, src_thread_id, max_thread_id)};

    SetInBoundsFlag(inst, in_range);
    return SelectValue(ctx, in_range, value, src_thread_id);
}

Id EmitShuffleUp(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id clamp,
                 Id segmentation_mask) {
    const Id thread_id{ctx.OpLoad(ctx.U32[1], ctx.subgroup_local_invocation_id)};
    const Id max_thread_id{GetMaxThreadId(ctx, thread_id, clamp, segmentation_mask)};
    const Id src_thread_id{ctx.OpISub(ctx.U32[1], thread_id, index)};
    const Id in_range{ctx.OpSGreaterThanEqual(ctx.U1, src_thread_id, max_thread_id)};

    SetInBoundsFlag(inst, in_range);
    return SelectValue(ctx, in_range, value, src_thread_id);
}

Id EmitShuffleDown(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id clamp,
                   Id segmentation_mask) {
    const Id thread_id{ctx.OpLoad(ctx.U32[1], ctx.subgroup_local_invocation_id)};
    const Id max_thread_id{GetMaxThreadId(ctx, thread_id, clamp, segmentation_mask)};
    const Id src_thread_id{ctx.OpIAdd(ctx.U32[1], thread_id, index)};
    const Id in_range{ctx.OpSLessThanEqual(ctx.U1, src_thread_id, max_thread_id)};

    SetInBoundsFlag(inst, in_range);
    return SelectValue(ctx, in_range, value, src_thread_id);
}

Id EmitShuffleButterfly(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id clamp,
                        Id segmentation_mask) {
    const Id thread_id{ctx.OpLoad(ctx.U32[1], ctx.subgroup_local_invocation_id)};
    const Id max_thread_id{GetMaxThreadId(ctx, thread_id, clamp, segmentation_mask)};
    const Id src_thread_id{ctx.OpBitwiseXor(ctx.U32[1], thread_id, index)};
    const Id in_range{ctx.OpSLessThanEqual(ctx.U1, src_thread_id, max_thread_id)};

    SetInBoundsFlag(inst, in_range);
    return SelectValue(ctx, in_range, value, src_thread_id);
}

Id EmitFSwizzleAdd(EmitContext& ctx, Id op_a, Id op_b, Id swizzle) {
    const Id three{ctx.Constant(ctx.U32[1], 3)};
    Id mask{ctx.OpLoad(ctx.U32[1], ctx.subgroup_local_invocation_id)};
    mask = ctx.OpBitwiseAnd(ctx.U32[1], mask, three);
    mask = ctx.OpShiftLeftLogical(ctx.U32[1], mask, ctx.Constant(ctx.U32[1], 1));
    mask = ctx.OpShiftRightLogical(ctx.U32[1], swizzle, mask);
    mask = ctx.OpBitwiseAnd(ctx.U32[1], mask, three);

    const Id modifier_a{ctx.OpVectorExtractDynamic(ctx.F32[1], ctx.fswzadd_lut_a, mask)};
    const Id modifier_b{ctx.OpVectorExtractDynamic(ctx.F32[1], ctx.fswzadd_lut_b, mask)};

    const Id result_a{ctx.OpFMul(ctx.F32[1], op_a, modifier_a)};
    const Id result_b{ctx.OpFMul(ctx.F32[1], op_b, modifier_b)};
    return ctx.OpFAdd(ctx.F32[1], result_a, result_b);
}

} // namespace Shader::Backend::SPIRV
generated by cgit v1.2.3 (git 2.25.1) at 2024-07-05 09:23:42 +0200