Skip to content

Commit c5f7b55

Browse files
HansKristian-WorkHansKristian-Work
authored
Merge pull request #1284 from KhronosGroup/fix-1282
MSL: Reintroduce workarounds for arrays not being value types
2 parents f19fdb9 + 16796e9 commit c5f7b55

20 files changed

Lines changed: 881 additions & 32 deletions

‎CMakeLists.txt‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -323,7 +323,7 @@ if (SPIRV_CROSS_STATIC)
323323
endif()
324324

325325
set(spirv-cross-abi-major 0)
326-
set(spirv-cross-abi-minor 24)
326+
set(spirv-cross-abi-minor 25)
327327
set(spirv-cross-abi-patch 0)
328328

329329
if (SPIRV_CROSS_SHARED)

‎main.cpp‎

Lines changed: 6 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -522,6 +522,7 @@ struct CLIArguments
522522
bool msl_dispatch_base = false;
523523
bool msl_decoration_binding = false;
524524
bool msl_force_active_argument_buffer_resources = false;
525+
bool msl_force_native_arrays = false;
525526
bool glsl_emit_push_constant_as_ubo = false;
526527
bool glsl_emit_ubo_as_plain_uniforms = false;
527528
bool vulkan_glsl_disable_ext_samplerless_texture_functions = false;
@@ -616,6 +617,7 @@ static void print_help()
616617
"\t[--msl-inline-uniform-block <set index> <binding>]\n"
617618
"\t[--msl-decoration-binding]\n"
618619
"\t[--msl-force-active-argument-buffer-resources]\n"
620+
"\t[--msl-force-native-arrays]\n"
619621
"\t[--hlsl]\n"
620622
"\t[--reflect]\n"
621623
"\t[--shader-model]\n"
@@ -806,6 +808,7 @@ static string compile_iteration(const CLIArguments &args, std::vector<uint32_t>
806808
msl_opts.dispatch_base = args.msl_dispatch_base;
807809
msl_opts.enable_decoration_binding = args.msl_decoration_binding;
808810
msl_opts.force_active_argument_buffer_resources = args.msl_force_active_argument_buffer_resources;
811+
msl_opts.force_native_arrays = args.msl_force_native_arrays;
809812
msl_comp->set_msl_options(msl_opts);
810813
for (auto &v : args.msl_discrete_descriptor_sets)
811814
msl_comp->add_discrete_descriptor_set(v);
@@ -1164,6 +1167,9 @@ static int main_inner(int argc, char *argv[])
11641167
uint32_t binding = parser.next_uint();
11651168
args.msl_inline_uniform_blocks.push_back(make_pair(desc_set, binding));
11661169
});
1170+
cbs.add("--msl-force-native-arrays", [&args](CLIParser &) {
1171+
args.msl_force_native_arrays = true;
1172+
});
11671173
cbs.add("--extension", [&args](CLIParser &parser) { args.extensions.push_back(parser.next_string()); });
11681174
cbs.add("--rename-entry-point", [&args](CLIParser &parser) {
11691175
auto old_name = parser.next_string();

‎reference/opt/shaders-msl/comp/composite-array-initialization.force-native-array.comp‎

Lines changed: 94 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,94 @@
1+
#pragma clang diagnostic ignored "-Wmissing-prototypes"
2+
3+
#include <metal_stdlib>
4+
#include <simd/simd.h>
5+
6+
using namespace metal;
7+
8+
struct Data
9+
{
10+
float a;
11+
float b;
12+
};
13+
14+
constant float X_tmp [[function_constant(0)]];
15+
constant float X = is_function_constant_defined(X_tmp) ? X_tmp : 4.0;
16+
17+
struct Data_1
18+
{
19+
float a;
20+
float b;
21+
};
22+
23+
struct SSBO
24+
{
25+
Data_1 outdata[1];
26+
};
27+
28+
constant uint3 gl_WorkGroupSize [[maybe_unused]] = uint3(2u, 1u, 1u);
29+
30+
template<typename T, uint A>
31+
inline void spvArrayCopyFromConstantToStack1(thread T (&dst)[A], constant T (&src)[A])
32+
{
33+
for (uint i = 0; i < A; i++)
34+
{
35+
dst[i] = src[i];
36+
}
37+
}
38+
39+
template<typename T, uint A>
40+
inline void spvArrayCopyFromConstantToThreadGroup1(threadgroup T (&dst)[A], constant T (&src)[A])
41+
{
42+
for (uint i = 0; i < A; i++)
43+
{
44+
dst[i] = src[i];
45+
}
46+
}
47+
48+
template<typename T, uint A>
49+
inline void spvArrayCopyFromStackToStack1(thread T (&dst)[A], thread const T (&src)[A])
50+
{
51+
for (uint i = 0; i < A; i++)
52+
{
53+
dst[i] = src[i];
54+
}
55+
}
56+
57+
template<typename T, uint A>
58+
inline void spvArrayCopyFromStackToThreadGroup1(threadgroup T (&dst)[A], thread const T (&src)[A])
59+
{
60+
for (uint i = 0; i < A; i++)
61+
{
62+
dst[i] = src[i];
63+
}
64+
}
65+
66+
template<typename T, uint A>
67+
inline void spvArrayCopyFromThreadGroupToStack1(thread T (&dst)[A], threadgroup const T (&src)[A])
68+
{
69+
for (uint i = 0; i < A; i++)
70+
{
71+
dst[i] = src[i];
72+
}
73+
}
74+
75+
template<typename T, uint A>
76+
inline void spvArrayCopyFromThreadGroupToThreadGroup1(threadgroup T (&dst)[A], threadgroup const T (&src)[A])
77+
{
78+
for (uint i = 0; i < A; i++)
79+
{
80+
dst[i] = src[i];
81+
}
82+
}
83+
84+
kernel void main0(device SSBO& _53 [[buffer(0)]], uint3 gl_WorkGroupID [[threadgroup_position_in_grid]], uint3 gl_LocalInvocationID [[thread_position_in_threadgroup]])
85+
{
86+
Data _25[2] = { Data{ 1.0, 2.0 }, Data{ 3.0, 4.0 } };
87+
88+
Data _31[2] = { Data{ X, 2.0 }, Data{ 3.0, 5.0 } };
89+
Data data2[2];
90+
spvArrayCopyFromStackToStack1(data2, _31);
91+
_53.outdata[gl_WorkGroupID.x].a = _25[gl_LocalInvocationID.x].a + data2[gl_LocalInvocationID.x].a;
92+
_53.outdata[gl_WorkGroupID.x].b = _25[gl_LocalInvocationID.x].b + data2[gl_LocalInvocationID.x].b;
93+
}
94+

‎reference/opt/shaders-msl/comp/copy-array-of-arrays.force-native-array.comp‎

Lines changed: 20 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,20 @@
1+
#include <metal_stdlib>
2+
#include <simd/simd.h>
3+
4+
using namespace metal;
5+
6+
struct BUF
7+
{
8+
int a;
9+
float b;
10+
float c;
11+
};
12+
13+
constant uint3 gl_WorkGroupSize [[maybe_unused]] = uint3(1u);
14+
15+
kernel void main0(device BUF& o [[buffer(0)]])
16+
{
17+
o.a = 4;
18+
o.b = o.c;
19+
}
20+

‎reference/opt/shaders-msl/vert/return-array.force-native-array.vert‎

Lines changed: 22 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,22 @@
1+
#include <metal_stdlib>
2+
#include <simd/simd.h>
3+
4+
using namespace metal;
5+
6+
struct main0_out
7+
{
8+
float4 gl_Position [[position]];
9+
};
10+
11+
struct main0_in
12+
{
13+
float4 vInput1 [[attribute(1)]];
14+
};
15+
16+
vertex main0_out main0(main0_in in [[stage_in]])
17+
{
18+
main0_out out = {};
19+
out.gl_Position = float4(10.0) + in.vInput1;
20+
return out;
21+
}
22+

‎reference/shaders-msl-no-opt/vert/pass-array-by-value.force-native-array.vert‎

Lines changed: 103 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,103 @@
1+
#pragma clang diagnostic ignored "-Wmissing-prototypes"
2+
3+
#include <metal_stdlib>
4+
#include <simd/simd.h>
5+
6+
using namespace metal;
7+
8+
constant float4 _68[4] = { float4(0.0), float4(1.0), float4(2.0), float4(3.0) };
9+
10+
struct main0_out
11+
{
12+
float4 gl_Position [[position]];
13+
};
14+
15+
struct main0_in
16+
{
17+
int Index1 [[attribute(0)]];
18+
int Index2 [[attribute(1)]];
19+
};
20+
21+
template<typename T, uint A>
22+
inline void spvArrayCopyFromConstantToStack1(thread T (&dst)[A], constant T (&src)[A])
23+
{
24+
for (uint i = 0; i < A; i++)
25+
{
26+
dst[i] = src[i];
27+
}
28+
}
29+
30+
template<typename T, uint A>
31+
inline void spvArrayCopyFromConstantToThreadGroup1(threadgroup T (&dst)[A], constant T (&src)[A])
32+
{
33+
for (uint i = 0; i < A; i++)
34+
{
35+
dst[i] = src[i];
36+
}
37+
}
38+
39+
template<typename T, uint A>
40+
inline void spvArrayCopyFromStackToStack1(thread T (&dst)[A], thread const T (&src)[A])
41+
{
42+
for (uint i = 0; i < A; i++)
43+
{
44+
dst[i] = src[i];
45+
}
46+
}
47+
48+
template<typename T, uint A>
49+
inline void spvArrayCopyFromStackToThreadGroup1(threadgroup T (&dst)[A], thread const T (&src)[A])
50+
{
51+
for (uint i = 0; i < A; i++)
52+
{
53+
dst[i] = src[i];
54+
}
55+
}
56+
57+
template<typename T, uint A>
58+
inline void spvArrayCopyFromThreadGroupToStack1(thread T (&dst)[A], threadgroup const T (&src)[A])
59+
{
60+
for (uint i = 0; i < A; i++)
61+
{
62+
dst[i] = src[i];
63+
}
64+
}
65+
66+
template<typename T, uint A>
67+
inline void spvArrayCopyFromThreadGroupToThreadGroup1(threadgroup T (&dst)[A], threadgroup const T (&src)[A])
68+
{
69+
for (uint i = 0; i < A; i++)
70+
{
71+
dst[i] = src[i];
72+
}
73+
}
74+
75+
static inline __attribute__((always_inline))
76+
float4 consume_constant_arrays2(thread const float4 (&positions)[4], thread const float4 (&positions2)[4], thread int& Index1, thread int& Index2)
77+
{
78+
float4 indexable[4];
79+
spvArrayCopyFromStackToStack1(indexable, positions);
80+
float4 indexable_1[4];
81+
spvArrayCopyFromStackToStack1(indexable_1, positions2);
82+
return indexable[Index1] + indexable_1[Index2];
83+
}
84+
85+
static inline __attribute__((always_inline))
86+
float4 consume_constant_arrays(thread const float4 (&positions)[4], thread const float4 (&positions2)[4], thread int& Index1, thread int& Index2)
87+
{
88+
return consume_constant_arrays2(positions, positions2, Index1, Index2);
89+
}
90+
91+
vertex main0_out main0(main0_in in [[stage_in]])
92+
{
93+
float4 _68_array_copy[4] = { float4(0.0), float4(1.0), float4(2.0), float4(3.0) };
94+
main0_out out = {};
95+
float4 LUT2[4];
96+
LUT2[0] = float4(10.0);
97+
LUT2[1] = float4(11.0);
98+
LUT2[2] = float4(12.0);
99+
LUT2[3] = float4(13.0);
100+
out.gl_Position = consume_constant_arrays(_68_array_copy, LUT2, in.Index1, in.Index2);
101+
return out;
102+
}
103+

‎reference/shaders-msl/comp/composite-array-initialization.force-native-array.comp‎

Lines changed: 104 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,104 @@
1+
#pragma clang diagnostic ignored "-Wmissing-prototypes"
2+
3+
#include <metal_stdlib>
4+
#include <simd/simd.h>
5+
6+
using namespace metal;
7+
8+
struct Data
9+
{
10+
float a;
11+
float b;
12+
};
13+
14+
constant float X_tmp [[function_constant(0)]];
15+
constant float X = is_function_constant_defined(X_tmp) ? X_tmp : 4.0;
16+
17+
struct Data_1
18+
{
19+
float a;
20+
float b;
21+
};
22+
23+
struct SSBO
24+
{
25+
Data_1 outdata[1];
26+
};
27+
28+
constant uint3 gl_WorkGroupSize [[maybe_unused]] = uint3(2u, 1u, 1u);
29+
30+
constant Data _25[2] = { Data{ 1.0, 2.0 }, Data{ 3.0, 4.0 } };
31+
32+
template<typename T, uint A>
33+
inline void spvArrayCopyFromConstantToStack1(thread T (&dst)[A], constant T (&src)[A])
34+
{
35+
for (uint i = 0; i < A; i++)
36+
{
37+
dst[i] = src[i];
38+
}
39+
}
40+
41+
template<typename T, uint A>
42+
inline void spvArrayCopyFromConstantToThreadGroup1(threadgroup T (&dst)[A], constant T (&src)[A])
43+
{
44+
for (uint i = 0; i < A; i++)
45+
{
46+
dst[i] = src[i];
47+
}
48+
}
49+
50+
template<typename T, uint A>
51+
inline void spvArrayCopyFromStackToStack1(thread T (&dst)[A], thread const T (&src)[A])
52+
{
53+
for (uint i = 0; i < A; i++)
54+
{
55+
dst[i] = src[i];
56+
}
57+
}
58+
59+
template<typename T, uint A>
60+
inline void spvArrayCopyFromStackToThreadGroup1(threadgroup T (&dst)[A], thread const T (&src)[A])
61+
{
62+
for (uint i = 0; i < A; i++)
63+
{
64+
dst[i] = src[i];
65+
}
66+
}
67+
68+
template<typename T, uint A>
69+
inline void spvArrayCopyFromThreadGroupToStack1(thread T (&dst)[A], threadgroup const T (&src)[A])
70+
{
71+
for (uint i = 0; i < A; i++)
72+
{
73+
dst[i] = src[i];
74+
}
75+
}
76+
77+
template<typename T, uint A>
78+
inline void spvArrayCopyFromThreadGroupToThreadGroup1(threadgroup T (&dst)[A], threadgroup const T (&src)[A])
79+
{
80+
for (uint i = 0; i < A; i++)
81+
{
82+
dst[i] = src[i];
83+
}
84+
}
85+
86+
static inline __attribute__((always_inline))
87+
Data combine(thread const Data& a, thread const Data& b)
88+
{
89+
return Data{ a.a + b.a, a.b + b.b };
90+
}
91+
92+
kernel void main0(device SSBO& _53 [[buffer(0)]], uint3 gl_WorkGroupID [[threadgroup_position_in_grid]], uint3 gl_LocalInvocationID [[thread_position_in_threadgroup]])
93+
{
94+
Data data[2] = { Data{ 1.0, 2.0 }, Data{ 3.0, 4.0 } };
95+
Data _31[2] = { Data{ X, 2.0 }, Data{ 3.0, 5.0 } };
96+
Data data2[2];
97+
spvArrayCopyFromStackToStack1(data2, _31);
98+
Data param = data[gl_LocalInvocationID.x];
99+
Data param_1 = data2[gl_LocalInvocationID.x];
100+
Data _73 = combine(param, param_1);
101+
_53.outdata[gl_WorkGroupID.x].a = _73.a;
102+
_53.outdata[gl_WorkGroupID.x].b = _73.b;
103+
}
104+

0 commit comments

Comments
 (0)