Welcome to mirror list, hosted at ThFree Co, Russian Federation.

github.com/KhronosGroup/SPIRV-Cross.git - Unnamed repository; edit this file 'description' to name the repository.
summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorHans-Kristian Arntzen <post@arntzen-software.no>2019-03-21 15:52:00 +0300
committerGitHub <noreply@github.com>2019-03-21 15:52:00 +0300
commit5dacfa9dc20d0380a67cd4f84dd17cfd8f96fd5d (patch)
tree47d6ff83956eb3abc2f932135a3320162357155e
parent1b00a6e48a4d9e1bd80bedc0e8243246f0a33a72 (diff)
parent0b20180537b620e210dc098991168a9b76e428d3 (diff)
Merge pull request #898 from KhronosGroup/fix-897
GLSL: Fix some complex cases for loading arrays
-rw-r--r--reference/opt/shaders/asm/geom/unroll-glposition-load.asm.geom22
-rw-r--r--reference/opt/shaders/asm/tese/unroll-input-array-load.asm.tese17
-rw-r--r--reference/shaders/asm/geom/unroll-glposition-load.asm.geom35
-rw-r--r--reference/shaders/asm/tese/unroll-input-array-load.asm.tese48
-rw-r--r--shaders/asm/geom/unroll-glposition-load.asm.geom102
-rw-r--r--shaders/asm/tese/unroll-input-array-load.asm.tese131
-rw-r--r--spirv_glsl.cpp58
-rw-r--r--spirv_glsl.hpp1
8 files changed, 414 insertions, 0 deletions
diff --git a/reference/opt/shaders/asm/geom/unroll-glposition-load.asm.geom b/reference/opt/shaders/asm/geom/unroll-glposition-load.asm.geom
new file mode 100644
index 00000000..d1f8963f
--- /dev/null
+++ b/reference/opt/shaders/asm/geom/unroll-glposition-load.asm.geom
@@ -0,0 +1,22 @@
+#version 450
+layout(triangles) in;
+layout(max_vertices = 3, triangle_strip) out;
+
+void main()
+{
+ vec4 _35_unrolled[3];
+ for (int i = 0; i < int(3); i++)
+ {
+ _35_unrolled[i] = gl_in[i].gl_Position;
+ }
+ vec4 param[3] = _35_unrolled;
+ for (int _73 = 0; _73 < 3; )
+ {
+ gl_Position = param[_73];
+ EmitVertex();
+ _73++;
+ continue;
+ }
+ EndPrimitive();
+}
+
diff --git a/reference/opt/shaders/asm/tese/unroll-input-array-load.asm.tese b/reference/opt/shaders/asm/tese/unroll-input-array-load.asm.tese
new file mode 100644
index 00000000..731ed3fa
--- /dev/null
+++ b/reference/opt/shaders/asm/tese/unroll-input-array-load.asm.tese
@@ -0,0 +1,17 @@
+#version 450
+layout(quads) in;
+
+struct ControlPoint
+{
+ vec4 baz;
+};
+
+layout(location = 0) patch in vec4 input_foo;
+layout(location = 1) patch in vec4 input_bar;
+layout(location = 2) in ControlPoint CPData[];
+
+void main()
+{
+ gl_Position = (((input_foo + input_bar) + vec2(gl_TessCoord.xy).xyxy) + CPData[0u].baz) + CPData[3u].baz;
+}
+
diff --git a/reference/shaders/asm/geom/unroll-glposition-load.asm.geom b/reference/shaders/asm/geom/unroll-glposition-load.asm.geom
new file mode 100644
index 00000000..de437fae
--- /dev/null
+++ b/reference/shaders/asm/geom/unroll-glposition-load.asm.geom
@@ -0,0 +1,35 @@
+#version 450
+layout(triangles) in;
+layout(max_vertices = 3, triangle_strip) out;
+
+struct SceneOut
+{
+ vec4 pos;
+};
+
+void _main(vec4 positions[3], SceneOut OUT)
+{
+ for (int i = 0; i < 3; i++)
+ {
+ SceneOut o;
+ o.pos = positions[i];
+ gl_Position = o.pos;
+ EmitVertex();
+ }
+ EndPrimitive();
+}
+
+void main()
+{
+ vec4 _35_unrolled[3];
+ for (int i = 0; i < int(3); i++)
+ {
+ _35_unrolled[i] = gl_in[i].gl_Position;
+ }
+ vec4 positions[3] = _35_unrolled;
+ vec4 param[3] = positions;
+ SceneOut param_1;
+ _main(param, param_1);
+ SceneOut OUT = param_1;
+}
+
diff --git a/reference/shaders/asm/tese/unroll-input-array-load.asm.tese b/reference/shaders/asm/tese/unroll-input-array-load.asm.tese
new file mode 100644
index 00000000..34970b82
--- /dev/null
+++ b/reference/shaders/asm/tese/unroll-input-array-load.asm.tese
@@ -0,0 +1,48 @@
+#version 450
+layout(quads) in;
+
+struct HS_INPUT
+{
+ vec4 foo;
+ vec4 bar;
+};
+
+struct ControlPoint
+{
+ vec4 baz;
+};
+
+struct DS_OUTPUT
+{
+ vec4 pos;
+};
+
+layout(location = 0) patch in vec4 input_foo;
+layout(location = 1) patch in vec4 input_bar;
+layout(location = 2) in ControlPoint CPData[];
+
+DS_OUTPUT _main(HS_INPUT _input, vec2 uv, ControlPoint CPData_1[4])
+{
+ DS_OUTPUT o;
+ o.pos = (((_input.foo + _input.bar) + uv.xyxy) + CPData_1[0].baz) + CPData_1[3].baz;
+ return o;
+}
+
+void main()
+{
+ HS_INPUT _input;
+ _input.foo = input_foo;
+ _input.bar = input_bar;
+ vec2 uv = vec2(gl_TessCoord.xy);
+ ControlPoint _54_unrolled[4];
+ for (int i = 0; i < int(4); i++)
+ {
+ _54_unrolled[i] = CPData[i];
+ }
+ ControlPoint CPData_1[4] = _54_unrolled;
+ HS_INPUT param = _input;
+ vec2 param_1 = uv;
+ ControlPoint param_2[4] = CPData_1;
+ gl_Position = _main(param, param_1, param_2).pos;
+}
+
diff --git a/shaders/asm/geom/unroll-glposition-load.asm.geom b/shaders/asm/geom/unroll-glposition-load.asm.geom
new file mode 100644
index 00000000..8c10de39
--- /dev/null
+++ b/shaders/asm/geom/unroll-glposition-load.asm.geom
@@ -0,0 +1,102 @@
+; SPIR-V
+; Version: 1.0
+; Generator: Khronos Glslang Reference Front End; 7
+; Bound: 55
+; Schema: 0
+ OpCapability Geometry
+ %1 = OpExtInstImport "GLSL.std.450"
+ OpMemoryModel Logical GLSL450
+ OpEntryPoint Geometry %main "main" %OUT_pos %positions_1
+ OpExecutionMode %main Triangles
+ OpExecutionMode %main Invocations 1
+ OpExecutionMode %main OutputTriangleStrip
+ OpExecutionMode %main OutputVertices 3
+ OpSource HLSL 500
+ OpName %main "main"
+ OpName %SceneOut "SceneOut"
+ OpMemberName %SceneOut 0 "pos"
+ OpName %_main_vf4_3__struct_SceneOut_vf41_ "@main(vf4[3];struct-SceneOut-vf41;"
+ OpName %positions "positions"
+ OpName %OUT "OUT"
+ OpName %i "i"
+ OpName %o "o"
+ OpName %OUT_pos "OUT.pos"
+ OpName %positions_0 "positions"
+ OpName %positions_1 "positions"
+ OpName %OUT_0 "OUT"
+ OpName %param "param"
+ OpName %param_0 "param"
+ OpDecorate %OUT_pos BuiltIn Position
+ OpDecorate %positions_1 BuiltIn Position
+ %void = OpTypeVoid
+ %3 = OpTypeFunction %void
+ %float = OpTypeFloat 32
+ %v4float = OpTypeVector %float 4
+ %uint = OpTypeInt 32 0
+ %uint_3 = OpConstant %uint 3
+%_arr_v4float_uint_3 = OpTypeArray %v4float %uint_3
+%_ptr_Function__arr_v4float_uint_3 = OpTypePointer Function %_arr_v4float_uint_3
+ %SceneOut = OpTypeStruct %v4float
+%_ptr_Function_SceneOut = OpTypePointer Function %SceneOut
+ %14 = OpTypeFunction %void %_ptr_Function__arr_v4float_uint_3 %_ptr_Function_SceneOut
+ %int = OpTypeInt 32 1
+%_ptr_Function_int = OpTypePointer Function %int
+ %int_0 = OpConstant %int 0
+ %int_3 = OpConstant %int 3
+ %bool = OpTypeBool
+%_ptr_Function_v4float = OpTypePointer Function %v4float
+%_ptr_Output_v4float = OpTypePointer Output %v4float
+ %OUT_pos = OpVariable %_ptr_Output_v4float Output
+ %int_1 = OpConstant %int 1
+%_ptr_Input__arr_v4float_uint_3 = OpTypePointer Input %_arr_v4float_uint_3
+%positions_1 = OpVariable %_ptr_Input__arr_v4float_uint_3 Input
+ %main = OpFunction %void None %3
+ %5 = OpLabel
+%positions_0 = OpVariable %_ptr_Function__arr_v4float_uint_3 Function
+ %OUT_0 = OpVariable %_ptr_Function_SceneOut Function
+ %param = OpVariable %_ptr_Function__arr_v4float_uint_3 Function
+ %param_0 = OpVariable %_ptr_Function_SceneOut Function
+ %48 = OpLoad %_arr_v4float_uint_3 %positions_1
+ OpStore %positions_0 %48
+ %51 = OpLoad %_arr_v4float_uint_3 %positions_0
+ OpStore %param %51
+ %53 = OpFunctionCall %void %_main_vf4_3__struct_SceneOut_vf41_ %param %param_0
+ %54 = OpLoad %SceneOut %param_0
+ OpStore %OUT_0 %54
+ OpReturn
+ OpFunctionEnd
+%_main_vf4_3__struct_SceneOut_vf41_ = OpFunction %void None %14
+ %positions = OpFunctionParameter %_ptr_Function__arr_v4float_uint_3
+ %OUT = OpFunctionParameter %_ptr_Function_SceneOut
+ %18 = OpLabel
+ %i = OpVariable %_ptr_Function_int Function
+ %o = OpVariable %_ptr_Function_SceneOut Function
+ OpStore %i %int_0
+ OpBranch %23
+ %23 = OpLabel
+ OpLoopMerge %25 %26 None
+ OpBranch %27
+ %27 = OpLabel
+ %28 = OpLoad %int %i
+ %31 = OpSLessThan %bool %28 %int_3
+ OpBranchConditional %31 %24 %25
+ %24 = OpLabel
+ %33 = OpLoad %int %i
+ %35 = OpAccessChain %_ptr_Function_v4float %positions %33
+ %36 = OpLoad %v4float %35
+ %37 = OpAccessChain %_ptr_Function_v4float %o %int_0
+ OpStore %37 %36
+ %40 = OpAccessChain %_ptr_Function_v4float %o %int_0
+ %41 = OpLoad %v4float %40
+ OpStore %OUT_pos %41
+ OpEmitVertex
+ OpBranch %26
+ %26 = OpLabel
+ %42 = OpLoad %int %i
+ %44 = OpIAdd %int %42 %int_1
+ OpStore %i %44
+ OpBranch %23
+ %25 = OpLabel
+ OpEndPrimitive
+ OpReturn
+ OpFunctionEnd
diff --git a/shaders/asm/tese/unroll-input-array-load.asm.tese b/shaders/asm/tese/unroll-input-array-load.asm.tese
new file mode 100644
index 00000000..960b8fa2
--- /dev/null
+++ b/shaders/asm/tese/unroll-input-array-load.asm.tese
@@ -0,0 +1,131 @@
+; SPIR-V
+; Version: 1.0
+; Generator: Khronos Glslang Reference Front End; 7
+; Bound: 79
+; Schema: 0
+ OpCapability Tessellation
+ %1 = OpExtInstImport "GLSL.std.450"
+ OpMemoryModel Logical GLSL450
+ OpEntryPoint TessellationEvaluation %main "main" %input_foo %input_bar %uv_1 %CPData_1 %_entryPointOutput_pos
+ OpExecutionMode %main Quads
+ OpSource HLSL 500
+ OpName %main "main"
+ OpName %HS_INPUT "HS_INPUT"
+ OpMemberName %HS_INPUT 0 "foo"
+ OpMemberName %HS_INPUT 1 "bar"
+ OpName %ControlPoint "ControlPoint"
+ OpMemberName %ControlPoint 0 "baz"
+ OpName %DS_OUTPUT "DS_OUTPUT"
+ OpMemberName %DS_OUTPUT 0 "pos"
+ OpName %_main_struct_HS_INPUT_vf4_vf41_vf2_struct_ControlPoint_vf41_4__ "@main(struct-HS_INPUT-vf4-vf41;vf2;struct-ControlPoint-vf41[4];"
+ OpName %input "input"
+ OpName %uv "uv"
+ OpName %CPData "CPData"
+ OpName %o "o"
+ OpName %input_0 "input"
+ OpName %input_foo "input.foo"
+ OpName %input_bar "input.bar"
+ OpName %uv_0 "uv"
+ OpName %uv_1 "uv"
+ OpName %CPData_0 "CPData"
+ OpName %CPData_1 "CPData"
+ OpName %_entryPointOutput_pos "@entryPointOutput.pos"
+ OpName %param "param"
+ OpName %param_0 "param"
+ OpName %param_1 "param"
+ OpDecorate %input_foo Patch
+ OpDecorate %input_foo Location 0
+ OpDecorate %input_bar Patch
+ OpDecorate %input_bar Location 1
+ OpDecorate %uv_1 Patch
+ OpDecorate %uv_1 BuiltIn TessCoord
+ OpDecorate %CPData_1 Location 2
+ OpDecorate %_entryPointOutput_pos BuiltIn Position
+ %void = OpTypeVoid
+ %3 = OpTypeFunction %void
+ %float = OpTypeFloat 32
+ %v4float = OpTypeVector %float 4
+ %HS_INPUT = OpTypeStruct %v4float %v4float
+%_ptr_Function_HS_INPUT = OpTypePointer Function %HS_INPUT
+ %v2float = OpTypeVector %float 2
+%_ptr_Function_v2float = OpTypePointer Function %v2float
+%ControlPoint = OpTypeStruct %v4float
+ %uint = OpTypeInt 32 0
+ %uint_4 = OpConstant %uint 4
+%_arr_ControlPoint_uint_4 = OpTypeArray %ControlPoint %uint_4
+%_ptr_Function__arr_ControlPoint_uint_4 = OpTypePointer Function %_arr_ControlPoint_uint_4
+ %DS_OUTPUT = OpTypeStruct %v4float
+ %18 = OpTypeFunction %DS_OUTPUT %_ptr_Function_HS_INPUT %_ptr_Function_v2float %_ptr_Function__arr_ControlPoint_uint_4
+%_ptr_Function_DS_OUTPUT = OpTypePointer Function %DS_OUTPUT
+ %int = OpTypeInt 32 1
+ %int_0 = OpConstant %int 0
+%_ptr_Function_v4float = OpTypePointer Function %v4float
+ %int_1 = OpConstant %int 1
+ %int_3 = OpConstant %int 3
+%_ptr_Input_v4float = OpTypePointer Input %v4float
+ %input_foo = OpVariable %_ptr_Input_v4float Input
+ %input_bar = OpVariable %_ptr_Input_v4float Input
+ %v3float = OpTypeVector %float 3
+%_ptr_Input_v3float = OpTypePointer Input %v3float
+ %uv_1 = OpVariable %_ptr_Input_v3float Input
+%_ptr_Input__arr_ControlPoint_uint_4 = OpTypePointer Input %_arr_ControlPoint_uint_4
+ %CPData_1 = OpVariable %_ptr_Input__arr_ControlPoint_uint_4 Input
+%_ptr_Output_v4float = OpTypePointer Output %v4float
+%_entryPointOutput_pos = OpVariable %_ptr_Output_v4float Output
+ %main = OpFunction %void None %3
+ %5 = OpLabel
+ %input_0 = OpVariable %_ptr_Function_HS_INPUT Function
+ %uv_0 = OpVariable %_ptr_Function_v2float Function
+ %CPData_0 = OpVariable %_ptr_Function__arr_ControlPoint_uint_4 Function
+ %param = OpVariable %_ptr_Function_HS_INPUT Function
+ %param_0 = OpVariable %_ptr_Function_v2float Function
+ %param_1 = OpVariable %_ptr_Function__arr_ControlPoint_uint_4 Function
+ %52 = OpLoad %v4float %input_foo
+ %53 = OpAccessChain %_ptr_Function_v4float %input_0 %int_0
+ OpStore %53 %52
+ %55 = OpLoad %v4float %input_bar
+ %56 = OpAccessChain %_ptr_Function_v4float %input_0 %int_1
+ OpStore %56 %55
+ %61 = OpLoad %v3float %uv_1
+ %62 = OpCompositeExtract %float %61 0
+ %63 = OpCompositeExtract %float %61 1
+ %64 = OpCompositeConstruct %v2float %62 %63
+ OpStore %uv_0 %64
+ %68 = OpLoad %_arr_ControlPoint_uint_4 %CPData_1
+ OpStore %CPData_0 %68
+ %72 = OpLoad %HS_INPUT %input_0
+ OpStore %param %72
+ %74 = OpLoad %v2float %uv_0
+ OpStore %param_0 %74
+ %76 = OpLoad %_arr_ControlPoint_uint_4 %CPData_0
+ OpStore %param_1 %76
+ %77 = OpFunctionCall %DS_OUTPUT %_main_struct_HS_INPUT_vf4_vf41_vf2_struct_ControlPoint_vf41_4__ %param %param_0 %param_1
+ %78 = OpCompositeExtract %v4float %77 0
+ OpStore %_entryPointOutput_pos %78
+ OpReturn
+ OpFunctionEnd
+%_main_struct_HS_INPUT_vf4_vf41_vf2_struct_ControlPoint_vf41_4__ = OpFunction %DS_OUTPUT None %18
+ %input = OpFunctionParameter %_ptr_Function_HS_INPUT
+ %uv = OpFunctionParameter %_ptr_Function_v2float
+ %CPData = OpFunctionParameter %_ptr_Function__arr_ControlPoint_uint_4
+ %23 = OpLabel
+ %o = OpVariable %_ptr_Function_DS_OUTPUT Function
+ %29 = OpAccessChain %_ptr_Function_v4float %input %int_0
+ %30 = OpLoad %v4float %29
+ %32 = OpAccessChain %_ptr_Function_v4float %input %int_1
+ %33 = OpLoad %v4float %32
+ %34 = OpFAdd %v4float %30 %33
+ %35 = OpLoad %v2float %uv
+ %36 = OpVectorShuffle %v4float %35 %35 0 1 0 1
+ %37 = OpFAdd %v4float %34 %36
+ %38 = OpAccessChain %_ptr_Function_v4float %CPData %int_0 %int_0
+ %39 = OpLoad %v4float %38
+ %40 = OpFAdd %v4float %37 %39
+ %42 = OpAccessChain %_ptr_Function_v4float %CPData %int_3 %int_0
+ %43 = OpLoad %v4float %42
+ %44 = OpFAdd %v4float %40 %43
+ %45 = OpAccessChain %_ptr_Function_v4float %o %int_0
+ OpStore %45 %44
+ %46 = OpLoad %DS_OUTPUT %o
+ OpReturnValue %46
+ OpFunctionEnd
diff --git a/spirv_glsl.cpp b/spirv_glsl.cpp
index e8260b1d..d04adae4 100644
--- a/spirv_glsl.cpp
+++ b/spirv_glsl.cpp
@@ -6909,6 +6909,11 @@ void CompilerGLSL::emit_instruction(const Instruction &instruction)
// We might need to bitcast in order to load from a builtin.
bitcast_from_builtin_load(ptr, expr, get<SPIRType>(result_type));
+ // We might be trying to load a gl_Position[N], where we should be
+ // doing float4[](gl_in[i].gl_Position, ...) instead.
+ // Similar workarounds are required for input arrays in tessellation.
+ unroll_array_from_complex_load(id, ptr, expr);
+
if (ptr_expression)
ptr_expression->need_transpose = old_need_transpose;
@@ -11041,6 +11046,59 @@ void CompilerGLSL::emit_array_copy(const string &lhs, uint32_t rhs_id)
statement(lhs, " = ", to_expression(rhs_id), ";");
}
+void CompilerGLSL::unroll_array_from_complex_load(uint32_t target_id, uint32_t source_id, std::string &expr)
+{
+ if (!backend.force_gl_in_out_block)
+ return;
+ // This path is only relevant for GL backends.
+
+ auto *var = maybe_get<SPIRVariable>(source_id);
+ if (!var)
+ return;
+
+ if (var->storage != StorageClassInput)
+ return;
+
+ auto &type = get_variable_data_type(*var);
+ if (type.array.empty())
+ return;
+
+ auto builtin = BuiltIn(get_decoration(var->self, DecorationBuiltIn));
+ bool is_builtin = is_builtin_variable(*var) && (builtin == BuiltInPointSize || builtin == BuiltInPosition);
+ bool is_tess = is_tessellation_shader();
+
+ // Tessellation input arrays are special in that they are unsized, so we cannot directly copy from it.
+ // We must unroll the array load.
+ // For builtins, we couldn't catch this case normally,
+ // because this is resolved in the OpAccessChain in most cases.
+ // If we load the entire array, we have no choice but to unroll here.
+ if (is_builtin || is_tess)
+ {
+ auto new_expr = join("_", target_id, "_unrolled");
+ statement(variable_decl(type, new_expr, target_id), ";");
+ string array_expr;
+ if (type.array_size_literal.front())
+ {
+ array_expr = convert_to_string(type.array.front());
+ if (type.array.front() == 0)
+ SPIRV_CROSS_THROW("Cannot unroll an array copy from unsized array.");
+ }
+ else
+ array_expr = to_expression(type.array.front());
+
+ // The array size might be a specialization constant, so use a for-loop instead.
+ statement("for (int i = 0; i < int(", array_expr, "); i++)");
+ begin_scope();
+ if (is_builtin)
+ statement(new_expr, "[i] = gl_in[i].", expr, ";");
+ else
+ statement(new_expr, "[i] = ", expr, "[i];");
+ end_scope();
+
+ expr = move(new_expr);
+ }
+}
+
void CompilerGLSL::bitcast_from_builtin_load(uint32_t source_id, std::string &expr,
const spirv_cross::SPIRType &expr_type)
{
diff --git a/spirv_glsl.hpp b/spirv_glsl.hpp
index e105c948..57576cf2 100644
--- a/spirv_glsl.hpp
+++ b/spirv_glsl.hpp
@@ -619,6 +619,7 @@ protected:
// Sometimes we will need to automatically perform bitcasts on load and store to make this work.
virtual void bitcast_to_builtin_store(uint32_t target_id, std::string &expr, const SPIRType &expr_type);
virtual void bitcast_from_builtin_load(uint32_t source_id, std::string &expr, const SPIRType &expr_type);
+ void unroll_array_from_complex_load(uint32_t target_id, uint32_t source_id, std::string &expr);
void handle_store_to_invariant_variable(uint32_t store_id, uint32_t value_id);
void disallow_forwarding_in_expression_chain(const SPIRExpression &expr);