Merge pull request #1409 from KhronosGroup/fix-1405

Improve expression usage tracking

reference/opt/shaders/asm/frag/loop-header-to-continue.asm.frag

@@ -22,17 +22,19 @@ layout(location = 0) out vec4 _entryPointOutput;
 
 void main()
 {
+    vec2 _45 = vec2(0.0, _8.CB1.TextureSize.w);
     vec4 _49 = texture(SPIRV_Cross_CombinedmapTexturemapSampler, IN_uv);
     float _50 = _49.y;
+    float _53 = clamp(_50 * 0.06399999558925628662109375, 7.999999797903001308441162109375e-05, 0.008000000379979610443115234375);
     float _55;
     float _58;
     _55 = 0.0;
     _58 = 0.0;
     for (int _60 = -3; _60 <= 3; )
     {
         float _64 = float(_60);
-        vec4 _72 = texture(SPIRV_Cross_CombinedmapTexturemapSampler, IN_uv + (vec2(0.0, _8.CB1.TextureSize.w) * _64));
-        float _78 = exp(((-_64) * _64) * 0.2222220003604888916015625) * float(abs(_72.y - _50) < clamp(_50 * 0.06399999558925628662109375, 7.999999797903001308441162109375e-05, 0.008000000379979610443115234375));
+        vec4 _72 = texture(SPIRV_Cross_CombinedmapTexturemapSampler, IN_uv + (_45 * _64));
+        float _78 = exp(((-_64) * _64) * 0.2222220003604888916015625) * float(abs(_72.y - _50) < _53);
         _55 += (_72.x * _78);
         _58 += _78;
         _60++;

reference/opt/shaders/frag/avoid-expression-lowering-to-loop.frag

@@ -0,0 +1,29 @@
+#version 310 es
+precision mediump float;
+precision highp int;
+
+layout(binding = 1, std140) uniform Count
+{
+    float count;
+} _44;
+
+layout(binding = 0) uniform mediump sampler2D tex;
+
+layout(location = 0) in highp vec4 vertex;
+layout(location = 0) out vec4 fragColor;
+
+void main()
+{
+    highp float _24 = 1.0 / float(textureSize(tex, 0).x);
+    highp float _34 = dFdx(vertex.x);
+    float _62;
+    _62 = 0.0;
+    for (float _61 = 0.0; _61 < _44.count; )
+    {
+        _62 += (_24 * _34);
+        _61 += 1.0;
+        continue;
+    }
+    fragColor = vec4(_62);
+}
+

reference/shaders-msl/asm/comp/variable-pointers-2.asm.comp

@@ -33,6 +33,7 @@ kernel void main0(device foo& buf [[buffer(0)]], constant bar& cb [[buffer(1)]],
 {
     device foo* _46 = select_buffer(buf, cb);
     device foo* _45 = _46;
+    thread uint3* _47 = select_input(gl_GlobalInvocationID, gl_LocalInvocationID, cb);
     device foo* _48 = _45;
     device int* _52;
     device int* _55;
@@ -46,7 +47,7 @@ kernel void main0(device foo& buf [[buffer(0)]], constant bar& cb [[buffer(1)]],
         _58 = *_55;
         if (_57 != _58)
         {
-            int _66 = (_57 + _58) + int((*select_input(gl_GlobalInvocationID, gl_LocalInvocationID, cb)).x);
+            int _66 = (_57 + _58) + int((*_47).x);
             *_52 = _66;
             *_55 = _66;
             _52 = &_52[1u];

reference/shaders-no-opt/asm/frag/inliner-dominator-inside-loop.asm.frag

@@ -125,6 +125,7 @@ void main()
     _151.UvStuds = IN_UvStuds_EdgeDistance2.xy;
     SurfaceInput _156 = _151;
     _156.UvStuds.y = (fract(_151.UvStuds.y) + IN_studIndex) * 0.25;
+    float _160 = clamp(1.0 - (_146.View_Depth.w * 0.00333332992158830165863037109375), 0.0, 1.0);
     float _163 = _146.View_Depth.w * _19.CB0.RefractionBias_FadeDistance_GlowFactor.y;
     float _165 = clamp(1.0 - _163, 0.0, 1.0);
     vec2 _166 = IN_Uv_EdgeDistance1.xy * 1.0;
@@ -141,7 +142,7 @@ void main()
         else
         {
             float _180 = 1.0 / (1.0 - 0.0);
-            _193 = mix(texture(SPIRV_Cross_CombinedDiffuseMapTextureDiffuseMapSampler, _166 * 0.25), texture(SPIRV_Cross_CombinedDiffuseMapTextureDiffuseMapSampler, _166), vec4(clamp((clamp(1.0 - (_146.View_Depth.w * 0.00333332992158830165863037109375), 0.0, 1.0) * _180) - (0.0 * _180), 0.0, 1.0)));
+            _193 = mix(texture(SPIRV_Cross_CombinedDiffuseMapTextureDiffuseMapSampler, _166 * 0.25), texture(SPIRV_Cross_CombinedDiffuseMapTextureDiffuseMapSampler, _166), vec4(clamp((_160 * _180) - (0.0 * _180), 0.0, 1.0)));
             break;
         }
         _193 = _192;
@@ -167,12 +168,14 @@ void main()
     vec2 _223 = vec2(1.0);
     vec2 _224 = (_220.wy * 2.0) - _223;
     vec3 _232 = vec3(_224, sqrt(clamp(1.0 + dot(-_224, _224), 0.0, 1.0)));
-    vec2 _240 = (texture(SPIRV_Cross_CombinedNormalDetailMapTextureNormalDetailMapSampler, _166 * 0.0).wy * 2.0) - _223;
+    vec4 _237 = texture(SPIRV_Cross_CombinedNormalDetailMapTextureNormalDetailMapSampler, _166 * 0.0);
+    vec2 _240 = (_237.wy * 2.0) - _223;
     vec2 _252 = _232.xy + (vec3(_240, sqrt(clamp(1.0 + dot(-_240, _240), 0.0, 1.0))).xy * 0.0);
     vec3 _253 = vec3(_252.x, _252.y, _232.z);
     vec2 _255 = _253.xy * _165;
     vec3 _256 = vec3(_255.x, _255.y, _253.z);
-    vec3 _271 = ((IN_Color.xyz * (_193 * 1.0).xyz) * (1.0 + (_256.x * 0.300000011920928955078125))) * (texture(SPIRV_Cross_CombinedStudsMapTextureStudsMapSampler, _156.UvStuds).x * 2.0);
+    vec4 _268 = texture(SPIRV_Cross_CombinedStudsMapTextureStudsMapSampler, _156.UvStuds);
+    vec3 _271 = ((IN_Color.xyz * (_193 * 1.0).xyz) * (1.0 + (_256.x * 0.300000011920928955078125))) * (_268.x * 2.0);
     vec4 _298;
     for (;;)
     {

reference/shaders-ue4-no-opt/asm/frag/accesschain-invalid-expression.asm.invalid.frag

@@ -216,11 +216,14 @@ fragment main0_out main0(main0_in in [[stage_in]], constant type_View& View [[bu
 {
     main0_out out = {};
     float4 _177 = float4((((gl_FragCoord.xy - View.View_ViewRectMin.xy) * View.View_ViewSizeAndInvSize.zw) - float2(0.5)) * float2(2.0, -2.0), _138, 1.0) * float4(gl_FragCoord.w);
+    float3 _179 = in.in_var_TEXCOORD8.xyz - float3(View.View_PreViewTranslation);
     float3 _181 = normalize(-in.in_var_TEXCOORD8.xyz);
-    float2 _190 = (Material_Texture2D_0.sample(Material_Texture2D_0Sampler, (in.in_var_TEXCOORD0 * float2(10.0))).xy * float2(2.0)) - float2(1.0);
+    float4 _187 = Material_Texture2D_0.sample(Material_Texture2D_0Sampler, (in.in_var_TEXCOORD0 * float2(10.0)));
+    float2 _190 = (_187.xy * float2(2.0)) - float2(1.0);
     float3 _206 = normalize(float3x3(float3(1.0, 0.0, 0.0), float3(0.0, 1.0, 0.0), float3(0.0, 0.0, 1.0)) * (((float4(_190, sqrt(fast::clamp(1.0 - dot(_190, _190), 0.0, 1.0)), 1.0).xyz * float3(0.300000011920928955078125, 0.300000011920928955078125, 1.0)) * float3(View.View_NormalOverrideParameter.w)) + View.View_NormalOverrideParameter.xyz));
     float _208 = dot(_206, _181);
-    float _219 = mix(0.4000000059604644775390625, 1.0, Material_Texture2D_1.sample(Material_Texture2D_1Sampler, (in.in_var_TEXCOORD0 * float2(20.0))).x);
+    float4 _217 = Material_Texture2D_1.sample(Material_Texture2D_1Sampler, (in.in_var_TEXCOORD0 * float2(20.0)));
+    float _219 = mix(0.4000000059604644775390625, 1.0, _217.x);
     float4 _223 = Material_Texture2D_1.sample(Material_Texture2D_1Sampler, (in.in_var_TEXCOORD0 * float2(5.0)));
     float _224 = _177.w;
     float _228 = fast::min(fast::max((_224 - 24.0) * 0.000666666659526526927947998046875, 0.0), 1.0);
@@ -328,7 +331,7 @@ fragment main0_out main0(main0_in in [[stage_in]], constant type_View& View [[bu
     {
         if (_491 < _Globals.NumDynamicPointLights)
         {
-            float3 _501 = _Globals.LightPositionAndInvRadius[_491].xyz - (in.in_var_TEXCOORD8.xyz - float3(View.View_PreViewTranslation));
+            float3 _501 = _Globals.LightPositionAndInvRadius[_491].xyz - _179;
             float _502 = dot(_501, _501);
             float3 _505 = _501 * float3(rsqrt(_502));
             _507 = normalize(_181 + _505);

reference/shaders-ue4-no-opt/asm/frag/array-copy-error.asm.invalid.frag

@@ -274,8 +274,11 @@ fragment main0_out main0(main0_in in [[stage_in]], constant type_View& View [[bu
     float3 _151 = normalize(-_148);
     float3 _152 = _151 * float3x3(in.in_var_TEXCOORD10_centroid.xyz, cross(in.in_var_TEXCOORD11_centroid.xyz, in.in_var_TEXCOORD10_centroid.xyz) * float3(in.in_var_TEXCOORD11_centroid.w), in.in_var_TEXCOORD11_centroid.xyz);
     float _170 = mix(Material.Material_ScalarExpressions[0].y, Material.Material_ScalarExpressions[0].z, fast::min(fast::max(abs(dot(_151, in.in_var_TEXCOORD11_centroid.xyz)), 0.0), 1.0));
+    float _171 = floor(_170);
     float _172 = 1.0 / _170;
     float2 _174 = (float2(Material.Material_ScalarExpressions[0].x) * ((_152.xy * float2(-1.0)) / float2(_152.z))) * float2(_172);
+    float2 _175 = dfdx(float2(in_var_TEXCOORD0[0].x, in_var_TEXCOORD0[0].y));
+    float2 _176 = dfdy(float2(in_var_TEXCOORD0[0].x, in_var_TEXCOORD0[0].y));
     float _180_copy;
     float2 _183;
     _183 = float2(0.0);
@@ -288,9 +291,9 @@ fragment main0_out main0(main0_in in [[stage_in]], constant type_View& View [[bu
     float _189 = 1.0;
     for (;;)
     {
-        if (float(_185) < (floor(_170) + 2.0))
+        if (float(_185) < (_171 + 2.0))
         {
-            _188 = Material_Texture2D_0.sample(Material_Texture2D_0Sampler, (float2(in_var_TEXCOORD0[0].x, in_var_TEXCOORD0[0].y) + _183), gradient2d(dfdx(float2(in_var_TEXCOORD0[0].x, in_var_TEXCOORD0[0].y)), dfdy(float2(in_var_TEXCOORD0[0].x, in_var_TEXCOORD0[0].y)))).y;
+            _188 = Material_Texture2D_0.sample(Material_Texture2D_0Sampler, (float2(in_var_TEXCOORD0[0].x, in_var_TEXCOORD0[0].y) + _183), gradient2d(_175, _176)).y;
             if (_180 < _188)
             {
                 float _201 = _188 - _180;

reference/shaders-ue4-no-opt/asm/frag/phi-variable-declaration.asm.invalid.frag

@@ -274,8 +274,11 @@ fragment main0_out main0(main0_in in [[stage_in]], constant type_View& View [[bu
     float3 _151 = normalize(-_148);
     float3 _152 = _151 * float3x3(in.in_var_TEXCOORD10_centroid.xyz, cross(in.in_var_TEXCOORD11_centroid.xyz, in.in_var_TEXCOORD10_centroid.xyz) * float3(in.in_var_TEXCOORD11_centroid.w), in.in_var_TEXCOORD11_centroid.xyz);
     float _170 = mix(Material.Material_ScalarExpressions[0].y, Material.Material_ScalarExpressions[0].z, fast::min(fast::max(abs(dot(_151, in.in_var_TEXCOORD11_centroid.xyz)), 0.0), 1.0));
+    float _171 = floor(_170);
     float _172 = 1.0 / _170;
     float2 _174 = (float2(Material.Material_ScalarExpressions[0].x) * ((_152.xy * float2(-1.0)) / float2(_152.z))) * float2(_172);
+    float2 _175 = dfdx(float2(in_var_TEXCOORD0[0].x, in_var_TEXCOORD0[0].y));
+    float2 _176 = dfdy(float2(in_var_TEXCOORD0[0].x, in_var_TEXCOORD0[0].y));
     float _180_copy;
     float2 _183;
     _183 = float2(0.0);
@@ -288,9 +291,9 @@ fragment main0_out main0(main0_in in [[stage_in]], constant type_View& View [[bu
     float _189 = 1.0;
     for (;;)
     {
-        if (float(_185) < (floor(_170) + 2.0))
+        if (float(_185) < (_171 + 2.0))
         {
-            _188 = Material_Texture2D_0.sample(Material_Texture2D_0Sampler, (float2(in_var_TEXCOORD0[0].x, in_var_TEXCOORD0[0].y) + _183), gradient2d(dfdx(float2(in_var_TEXCOORD0[0].x, in_var_TEXCOORD0[0].y)), dfdy(float2(in_var_TEXCOORD0[0].x, in_var_TEXCOORD0[0].y)))).y;
+            _188 = Material_Texture2D_0.sample(Material_Texture2D_0Sampler, (float2(in_var_TEXCOORD0[0].x, in_var_TEXCOORD0[0].y) + _183), gradient2d(_175, _176)).y;
             if (_180 < _188)
             {
                 float _201 = _188 - _180;

reference/shaders-ue4/asm/frag/texture-atomics.asm.frag

@@ -84,6 +84,7 @@ fragment main0_out main0(main0_in in [[stage_in]], const device type_StructuredB
     uint _107 = _103 + 1u;
     if (all(CulledObjectBoxBounds._m0[_107].xy > _96.xy) && all(CulledObjectBoxBounds._m0[_103].xyz < _102))
     {
+        float3 _121 = float3(0.5) * (CulledObjectBoxBounds._m0[_103].xyz + CulledObjectBoxBounds._m0[_107].xyz);
         float _122 = _96.x;
         float _123 = _96.y;
         spvUnsafeArray<float3, 8> _73;
@@ -103,7 +104,7 @@ fragment main0_out main0(main0_in in [[stage_in]], const device type_StructuredB
         _158 = float3(500000.0);
         for (int _160 = 0; _160 < 8; )
         {
-            float3 _166 = _73[_160] - (float3(0.5) * (CulledObjectBoxBounds._m0[_103].xyz + CulledObjectBoxBounds._m0[_107].xyz));
+            float3 _166 = _73[_160] - _121;
             float3 _170 = float3(dot(_166, CulledObjectBoxBounds._m0[_103 + 2u].xyz), dot(_166, CulledObjectBoxBounds._m0[_103 + 3u].xyz), dot(_166, CulledObjectBoxBounds._m0[_103 + 4u].xyz));
             _155 = fast::max(_155, _170);
             _158 = fast::min(_158, _170);

reference/shaders-ue4/asm/frag/texture-atomics.asm.graphics-robust-access.frag

@@ -84,6 +84,7 @@ fragment main0_out main0(main0_in in [[stage_in]], const device type_StructuredB
     uint _107 = _103 + 1u;
     if (all(CulledObjectBoxBounds._m0[_107].xy > _96.xy) && all(CulledObjectBoxBounds._m0[_103].xyz < _102))
     {
+        float3 _121 = float3(0.5) * (CulledObjectBoxBounds._m0[_103].xyz + CulledObjectBoxBounds._m0[_107].xyz);
         float _122 = _96.x;
         float _123 = _96.y;
         spvUnsafeArray<float3, 8> _73;
@@ -103,7 +104,7 @@ fragment main0_out main0(main0_in in [[stage_in]], const device type_StructuredB
         _158 = float3(500000.0);
         for (int _160 = 0; _160 < 8; )
         {
-            float3 _166 = _73[_160] - (float3(0.5) * (CulledObjectBoxBounds._m0[_103].xyz + CulledObjectBoxBounds._m0[_107].xyz));
+            float3 _166 = _73[_160] - _121;
             float3 _170 = float3(dot(_166, CulledObjectBoxBounds._m0[_103 + 2u].xyz), dot(_166, CulledObjectBoxBounds._m0[_103 + 3u].xyz), dot(_166, CulledObjectBoxBounds._m0[_103 + 4u].xyz));
             _155 = fast::max(_155, _170);
             _158 = fast::min(_158, _170);

reference/shaders/asm/frag/loop-header-to-continue.asm.frag

@@ -22,17 +22,19 @@ layout(location = 0) out vec4 _entryPointOutput;
 
 void main()
 {
+    vec2 _45 = vec2(0.0, _8.CB1.TextureSize.w);
     vec4 _49 = texture(SPIRV_Cross_CombinedmapTexturemapSampler, IN_uv);
     float _50 = _49.y;
+    float _53 = clamp((_50 * 80.0) * 0.0007999999797903001308441162109375, 7.999999797903001308441162109375e-05, 0.008000000379979610443115234375);
     float _55;
     float _58;
     _55 = 0.0;
     _58 = 0.0;
     for (int _60 = -3; _60 <= 3; )
     {
         float _64 = float(_60);
-        vec4 _72 = texture(SPIRV_Cross_CombinedmapTexturemapSampler, IN_uv + (vec2(0.0, _8.CB1.TextureSize.w) * _64));
-        float _78 = exp(((-_64) * _64) * 0.2222220003604888916015625) * float(abs(_72.y - _50) < clamp((_50 * 80.0) * 0.0007999999797903001308441162109375, 7.999999797903001308441162109375e-05, 0.008000000379979610443115234375));
+        vec4 _72 = texture(SPIRV_Cross_CombinedmapTexturemapSampler, IN_uv + (_45 * _64));
+        float _78 = exp(((-_64) * _64) * 0.2222220003604888916015625) * float(abs(_72.y - _50) < _53);
         _55 += (_72.x * _78);
         _58 += _78;
         _60++;

reference/shaders/frag/avoid-expression-lowering-to-loop.frag

@@ -0,0 +1,26 @@
+#version 310 es
+precision mediump float;
+precision highp int;
+
+layout(binding = 1, std140) uniform Count
+{
+    float count;
+} _44;
+
+layout(binding = 0) uniform mediump sampler2D tex;
+
+layout(location = 0) in highp vec4 vertex;
+layout(location = 0) out vec4 fragColor;
+
+void main()
+{
+    highp float size = 1.0 / float(textureSize(tex, 0).x);
+    float r = 0.0;
+    float d = dFdx(vertex.x);
+    for (float i = 0.0; i < _44.count; i += 1.0)
+    {
+        r += (size * d);
+    }
+    fragColor = vec4(r);
+}
+

shaders/frag/avoid-expression-lowering-to-loop.frag

@@ -0,0 +1,23 @@
+#version 310 es
+precision mediump float;
+precision mediump int;
+
+layout(binding = 0) uniform mediump sampler2D tex;
+layout(binding = 1) uniform Count
+{
+	float count;
+};
+
+layout(location = 0) in highp vec4 vertex;
+layout(location = 0) out vec4 fragColor;
+
+void main() {
+
+	highp float size = 1.0 / float(textureSize(tex, 0).x);
+	float r = 0.0;
+	float d = dFdx(vertex.x);
+	for (float i = 0.0; i < count ; i += 1.0)
+		r += size * d;
+
+	fragColor = vec4(r);
+}

spirv_common.hpp

@@ -262,6 +262,29 @@ inline std::string convert_to_string(double t, char locale_radix_point)
 	return buf;
 }
 
+template <typename T>
+struct ValueSaver
+{
+	explicit ValueSaver(T &current_)
+		: current(current_)
+		, saved(current_)
+	{
+	}
+
+	void release()
+	{
+		current = saved;
+	}
+
+	~ValueSaver()
+	{
+		release();
+	}
+
+	T &current;
+	T saved;
+};
+
 #if defined(__clang__) || defined(__GNUC__)
 #pragma GCC diagnostic pop
 #elif defined(_MSC_VER)
@@ -699,6 +722,9 @@ struct SPIRExpression : IVariant
 	// Used by access chain Store and Load since we read multiple expressions in this case.
 	SmallVector<ID> implied_read_expressions;
 
+	// The expression was emitted at a certain scope. Lets us track when an expression read means multiple reads.
+	uint32_t emitted_loop_level = 0;
+
 	SPIRV_CROSS_DECLARE_CLONE(SPIRExpression)
 };
 

spirv_cross.cpp

@@ -4674,3 +4674,8 @@ bool Compiler::flush_phi_required(BlockID from, BlockID to) const
 			return true;
 	return false;
 }
+
+void Compiler::add_loop_level()
+{
+	current_loop_level++;
+}

spirv_cross.hpp

@@ -513,9 +513,22 @@ class Compiler
 
 	SPIRFunction *current_function = nullptr;
 	SPIRBlock *current_block = nullptr;
+	uint32_t current_loop_level = 0;
 	std::unordered_set<VariableID> active_interface_variables;
 	bool check_active_interface_variables = false;
 
+	void add_loop_level();
+
+	void set_initializers(SPIRExpression &e)
+	{
+		e.emitted_loop_level = current_loop_level;
+	}
+
+	template <typename T>
+	void set_initializers(const T &)
+	{
+	}
+
 	// If our IDs are out of range here as part of opcodes, throw instead of
 	// undefined behavior.
 	template <typename T, typename... P>
@@ -524,6 +537,7 @@ class Compiler
 		ir.add_typed_id(static_cast<Types>(T::type), id);
 		auto &var = variant_set<T>(ir.ids[id], std::forward<P>(args)...);
 		var.self = id;
+		set_initializers(var);
 		return var;
 	}