Merge pull request #715 from KhronosGroup/fix-709

Use GL_EXT_samplerless_texture_functions in Vulkan GLSL.

main.cpp

@@ -832,7 +832,8 @@ static int main_inner(int argc, char *argv[])
 	else
 	{
 		combined_image_samplers = !args.vulkan_semantics;
-		build_dummy_sampler = true;
+		if (!args.vulkan_semantics)
+			build_dummy_sampler = true;
 		compiler = unique_ptr<CompilerGLSL>(new CompilerGLSL(read_spirv_file(args.input)));
 	}
 

reference/opt/shaders/asm/frag/image-fetch-no-sampler.asm.vk.frag.vk

@@ -1,14 +1,14 @@
 #version 450
+#extension GL_EXT_samplerless_texture_functions : require
 
 layout(set = 0, binding = 0) uniform sampler Sampler;
 layout(set = 0, binding = 0) uniform texture2D SampledImage;
-layout(set = 0, binding = 0) uniform sampler SPIRV_Cross_DummySampler;
 
 layout(location = 0) out vec4 _entryPointOutput;
 
 void main()
 {
     ivec2 _152 = ivec3(int(gl_FragCoord.x * 1280.0), int(gl_FragCoord.y * 720.0), 0).xy;
-    _entryPointOutput = ((texelFetch(sampler2D(SampledImage, SPIRV_Cross_DummySampler), _152, 0) + texelFetch(sampler2D(SampledImage, SPIRV_Cross_DummySampler), _152, 0)) + texture(sampler2D(SampledImage, Sampler), gl_FragCoord.xy)) + texture(sampler2D(SampledImage, Sampler), gl_FragCoord.xy);
+    _entryPointOutput = ((texelFetch(SampledImage, _152, 0) + texelFetch(SampledImage, _152, 0)) + texture(sampler2D(SampledImage, Sampler), gl_FragCoord.xy)) + texture(sampler2D(SampledImage, Sampler), gl_FragCoord.xy);
 }
 

reference/shaders/asm/frag/image-fetch-no-sampler.asm.vk.frag.vk

@@ -1,14 +1,14 @@
 #version 450
+#extension GL_EXT_samplerless_texture_functions : require
 
 layout(set = 0, binding = 0) uniform sampler Sampler;
 layout(set = 0, binding = 0) uniform texture2D SampledImage;
-layout(set = 0, binding = 0) uniform sampler SPIRV_Cross_DummySampler;
 
 layout(location = 0) out vec4 _entryPointOutput;
 
 vec4 sample_fetch(texture2D tex, ivec3 UV)
 {
-    return texelFetch(sampler2D(tex, SPIRV_Cross_DummySampler), UV.xy, UV.z);
+    return texelFetch(tex, UV.xy, UV.z);
 }
 
 vec4 sample_sampler(texture2D tex, vec2 UV)
@@ -21,7 +21,7 @@ vec4 _main(vec4 xIn)
     ivec3 coord = ivec3(int(xIn.x * 1280.0), int(xIn.y * 720.0), 0);
     ivec3 param = coord;
     vec4 value = sample_fetch(SampledImage, param);
-    value += texelFetch(sampler2D(SampledImage, SPIRV_Cross_DummySampler), coord.xy, coord.z);
+    value += texelFetch(SampledImage, coord.xy, coord.z);
     vec2 param_1 = xIn.xy;
     value += sample_sampler(SampledImage, param_1);
     value += texture(sampler2D(SampledImage, Sampler), xIn.xy);

reference/shaders/asm/frag/image-query-no-sampler.vk.asm.frag.vk

@@ -1,14 +1,14 @@
 #version 450
+#extension GL_EXT_samplerless_texture_functions : require
 
 layout(set = 0, binding = 0) uniform texture2D uSampler2D;
 layout(set = 0, binding = 0) uniform texture2DMS uSampler2DMS;
-layout(set = 0, binding = 0) uniform sampler SPIRV_Cross_DummySampler;
 
 void main()
 {
-    ivec2 b = textureSize(sampler2D(uSampler2D, SPIRV_Cross_DummySampler), 0);
+    ivec2 b = textureSize(uSampler2D, 0);
-    ivec2 c = textureSize(sampler2DMS(uSampler2DMS, SPIRV_Cross_DummySampler));
+    ivec2 c = textureSize(uSampler2DMS);
-    int l1 = textureQueryLevels(sampler2D(uSampler2D, SPIRV_Cross_DummySampler));
+    int l1 = textureQueryLevels(uSampler2D);
-    int s0 = textureSamples(sampler2DMS(uSampler2DMS, SPIRV_Cross_DummySampler));
+    int s0 = textureSamples(uSampler2DMS);
 }
 

spirv_cross.cpp

@@ -1098,8 +1098,8 @@ const SPIRType &Compiler::get_non_pointer_type(uint32_t type_id) const
 
 bool Compiler::is_sampled_image_type(const SPIRType &type)
 {
-	return (type.basetype == SPIRType::Image || type.basetype == SPIRType::SampledImage) &&
+	return (type.basetype == SPIRType::Image || type.basetype == SPIRType::SampledImage) && type.image.sampled == 1 &&
-		type.image.sampled == 1 && type.image.dim != DimBuffer;
+	       type.image.dim != DimBuffer;
 }
 
 void Compiler::set_member_decoration_string(uint32_t id, uint32_t index, spv::Decoration decoration,

spirv_glsl.cpp

@@ -4083,30 +4083,32 @@ string CompilerGLSL::to_function_name(uint32_t tex, const SPIRType &imgtype, boo
 	return is_legacy() ? legacy_tex_op(fname, imgtype, lod, tex) : fname;
 }
 
-std::string CompilerGLSL::convert_separate_image_to_combined(uint32_t id)
+std::string CompilerGLSL::convert_separate_image_to_expression(uint32_t id)
 {
-	auto &imgtype = expression_type(id);
 	auto *var = maybe_get_backing_variable(id);
 
-	// If we are fetching from a plain OpTypeImage, we must combine with a dummy sampler.
+	// If we are fetching from a plain OpTypeImage, we must combine with a dummy sampler in GLSL.
+	// In Vulkan GLSL, we can make use of the newer GL_EXT_samplerless_texture_functions.
 	if (var)
 	{
 		auto &type = get<SPIRType>(var->basetype);
 		if (type.basetype == SPIRType::Image && type.image.sampled == 1 && type.image.dim != DimBuffer)
 		{
-			if (!dummy_sampler_id)
-				SPIRV_CROSS_THROW(
-				    "Cannot find dummy sampler ID. Was build_dummy_sampler_for_combined_images() called?");
-
 			if (options.vulkan_semantics)
 			{
-				auto sampled_type = imgtype;
+				// Newer glslang supports this extension to deal with texture2D as argument to texture functions.
-				sampled_type.basetype = SPIRType::SampledImage;
+				if (dummy_sampler_id)
-				return join(type_to_glsl(sampled_type), "(", to_expression(id), ", ", to_expression(dummy_sampler_id),
+					SPIRV_CROSS_THROW("Vulkan GLSL should not have a dummy sampler for combining.");
-				            ")");
+				require_extension_internal("GL_EXT_samplerless_texture_functions");
 			}
 			else
+			{
+				if (!dummy_sampler_id)
+					SPIRV_CROSS_THROW(
+					    "Cannot find dummy sampler ID. Was build_dummy_sampler_for_combined_images() called?");
+
 				return to_combined_image_sampler(id, dummy_sampler_id);
+			}
 		}
 	}
 
@@ -4121,7 +4123,7 @@ string CompilerGLSL::to_function_args(uint32_t img, const SPIRType &imgtype, boo
 {
 	string farg_str;
 	if (is_fetch)
-		farg_str = convert_separate_image_to_combined(img);
+		farg_str = convert_separate_image_to_expression(img);
 	else
 		farg_str = to_expression(img);
 
@@ -7526,7 +7528,7 @@ void CompilerGLSL::emit_instruction(const Instruction &instruction)
 		if (options.es)
 			SPIRV_CROSS_THROW("textureQueryLevels not supported in ES profile.");
 
-		auto expr = join("textureQueryLevels(", convert_separate_image_to_combined(ops[2]), ")");
+		auto expr = join("textureQueryLevels(", convert_separate_image_to_expression(ops[2]), ")");
 		auto &restype = get<SPIRType>(ops[0]);
 		expr = bitcast_expression(restype, SPIRType::Int, expr);
 		emit_op(result_type, id, expr, true);
@@ -7543,7 +7545,7 @@ void CompilerGLSL::emit_instruction(const Instruction &instruction)
 		if (type.image.sampled == 2)
 			expr = join("imageSamples(", to_expression(ops[2]), ")");
 		else
-			expr = join("textureSamples(", convert_separate_image_to_combined(ops[2]), ")");
+			expr = join("textureSamples(", convert_separate_image_to_expression(ops[2]), ")");
 
 		auto &restype = get<SPIRType>(ops[0]);
 		expr = bitcast_expression(restype, SPIRType::Int, expr);
@@ -7564,7 +7566,7 @@ void CompilerGLSL::emit_instruction(const Instruction &instruction)
 		uint32_t result_type = ops[0];
 		uint32_t id = ops[1];
 
-		auto expr = join("textureSize(", convert_separate_image_to_combined(ops[2]), ", ",
+		auto expr = join("textureSize(", convert_separate_image_to_expression(ops[2]), ", ",
 		                 bitcast_expression(SPIRType::Int, ops[3]), ")");
 		auto &restype = get<SPIRType>(ops[0]);
 		expr = bitcast_expression(restype, SPIRType::Int, expr);
@@ -7783,7 +7785,7 @@ void CompilerGLSL::emit_instruction(const Instruction &instruction)
 			else
 			{
 				// This path is hit for samplerBuffers and multisampled images which do not have LOD.
-				expr = join("textureSize(", convert_separate_image_to_combined(ops[2]), ")");
+				expr = join("textureSize(", convert_separate_image_to_expression(ops[2]), ")");
 			}
 
 			auto &restype = get<SPIRType>(ops[0]);
@@ -9282,8 +9284,7 @@ void CompilerGLSL::branch(uint32_t from, uint32_t to)
 		// Only sensible solution is to make a ladder variable, which we declare at the top of the switch block,
 		// write to the ladder here, and defer the break.
 		// The loop we're breaking out of must dominate the switch block, or there is no ladder breaking case.
-		if (current_emitting_switch && is_loop_break(to) &&
+		if (current_emitting_switch && is_loop_break(to) && current_emitting_switch->loop_dominator != -1u &&
-		    current_emitting_switch->loop_dominator != -1u &&
 		    get<SPIRBlock>(current_emitting_switch->loop_dominator).merge_block == to)
 		{
 			if (!current_emitting_switch->need_ladder_break)

spirv_glsl.hpp

@@ -574,7 +574,7 @@ protected:
 	std::string convert_float_to_string(const SPIRConstant &value, uint32_t col, uint32_t row);
 	std::string convert_double_to_string(const SPIRConstant &value, uint32_t col, uint32_t row);
 
-	std::string convert_separate_image_to_combined(uint32_t id);
+	std::string convert_separate_image_to_expression(uint32_t id);
 
 	// Builtins in GLSL are always specific signedness, but the SPIR-V can declare them
 	// as either unsigned or signed.

spirv_msl.cpp

@@ -3161,8 +3161,7 @@ string CompilerMSL::to_function_args(uint32_t img, const SPIRType &imgtype, bool
 		farg_str += to_expression(sample);
 	}
 
-	if (msl_options.swizzle_texture_samples && is_sampled_image_type(imgtype) &&
+	if (msl_options.swizzle_texture_samples && is_sampled_image_type(imgtype) && (!is_gather || !imgtype.image.depth))
-	    (!is_gather || !imgtype.image.depth))
 	{
 		// Add the swizzle constant from the swizzle buffer.
 		if (!is_gather)
@@ -4687,8 +4686,8 @@ bool CompilerMSL::SampledImageScanner::handle(spv::Op opcode, const uint32_t *ar
 	case OpImageSampleProjDrefImplicitLod:
 	case OpImageFetch:
 	case OpImageGather:
-		compiler.has_sampled_images = compiler.has_sampled_images ||
+		compiler.has_sampled_images =
-			compiler.is_sampled_image_type(compiler.expression_type(args[2]));
+		    compiler.has_sampled_images || compiler.is_sampled_image_type(compiler.expression_type(args[2]));
 		break;
 	default:
 		break;