Merge pull request #1495 from cdavis5e/1d-2d-offset-grad

MSL: Handle Offset and Grad operands for 1D-as-2D textures.

main.cpp

@@ -565,6 +565,7 @@ struct CLIArguments
 	bool msl_arrayed_subpass_input = false;
 	uint32_t msl_r32ui_linear_texture_alignment = 4;
 	uint32_t msl_r32ui_alignment_constant_id = 65535;
+	bool msl_texture_1d_as_2d = false;
 	bool glsl_emit_push_constant_as_ubo = false;
 	bool glsl_emit_ubo_as_plain_uniforms = false;
 	bool glsl_force_flattened_io_blocks = false;
@@ -774,7 +775,9 @@ static void print_help_msl()
 	                "\t[--msl-r32ui-linear-texture-align <alignment>]:\n\t\tThe required alignment of linear textures of format MTLPixelFormatR32Uint.\n"
 	                "\t\tThis is used to align the row stride for atomic accesses to such images.\n"
 	                "\t[--msl-r32ui-linear-texture-align-constant-id <id>]:\n\t\tThe function constant ID to use for the linear texture alignment.\n"
-	                "\t\tOn MSL 1.2 or later, you can override the alignment by setting this function constant.\n");
+	                "\t\tOn MSL 1.2 or later, you can override the alignment by setting this function constant.\n"
+	                "\t[--msl-texture-1d-as-2d]:\n\t\tEmit Image variables of dimension Dim1D as texture2d.\n"
+	                "\t\tIn Metal, 1D textures do not support all features that 2D textures do. Use this option if your code relies on these features.\n");
 	// clang-format on
 }
 
@@ -1015,6 +1018,7 @@ static string compile_iteration(const CLIArguments &args, std::vector<uint32_t>
 		msl_opts.arrayed_subpass_input = args.msl_arrayed_subpass_input;
 		msl_opts.r32ui_linear_texture_alignment = args.msl_r32ui_linear_texture_alignment;
 		msl_opts.r32ui_alignment_constant_id = args.msl_r32ui_alignment_constant_id;
+		msl_opts.texture_1D_as_2D = args.msl_texture_1d_as_2d;
 		msl_comp->set_msl_options(msl_opts);
 		for (auto &v : args.msl_discrete_descriptor_sets)
 			msl_comp->add_discrete_descriptor_set(v);
@@ -1439,6 +1443,7 @@ static int main_inner(int argc, char *argv[])
 	        [&args](CLIParser &parser) { args.msl_r32ui_linear_texture_alignment = parser.next_uint(); });
 	cbs.add("--msl-r32ui-linear-texture-align-constant-id",
 	        [&args](CLIParser &parser) { args.msl_r32ui_alignment_constant_id = parser.next_uint(); });
+	cbs.add("--msl-texture-1d-as-2d", [&args](CLIParser &) { args.msl_texture_1d_as_2d = true; });
 	cbs.add("--extension", [&args](CLIParser &parser) { args.extensions.push_back(parser.next_string()); });
 	cbs.add("--rename-entry-point", [&args](CLIParser &parser) {
 		auto old_name = parser.next_string();

reference/opt/shaders-msl/frag/bitcasting.1d-as-2d.frag

@@ -0,0 +1,26 @@
+#include <metal_stdlib>
+#include <simd/simd.h>
+
+using namespace metal;
+
+struct main0_out
+{
+    float4 FragColor0 [[color(0)]];
+    float4 FragColor1 [[color(1)]];
+};
+
+struct main0_in
+{
+    float4 VertGeom [[user(locn0)]];
+};
+
+fragment main0_out main0(main0_in in [[stage_in]], texture2d<float> TextureBase [[texture(0)]], texture2d<float> TextureDetail [[texture(1)]], sampler TextureBaseSmplr [[sampler(0)]], sampler TextureDetailSmplr [[sampler(1)]])
+{
+    main0_out out = {};
+    float4 _22 = TextureBase.sample(TextureBaseSmplr, float2(in.VertGeom.x, 0.5));
+    float4 _30 = TextureDetail.sample(TextureDetailSmplr, float2(in.VertGeom.x, 0.5), int2(3, 0));
+    out.FragColor0 = as_type<float4>(as_type<int4>(_22)) * as_type<float4>(as_type<int4>(_30));
+    out.FragColor1 = as_type<float4>(as_type<uint4>(_22)) * as_type<float4>(as_type<uint4>(_30));
+    return out;
+}
+

reference/opt/shaders-msl/frag/sampler-1d-lod.1d-as-2d.frag

@@ -0,0 +1,22 @@
+#include <metal_stdlib>
+#include <simd/simd.h>
+
+using namespace metal;
+
+struct main0_out
+{
+    float4 FragColor [[color(0)]];
+};
+
+struct main0_in
+{
+    float vTex [[user(locn0), flat]];
+};
+
+fragment main0_out main0(main0_in in [[stage_in]], texture2d<float> uSampler [[texture(0)]], sampler uSamplerSmplr [[sampler(0)]])
+{
+    main0_out out = {};
+    out.FragColor += ((uSampler.sample(uSamplerSmplr, float2(in.vTex, 0.5), bias(2.0)) + uSampler.sample(uSamplerSmplr, float2(in.vTex, 0.5), level(3.0))) + uSampler.sample(uSamplerSmplr, float2(in.vTex, 0.5), gradient2d(5.0, 8.0)));
+    return out;
+}
+

reference/opt/shaders-msl/frag/texel-fetch-offset.1d-as-2d.frag

@@ -0,0 +1,18 @@
+#include <metal_stdlib>
+#include <simd/simd.h>
+
+using namespace metal;
+
+struct main0_out
+{
+    float4 FragColor [[color(0)]];
+};
+
+fragment main0_out main0(texture2d<float> uTexture [[texture(0)]], texture2d<float> uTexture2 [[texture(1)]], sampler uTextureSmplr [[sampler(0)]], sampler uTexture2Smplr [[sampler(1)]], float4 gl_FragCoord [[position]])
+{
+    main0_out out = {};
+    out.FragColor = uTexture.read(uint2(int2(gl_FragCoord.xy)) + uint2(int2(1)), 0);
+    out.FragColor += uTexture2.read(uint2(uint(int(gl_FragCoord.x)), 0) + uint2(uint(-1), 0), 0);
+    return out;
+}
+

reference/shaders-msl/frag/bitcasting.1d-as-2d.frag

@@ -0,0 +1,30 @@
+#include <metal_stdlib>
+#include <simd/simd.h>
+
+using namespace metal;
+
+struct main0_out
+{
+    float4 FragColor0 [[color(0)]];
+    float4 FragColor1 [[color(1)]];
+};
+
+struct main0_in
+{
+    float4 VertGeom [[user(locn0)]];
+};
+
+fragment main0_out main0(main0_in in [[stage_in]], texture2d<float> TextureBase [[texture(0)]], texture2d<float> TextureDetail [[texture(1)]], sampler TextureBaseSmplr [[sampler(0)]], sampler TextureDetailSmplr [[sampler(1)]])
+{
+    main0_out out = {};
+    float4 texSample0 = TextureBase.sample(TextureBaseSmplr, float2(in.VertGeom.x, 0.5));
+    float4 texSample1 = TextureDetail.sample(TextureDetailSmplr, float2(in.VertGeom.x, 0.5), int2(3, 0));
+    int4 iResult0 = as_type<int4>(texSample0);
+    int4 iResult1 = as_type<int4>(texSample1);
+    out.FragColor0 = as_type<float4>(iResult0) * as_type<float4>(iResult1);
+    uint4 uResult0 = as_type<uint4>(texSample0);
+    uint4 uResult1 = as_type<uint4>(texSample1);
+    out.FragColor1 = as_type<float4>(uResult0) * as_type<float4>(uResult1);
+    return out;
+}
+

reference/shaders-msl/frag/sampler-1d-lod.1d-as-2d.frag

@@ -0,0 +1,22 @@
+#include <metal_stdlib>
+#include <simd/simd.h>
+
+using namespace metal;
+
+struct main0_out
+{
+    float4 FragColor [[color(0)]];
+};
+
+struct main0_in
+{
+    float vTex [[user(locn0), flat]];
+};
+
+fragment main0_out main0(main0_in in [[stage_in]], texture2d<float> uSampler [[texture(0)]], sampler uSamplerSmplr [[sampler(0)]])
+{
+    main0_out out = {};
+    out.FragColor += ((uSampler.sample(uSamplerSmplr, float2(in.vTex, 0.5), bias(2.0)) + uSampler.sample(uSamplerSmplr, float2(in.vTex, 0.5), level(3.0))) + uSampler.sample(uSamplerSmplr, float2(in.vTex, 0.5), gradient2d(5.0, 8.0)));
+    return out;
+}
+

reference/shaders-msl/frag/texel-fetch-offset.1d-as-2d.frag

@@ -0,0 +1,18 @@
+#include <metal_stdlib>
+#include <simd/simd.h>
+
+using namespace metal;
+
+struct main0_out
+{
+    float4 FragColor [[color(0)]];
+};
+
+fragment main0_out main0(texture2d<float> uTexture [[texture(0)]], texture2d<float> uTexture2 [[texture(1)]], sampler uTextureSmplr [[sampler(0)]], sampler uTexture2Smplr [[sampler(1)]], float4 gl_FragCoord [[position]])
+{
+    main0_out out = {};
+    out.FragColor = uTexture.read(uint2(int2(gl_FragCoord.xy)) + uint2(int2(1)), 0);
+    out.FragColor += uTexture2.read(uint2(uint(int(gl_FragCoord.x)), 0) + uint2(uint(-1), 0), 0);
+    return out;
+}
+

shaders-msl/frag/bitcasting.1d-as-2d.frag

@@ -0,0 +1,23 @@
+#version 450
+
+layout(binding = 0) uniform sampler1D TextureBase;
+layout(binding = 1) uniform sampler1D TextureDetail;
+
+layout(location = 0) in vec4 VertGeom;
+
+layout(location = 0) out vec4 FragColor0;
+layout(location = 1) out vec4 FragColor1;
+
+void main()
+{
+	vec4 texSample0 = texture(TextureBase, VertGeom.x);
+	vec4 texSample1 = textureOffset(TextureDetail, VertGeom.x, 3);
+	
+    ivec4 iResult0 = floatBitsToInt(texSample0);
+    ivec4 iResult1 = floatBitsToInt(texSample1);
+    FragColor0 = (intBitsToFloat(iResult0) * intBitsToFloat(iResult1));    
+    
+    uvec4 uResult0 = floatBitsToUint(texSample0);
+    uvec4 uResult1 = floatBitsToUint(texSample1);
+    FragColor1 = (uintBitsToFloat(uResult0) * uintBitsToFloat(uResult1));    
+}

shaders-msl/frag/sampler-1d-lod.1d-as-2d.frag

@@ -0,0 +1,12 @@
+#version 450
+
+layout(location = 0) out vec4 FragColor;
+layout(location = 0) flat in float vTex;
+layout(binding = 0) uniform sampler1D uSampler;
+
+void main()
+{
+	FragColor += texture(uSampler, vTex, 2.0) +
+		textureLod(uSampler, vTex, 3.0) +
+		textureGrad(uSampler, vTex, 5.0, 8.0);
+}

shaders-msl/frag/texel-fetch-offset.1d-as-2d.frag

@@ -0,0 +1,10 @@
+#version 450
+layout(location = 0) out vec4 FragColor;
+layout(binding = 0) uniform sampler2D uTexture;
+layout(binding = 1) uniform sampler1D uTexture2;
+
+void main()
+{
+	FragColor = texelFetchOffset(uTexture, ivec2(gl_FragCoord.xy), 0, ivec2(1, 1));
+	FragColor += texelFetchOffset(uTexture2, int(gl_FragCoord.x), 0, int(-1));
+}

spirv_msl.cpp

@@ -8265,25 +8265,26 @@ string CompilerMSL::to_function_args(const TextureFunctionArguments &args, bool
 		break;
 	}
 
-	if (args.base.is_fetch && args.offset)
+	if (args.base.is_fetch && (args.offset || args.coffset))
 	{
+		uint32_t offset_expr = args.offset ? args.offset : args.coffset;
 		// Fetch offsets must be applied directly to the coordinate.
-		forward = forward && should_forward(args.offset);
-		auto &type = expression_type(args.offset);
-		if (type.basetype != SPIRType::UInt)
-			tex_coords += " + " + bitcast_expression(SPIRType::UInt, args.offset);
+		forward = forward && should_forward(offset_expr);
+		auto &type = expression_type(offset_expr);
+		if (imgtype.image.dim == Dim1D && msl_options.texture_1D_as_2D)
+		{
+			if (type.basetype != SPIRType::UInt)
+				tex_coords += join(" + uint2(", bitcast_expression(SPIRType::UInt, offset_expr), ", 0)");
+			else
+				tex_coords += join(" + uint2(", to_enclosed_expression(offset_expr), ", 0)");
+		}
 		else
-			tex_coords += " + " + to_enclosed_expression(args.offset);
-	}
-	else if (args.base.is_fetch && args.coffset)
-	{
-		// Fetch offsets must be applied directly to the coordinate.
-		forward = forward && should_forward(args.coffset);
-		auto &type = expression_type(args.coffset);
-		if (type.basetype != SPIRType::UInt)
-			tex_coords += " + " + bitcast_expression(SPIRType::UInt, args.coffset);
-		else
-			tex_coords += " + " + to_enclosed_expression(args.coffset);
+		{
+			if (type.basetype != SPIRType::UInt)
+				tex_coords += " + " + bitcast_expression(SPIRType::UInt, offset_expr);
+			else
+				tex_coords += " + " + to_enclosed_expression(offset_expr);
+		}
 	}
 
 	// If projection, use alt coord as divisor
@@ -8454,6 +8455,7 @@ string CompilerMSL::to_function_args(const TextureFunctionArguments &args, bool
 		string grad_opt;
 		switch (imgtype.image.dim)
 		{
+		case Dim1D:
 		case Dim2D:
 			grad_opt = "2d";
 			break;
@@ -8489,30 +8491,42 @@ string CompilerMSL::to_function_args(const TextureFunctionArguments &args, bool
 
 	// Add offsets
 	string offset_expr;
+	const SPIRType *offset_type = nullptr;
 	if (args.coffset && !args.base.is_fetch)
 	{
 		forward = forward && should_forward(args.coffset);
 		offset_expr = to_expression(args.coffset);
+		offset_type = &expression_type(args.coffset);
 	}
 	else if (args.offset && !args.base.is_fetch)
 	{
 		forward = forward && should_forward(args.offset);
 		offset_expr = to_expression(args.offset);
+		offset_type = &expression_type(args.offset);
 	}
 
 	if (!offset_expr.empty())
 	{
 		switch (imgtype.image.dim)
 		{
+		case Dim1D:
+			if (!msl_options.texture_1D_as_2D)
+				break;
+			if (offset_type->vecsize > 1)
+				offset_expr = enclose_expression(offset_expr) + ".x";
+
+			farg_str += join(", int2(", offset_expr, ", 0)");
+			break;
+
 		case Dim2D:
-			if (coord_type.vecsize > 2)
+			if (offset_type->vecsize > 2)
 				offset_expr = enclose_expression(offset_expr) + ".xy";
 
 			farg_str += ", " + offset_expr;
 			break;
 
 		case Dim3D:
-			if (coord_type.vecsize > 3)
+			if (offset_type->vecsize > 3)
 				offset_expr = enclose_expression(offset_expr) + ".xyz";
 
 			farg_str += ", " + offset_expr;

test_shaders.py

@@ -306,6 +306,8 @@ def cross_compile_msl(shader, spirv, opt, iterations, paths):
         msl_args.append('0x00000022')
     if '.arrayed-subpass.' in shader:
         msl_args.append('--msl-arrayed-subpass-input')
+    if '.1d-as-2d.' in shader:
+        msl_args.append('--msl-texture-1d-as-2d')
 
     subprocess.check_call(msl_args)