Merge pull request #2386 from CodeLinaro/SPV_QCOM_image_processing2

support for SPV_QCOM_image_processing2

checkout_glslang_spirv_tools.sh

@@ -2,7 +2,7 @@
 # Copyright 2016-2021 The Khronos Group Inc.
 # SPDX-License-Identifier: Apache-2.0
 
-GLSLANG_REV=10ee92feb00712f444783032f5c9ee798887670d
+GLSLANG_REV=142052fa30f9eca191aa9dcf65359fcaed09eeec
 SPIRV_TOOLS_REV=fd96922e9a1814d92df46df03277788c799a4fad
 SPIRV_HEADERS_REV=ea77f2a826bc820cb8f57f9b2a7c7eccb681c73
 PROTOCOL=https

reference/opt/shaders/asm/frag/locations-components.asm.frag

@@ -2,7 +2,7 @@
 
 layout(location = 1) in vec2 _8;
 layout(location = 1, component = 2) in float _16;
-layout(location = 2) flat in float _22;
+layout(location = 3) flat in float _22;
 layout(location = 2, component = 1) flat in uint _28;
 layout(location = 2, component = 2) flat in uint _33;
 layout(location = 0) out vec4 o0;

reference/opt/shaders/vulkan/vert/small-storage.vk.vert

@@ -30,7 +30,7 @@ layout(binding = 0, std140) uniform block
     i8vec2 c;
     u8vec2 d;
     f16vec2 e;
-} _26;
+} _28;
 
 layout(binding = 1, std430) readonly buffer storage
 {
@@ -39,7 +39,7 @@ layout(binding = 1, std430) readonly buffer storage
     i8vec3 h;
     u8vec3 i;
     f16vec3 j;
-} _53;
+} _55;
 
 struct pushconst
 {
@@ -50,20 +50,20 @@ struct pushconst
     f16vec4 o;
 };
 
-uniform pushconst _76;
+uniform pushconst _78;
 
 layout(location = 0) out i16vec4 p;
-layout(location = 0, component = 0) in int16_t foo;
+layout(location = 0, component = 0) in uint16_t foo;
 layout(location = 1) out u16vec4 q;
 layout(location = 0, component = 1) in uint16_t bar;
 layout(location = 2) out f16vec4 r;
 layout(location = 1) in float16_t baz;
 
 void main()
 {
-    p = i16vec4((((ivec4(int(foo)) + ivec4(ivec2(_26.a), ivec2(_26.c))) - ivec4(ivec3(_53.f) / ivec3(_53.h), 1)) + ivec4(_76.k)) + ivec4(_76.m));
-    q = u16vec4((((uvec4(uint(bar)) + uvec4(uvec2(_26.b), uvec2(_26.d))) - uvec4(uvec3(_53.g) / uvec3(_53.i), 1u)) + uvec4(_76.l)) + uvec4(_76.n));
-    r = f16vec4(((vec4(float(baz)) + vec4(vec2(_26.e), 0.0, 1.0)) - vec4(vec3(_53.j), 1.0)) + vec4(_76.o));
+    p = i16vec4((((ivec4(int(uint(foo))) + ivec4(ivec2(_28.a), ivec2(_28.c))) - ivec4(ivec3(_55.f) / ivec3(_55.h), 1)) + ivec4(_78.k)) + ivec4(_78.m));
+    q = u16vec4((((uvec4(uint(bar)) + uvec4(uvec2(_28.b), uvec2(_28.d))) - uvec4(uvec3(_55.g) / uvec3(_55.i), 1u)) + uvec4(_78.l)) + uvec4(_78.n));
+    r = f16vec4(((vec4(float(baz)) + vec4(vec2(_28.e), 0.0, 1.0)) - vec4(vec3(_55.j), 1.0)) + vec4(_78.o));
     gl_Position = vec4(0.0, 0.0, 0.0, 1.0);
 }
 

reference/opt/shaders/vulkan/vert/small-storage.vk.vert.vk

@@ -18,7 +18,7 @@ layout(set = 0, binding = 0, std140) uniform block
     i8vec2 c;
     u8vec2 d;
     f16vec2 e;
-} _26;
+} _28;
 
 layout(set = 0, binding = 1, std430) readonly buffer storage
 {
@@ -27,7 +27,7 @@ layout(set = 0, binding = 1, std430) readonly buffer storage
     i8vec3 h;
     u8vec3 i;
     f16vec3 j;
-} _53;
+} _55;
 
 layout(push_constant, std430) uniform pushconst
 {
@@ -36,20 +36,20 @@ layout(push_constant, std430) uniform pushconst
     i8vec4 m;
     u8vec4 n;
     f16vec4 o;
-} _76;
+} _78;
 
 layout(location = 0) out i16vec4 p;
-layout(location = 0, component = 0) in int16_t foo;
+layout(location = 0, component = 0) in uint16_t foo;
 layout(location = 1) out u16vec4 q;
 layout(location = 0, component = 1) in uint16_t bar;
 layout(location = 2) out f16vec4 r;
 layout(location = 1) in float16_t baz;
 
 void main()
 {
-    p = i16vec4((((ivec4(int(foo)) + ivec4(ivec2(_26.a), ivec2(_26.c))) - ivec4(ivec3(_53.f) / ivec3(_53.h), 1)) + ivec4(_76.k)) + ivec4(_76.m));
-    q = u16vec4((((uvec4(uint(bar)) + uvec4(uvec2(_26.b), uvec2(_26.d))) - uvec4(uvec3(_53.g) / uvec3(_53.i), 1u)) + uvec4(_76.l)) + uvec4(_76.n));
-    r = f16vec4(((vec4(float(baz)) + vec4(vec2(_26.e), 0.0, 1.0)) - vec4(vec3(_53.j), 1.0)) + vec4(_76.o));
+    p = i16vec4((((ivec4(int(uint(foo))) + ivec4(ivec2(_28.a), ivec2(_28.c))) - ivec4(ivec3(_55.f) / ivec3(_55.h), 1)) + ivec4(_78.k)) + ivec4(_78.m));
+    q = u16vec4((((uvec4(uint(bar)) + uvec4(uvec2(_28.b), uvec2(_28.d))) - uvec4(uvec3(_55.g) / uvec3(_55.i), 1u)) + uvec4(_78.l)) + uvec4(_78.n));
+    r = f16vec4(((vec4(float(baz)) + vec4(vec2(_28.e), 0.0, 1.0)) - vec4(vec3(_55.j), 1.0)) + vec4(_78.o));
     gl_Position = vec4(0.0, 0.0, 0.0, 1.0);
 }
 

reference/shaders/asm/frag/locations-components.asm.frag

@@ -2,7 +2,7 @@
 
 layout(location = 1) in vec2 _8;
 layout(location = 1, component = 2) in float _16;
-layout(location = 2) flat in float _22;
+layout(location = 3) flat in float _22;
 layout(location = 2, component = 1) flat in uint _28;
 layout(location = 2, component = 2) flat in uint _33;
 layout(location = 0) out vec4 o0;

reference/shaders/vulkan/vert/small-storage.vk.vert

@@ -30,7 +30,7 @@ layout(binding = 0, std140) uniform block
     i8vec2 c;
     u8vec2 d;
     f16vec2 e;
-} _26;
+} _28;
 
 layout(binding = 1, std430) readonly buffer storage
 {
@@ -39,7 +39,7 @@ layout(binding = 1, std430) readonly buffer storage
     i8vec3 h;
     u8vec3 i;
     f16vec3 j;
-} _53;
+} _55;
 
 struct pushconst
 {
@@ -50,20 +50,20 @@ struct pushconst
     f16vec4 o;
 };
 
-uniform pushconst _76;
+uniform pushconst _78;
 
 layout(location = 0) out i16vec4 p;
-layout(location = 0, component = 0) in int16_t foo;
+layout(location = 0, component = 0) in uint16_t foo;
 layout(location = 1) out u16vec4 q;
 layout(location = 0, component = 1) in uint16_t bar;
 layout(location = 2) out f16vec4 r;
 layout(location = 1) in float16_t baz;
 
 void main()
 {
-    p = i16vec4((((ivec4(int(foo)) + ivec4(ivec2(_26.a), ivec2(_26.c))) - ivec4(ivec3(_53.f) / ivec3(_53.h), 1)) + ivec4(_76.k)) + ivec4(_76.m));
-    q = u16vec4((((uvec4(uint(bar)) + uvec4(uvec2(_26.b), uvec2(_26.d))) - uvec4(uvec3(_53.g) / uvec3(_53.i), 1u)) + uvec4(_76.l)) + uvec4(_76.n));
-    r = f16vec4(((vec4(float(baz)) + vec4(vec2(_26.e), 0.0, 1.0)) - vec4(vec3(_53.j), 1.0)) + vec4(_76.o));
+    p = i16vec4((((ivec4(int(uint(foo))) + ivec4(ivec2(_28.a), ivec2(_28.c))) - ivec4(ivec3(_55.f) / ivec3(_55.h), 1)) + ivec4(_78.k)) + ivec4(_78.m));
+    q = u16vec4((((uvec4(uint(bar)) + uvec4(uvec2(_28.b), uvec2(_28.d))) - uvec4(uvec3(_55.g) / uvec3(_55.i), 1u)) + uvec4(_78.l)) + uvec4(_78.n));
+    r = f16vec4(((vec4(float(baz)) + vec4(vec2(_28.e), 0.0, 1.0)) - vec4(vec3(_55.j), 1.0)) + vec4(_78.o));
     gl_Position = vec4(0.0, 0.0, 0.0, 1.0);
 }
 

reference/shaders/vulkan/vert/small-storage.vk.vert.vk

@@ -18,7 +18,7 @@ layout(set = 0, binding = 0, std140) uniform block
     i8vec2 c;
     u8vec2 d;
     f16vec2 e;
-} _26;
+} _28;
 
 layout(set = 0, binding = 1, std430) readonly buffer storage
 {
@@ -27,7 +27,7 @@ layout(set = 0, binding = 1, std430) readonly buffer storage
     i8vec3 h;
     u8vec3 i;
     f16vec3 j;
-} _53;
+} _55;
 
 layout(push_constant, std430) uniform pushconst
 {
@@ -36,20 +36,20 @@ layout(push_constant, std430) uniform pushconst
     i8vec4 m;
     u8vec4 n;
     f16vec4 o;
-} _76;
+} _78;
 
 layout(location = 0) out i16vec4 p;
-layout(location = 0, component = 0) in int16_t foo;
+layout(location = 0, component = 0) in uint16_t foo;
 layout(location = 1) out u16vec4 q;
 layout(location = 0, component = 1) in uint16_t bar;
 layout(location = 2) out f16vec4 r;
 layout(location = 1) in float16_t baz;
 
 void main()
 {
-    p = i16vec4((((ivec4(int(foo)) + ivec4(ivec2(_26.a), ivec2(_26.c))) - ivec4(ivec3(_53.f) / ivec3(_53.h), 1)) + ivec4(_76.k)) + ivec4(_76.m));
-    q = u16vec4((((uvec4(uint(bar)) + uvec4(uvec2(_26.b), uvec2(_26.d))) - uvec4(uvec3(_53.g) / uvec3(_53.i), 1u)) + uvec4(_76.l)) + uvec4(_76.n));
-    r = f16vec4(((vec4(float(baz)) + vec4(vec2(_26.e), 0.0, 1.0)) - vec4(vec3(_53.j), 1.0)) + vec4(_76.o));
+    p = i16vec4((((ivec4(int(uint(foo))) + ivec4(ivec2(_28.a), ivec2(_28.c))) - ivec4(ivec3(_55.f) / ivec3(_55.h), 1)) + ivec4(_78.k)) + ivec4(_78.m));
+    q = u16vec4((((uvec4(uint(bar)) + uvec4(uvec2(_28.b), uvec2(_28.d))) - uvec4(uvec3(_55.g) / uvec3(_55.i), 1u)) + uvec4(_78.l)) + uvec4(_78.n));
+    r = f16vec4(((vec4(float(baz)) + vec4(vec2(_28.e), 0.0, 1.0)) - vec4(vec3(_55.j), 1.0)) + vec4(_78.o));
     gl_Position = vec4(0.0, 0.0, 0.0, 1.0);
 }
 

shaders/asm/frag/locations-components.asm.frag

@@ -15,7 +15,7 @@
                OpDecorate %8 Location 1
                OpDecorate %16 Location 1
                OpDecorate %16 Component 2
-               OpDecorate %22 Location 2
+               OpDecorate %22 Location 3
                OpDecorate %22 Flat
                OpDecorate %28 Location 2
                OpDecorate %28 Component 1

shaders/vulkan/vert/small-storage.vk.vert

@@ -5,7 +5,7 @@
 #extension GL_AMD_gpu_shader_int16 : require
 #extension GL_AMD_gpu_shader_half_float : require
 
-layout(location = 0, component = 0) in int16_t foo;
+layout(location = 0, component = 0) in uint16_t foo;
 layout(location = 0, component = 1) in uint16_t bar;
 layout(location = 1) in float16_t baz;
 

spirv.h

@@ -1,5 +1,5 @@
 /*
-** Copyright (c) 2014-2020 The Khronos Group Inc.
+** Copyright (c) 2014-2024 The Khronos Group Inc.
 ** 
 ** Permission is hereby granted, free of charge, to any person obtaining a copy
 ** of this software and/or associated documentation files (the "Materials"),
@@ -511,6 +511,7 @@ typedef enum SpvDecoration_ {
     SpvDecorationNoUnsignedWrap = 4470,
     SpvDecorationWeightTextureQCOM = 4487,
     SpvDecorationBlockMatchTextureQCOM = 4488,
+    SpvDecorationBlockMatchSamplerQCOM = 4499,
     SpvDecorationExplicitInterpAMD = 4999,
     SpvDecorationOverrideCoverageNV = 5248,
     SpvDecorationPassthroughNV = 5250,
@@ -996,6 +997,7 @@ typedef enum SpvCapability_ {
     SpvCapabilityTextureSampleWeightedQCOM = 4484,
     SpvCapabilityTextureBoxFilterQCOM = 4485,
     SpvCapabilityTextureBlockMatchQCOM = 4486,
+    SpvCapabilityTextureBlockMatch2QCOM = 4498,
     SpvCapabilityFloat16ImageAMD = 5008,
     SpvCapabilityImageGatherBiasLodAMD = 5009,
     SpvCapabilityFragmentMaskAMD = 5010,
@@ -1605,6 +1607,10 @@ typedef enum SpvOp_ {
     SpvOpImageBoxFilterQCOM = 4481,
     SpvOpImageBlockMatchSSDQCOM = 4482,
     SpvOpImageBlockMatchSADQCOM = 4483,
+    SpvOpImageBlockMatchWindowSSDQCOM = 4500,
+    SpvOpImageBlockMatchWindowSADQCOM = 4501,
+    SpvOpImageBlockMatchGatherSSDQCOM = 4502,
+    SpvOpImageBlockMatchGatherSADQCOM = 4503,
     SpvOpGroupIAddNonUniformAMD = 5000,
     SpvOpGroupFAddNonUniformAMD = 5001,
     SpvOpGroupFMinNonUniformAMD = 5002,
@@ -2284,6 +2290,10 @@ inline void SpvHasResultAndType(SpvOp opcode, bool *hasResult, bool *hasResultTy
     case SpvOpImageBoxFilterQCOM: *hasResult = true; *hasResultType = true; break;
     case SpvOpImageBlockMatchSSDQCOM: *hasResult = true; *hasResultType = true; break;
     case SpvOpImageBlockMatchSADQCOM: *hasResult = true; *hasResultType = true; break;
+    case SpvOpImageBlockMatchWindowSSDQCOM: *hasResult = true; *hasResultType = true; break;
+    case SpvOpImageBlockMatchWindowSADQCOM: *hasResult = true; *hasResultType = true; break;
+    case SpvOpImageBlockMatchGatherSSDQCOM: *hasResult = true; *hasResultType = true; break;
+    case SpvOpImageBlockMatchGatherSADQCOM: *hasResult = true; *hasResultType = true; break;
     case SpvOpGroupIAddNonUniformAMD: *hasResult = true; *hasResultType = true; break;
     case SpvOpGroupFAddNonUniformAMD: *hasResult = true; *hasResultType = true; break;
     case SpvOpGroupFMinNonUniformAMD: *hasResult = true; *hasResultType = true; break;

spirv.hpp

@@ -1,4 +1,4 @@
-// Copyright (c) 2014-2020 The Khronos Group Inc.
+// Copyright (c) 2014-2024 The Khronos Group Inc.
 // 
 // Permission is hereby granted, free of charge, to any person obtaining a copy
 // of this software and/or associated documentation files (the "Materials"),
@@ -507,6 +507,7 @@ enum Decoration {
     DecorationNoUnsignedWrap = 4470,
     DecorationWeightTextureQCOM = 4487,
     DecorationBlockMatchTextureQCOM = 4488,
+    DecorationBlockMatchSamplerQCOM = 4499,
     DecorationExplicitInterpAMD = 4999,
     DecorationOverrideCoverageNV = 5248,
     DecorationPassthroughNV = 5250,
@@ -992,6 +993,7 @@ enum Capability {
     CapabilityTextureSampleWeightedQCOM = 4484,
     CapabilityTextureBoxFilterQCOM = 4485,
     CapabilityTextureBlockMatchQCOM = 4486,
+    CapabilityTextureBlockMatch2QCOM = 4498,
     CapabilityFloat16ImageAMD = 5008,
     CapabilityImageGatherBiasLodAMD = 5009,
     CapabilityFragmentMaskAMD = 5010,
@@ -1601,6 +1603,10 @@ enum Op {
     OpImageBoxFilterQCOM = 4481,
     OpImageBlockMatchSSDQCOM = 4482,
     OpImageBlockMatchSADQCOM = 4483,
+    OpImageBlockMatchWindowSSDQCOM = 4500,
+    OpImageBlockMatchWindowSADQCOM = 4501,
+    OpImageBlockMatchGatherSSDQCOM = 4502,
+    OpImageBlockMatchGatherSADQCOM = 4503,
     OpGroupIAddNonUniformAMD = 5000,
     OpGroupFAddNonUniformAMD = 5001,
     OpGroupFMinNonUniformAMD = 5002,
@@ -2280,6 +2286,10 @@ inline void HasResultAndType(Op opcode, bool *hasResult, bool *hasResultType) {
     case OpImageBoxFilterQCOM: *hasResult = true; *hasResultType = true; break;
     case OpImageBlockMatchSSDQCOM: *hasResult = true; *hasResultType = true; break;
     case OpImageBlockMatchSADQCOM: *hasResult = true; *hasResultType = true; break;
+    case OpImageBlockMatchWindowSSDQCOM: *hasResult = true; *hasResultType = true; break;
+    case OpImageBlockMatchWindowSADQCOM: *hasResult = true; *hasResultType = true; break;
+    case OpImageBlockMatchGatherSSDQCOM: *hasResult = true; *hasResultType = true; break;
+    case OpImageBlockMatchGatherSADQCOM: *hasResult = true; *hasResultType = true; break;
     case OpGroupIAddNonUniformAMD: *hasResult = true; *hasResultType = true; break;
     case OpGroupFAddNonUniformAMD: *hasResult = true; *hasResultType = true; break;
     case OpGroupFMinNonUniformAMD: *hasResult = true; *hasResultType = true; break;

spirv_glsl.cpp

@@ -14481,6 +14481,50 @@ void CompilerGLSL::emit_instruction(const Instruction &instruction)
 		break;
 	}
 
+	case OpImageBlockMatchWindowSSDQCOM:
+	case OpImageBlockMatchWindowSADQCOM:
+	case OpImageBlockMatchGatherSSDQCOM:
+	case OpImageBlockMatchGatherSADQCOM:
+	{
+		require_extension_internal("GL_QCOM_image_processing2");
+		uint32_t result_type_id = ops[0];
+		uint32_t id = ops[1];
+		string expr;
+		switch (opcode)
+		{
+		case OpImageBlockMatchWindowSSDQCOM:
+			expr = "textureBlockMatchWindowSSDQCOM";
+			break;
+		case OpImageBlockMatchWindowSADQCOM:
+			expr = "textureBlockMatchWindowSADQCOM";
+			break;
+		case OpImageBlockMatchGatherSSDQCOM:
+			expr = "textureBlockMatchGatherSSDQCOM";
+			break;
+		case OpImageBlockMatchGatherSADQCOM:
+			expr = "textureBlockMatchGatherSADQCOM";
+			break;
+		default:
+			SPIRV_CROSS_THROW("Invalid opcode for QCOM_image_processing2.");
+		}
+		expr += "(";
+
+		bool forward = false;
+		expr += to_expression(ops[2]);
+		expr += ", " + to_expression(ops[3]);
+
+		expr += ", " + to_non_uniform_aware_expression(ops[4]);
+		expr += ", " + to_expression(ops[5]);
+		expr += ", " + to_expression(ops[6]);
+
+		expr += ")";
+		emit_op(result_type_id, id, expr, forward);
+
+		inherit_expression_dependencies(id, ops[3]);
+		inherit_expression_dependencies(id, ops[5]);
+		break;
+	}
+
 	// Compute
 	case OpControlBarrier:
 	case OpMemoryBarrier:

spirv_reflect.cpp

@@ -637,6 +637,8 @@ void CompilerReflection::emit_resources(const char *tag, const SmallVector<Resou
 			json_stream->emit_json_key_value("WeightTextureQCOM", get_decoration(res.id, DecorationWeightTextureQCOM));
 		if (mask.get(DecorationBlockMatchTextureQCOM))
 			json_stream->emit_json_key_value("BlockMatchTextureQCOM", get_decoration(res.id, DecorationBlockMatchTextureQCOM));
+		if (mask.get(DecorationBlockMatchSamplerQCOM))
+			json_stream->emit_json_key_value("BlockMatchSamplerQCOM", get_decoration(res.id, DecorationBlockMatchSamplerQCOM));
 
 		// For images, the type itself adds a layout qualifer.
 		// Only emit the format for storage images.