Skip to content

G
glslang
Commit 5ea6a192 by John Kessenich
Options

Merge branch 'amdrexu-feature'

parents 4d5bcd31 cabbb788
Showing with 993 additions and 23 deletions
SPIRV/GLSL.ext.AMD.h
......@@ -33,7 +33,7 @@ enum Decoration;
enum Op;
static const int GLSLextAMDVersion = 100;
static const int GLSLextAMDRevision = 3;
static const int GLSLextAMDRevision = 4;
// SPV_AMD_shader_ballot
static const char* const E_SPV_AMD_shader_ballot = "SPV_AMD_shader_ballot";
......@@ -119,4 +119,7 @@ static const char* const E_SPV_AMD_texture_gather_bias_lod = "SPV_AMD_texture_ga
static const Capability OpCapabilityImageGatherBiasLodAMD = static_cast<Capability>(5009);
// SPV_AMD_gpu_shader_int16
static const char* const E_SPV_AMD_gpu_shader_int16 = "SPV_AMD_gpu_shader_int16";
#endif // #ifndef GLSLextAMD_H
SPIRV/SpvBuilder.h
......@@ -214,6 +214,10 @@ public:
Id makeUintConstant(unsigned u, bool specConstant = false) { return makeIntConstant(makeUintType(32), u, specConstant); }
Id makeInt64Constant(long long i, bool specConstant = false) { return makeInt64Constant(makeIntType(64), (unsigned long long)i, specConstant); }
Id makeUint64Constant(unsigned long long u, bool specConstant = false) { return makeInt64Constant(makeUintType(64), u, specConstant); }
#ifdef AMD_EXTENSIONS
Id makeInt16Constant(short i, bool specConstant = false) { return makeIntConstant(makeIntType(16), (unsigned)((unsigned short)i), specConstant); }
Id makeUint16Constant(unsigned short u, bool specConstant = false) { return makeIntConstant(makeUintType(16), (unsigned)u, specConstant); }
#endif
Id makeFloatConstant(float f, bool specConstant = false);
Id makeDoubleConstant(double d, bool specConstant = false);
#ifdef AMD_EXTENSIONS
......
Test/spv.int16.frag 0 → 100644
#version 450 core
#extension GL_ARB_gpu_shader_int64: enable
#extension GL_AMD_gpu_shader_half_float: enable
#extension GL_AMD_gpu_shader_int16: enable
layout(binding = 0) uniform Uniforms
{
uint i;
};
// int16/uint16 in block
layout(std140, binding = 1) uniform Block
{
i16vec3 i16v;
uint16_t u16;
} block;
// int16/uint16 for input
layout(location = 0) in flat u16vec3 iu16v;
layout(location = 1) in flat int16_t ii16;
void literal()
{
const int16_t i16c[3] =
{
0x111S, // Hex
-2s, // Dec
0400s, // Oct
};
const uint16_t u16c[] =
{
0xFFFFus, // Hex
65535US, // Dec
0177777us, // Oct
};
uint16_t u16 = i16c[i] + u16c[i];
}
void operators()
{
u16vec3 u16v;
int16_t i16;
uint16_t u16;
int i;
uint u;
bool b;
// Unary
u16v++;
i16--;
++i16;
--u16v;
u16v = ~u16v;
i16 = +i16;
u16v = -u16v;
// Arithmetic
u16 += i16;
u16v -= u16v;
i16 *= i16;
u16v /= u16v;
u16v %= i16;
u16v = u16v + u16v;
u16 = i16 - u16;
u16v = u16v * i16;
i16 = i16 * i16;
i16 = i16 % i16;
// Shift
u16v <<= i;
i16 >>= u16v.y;
i16 = i16 << u16v.z;
u16v = u16v << i16;
// Relational
b = (u16v.x != i16);
b = (i16 == u16v.x);
b = (u16v.x > u16v.y);
b = (i16 < u);
b = (u16v.y >= u16v.x);
b = (i16 <= i);
// Bitwise
u16v |= i16;
u16 = i16 | u16;
i16 &= i16;
u16v = u16v & u16v;
u16v ^= i16;
u16v = u16v ^ i16;
}
void typeCast()
{
bvec2 bv;
ivec2 iv;
uvec2 uv;
vec2 fv;
dvec2 dv;
f16vec2 f16v;
i64vec2 i64v;
u64vec2 u64v;
i16vec2 i16v;
u16vec2 u16v;
i16v = i16vec2(bv); // bool -> int16
u16v = u16vec2(bv); // bool -> uint16
bv = bvec2(i16v); // int16 -> bool
bv = bvec2(u16v); // uint16 -> bool
i16v = i16vec2(iv); // int -> int16
u16v = u16vec2(iv); // int -> uint16
iv = i16v; // int16 -> int
iv = ivec2(u16v); // uint16 -> int
i16v = i16vec2(uv); // uint -> int16
u16v = u16vec2(uv); // uint -> uint16
uv = i16v; // int16 -> uint
uv = u16v; // uint16 -> uint
i16v = i16vec2(fv); // float -> int16
u16v = u16vec2(fv); // float -> uint16
fv = i16v; // int16 -> float
fv = u16v; // uint16 -> float
i16v = i16vec2(dv); // double -> int16
u16v = u16vec2(dv); // double -> uint16
dv = i16v; // int16 -> double
dv = u16v; // uint16 -> double
i16v = i16vec2(f16v); // float16 -> int16
u16v = u16vec2(f16v); // float16 -> uint16
f16v = i16v; // int16 -> float16
f16v = u16v; // uint16 -> float16
i16v = i16vec2(i64v); // int64 -> int16
u16v = u16vec2(i64v); // int64 -> uint16
i64v = i16v; // int16 -> int64
i64v = i64vec2(u16v); // uint16 -> int64
i16v = i16vec2(u64v); // uint64 -> int16
u16v = u16vec2(u64v); // uint64 -> uint16
u64v = i16v; // int16 -> uint64
u64v = u16v; // uint16 -> uint64
i16v = i16vec2(u16v); // uint16 -> int16
u16v = i16v; // int16 -> uint16
}
void builtinFuncs()
{
i16vec2 i16v;
u16vec3 u16v;
f16vec3 f16v;
bvec3 bv;
int16_t i16;
uint16_t u16;
// abs()
i16v = abs(i16v);
// sign()
i16v = sign(i16v);
// min()
i16v = min(i16v, i16);
i16v = min(i16v, i16vec2(-1s));
u16v = min(u16v, u16);
u16v = min(u16v, u16vec3(0us));
// max()
i16v = max(i16v, i16);
i16v = max(i16v, i16vec2(-1s));
u16v = max(u16v, u16);
u16v = max(u16v, u16vec3(0us));
// clamp()
i16v = clamp(i16v, -i16, i16);
i16v = clamp(i16v, -i16v, i16v);
u16v = clamp(u16v, -u16, u16);
u16v = clamp(u16v, -u16v, u16v);
// mix()
i16 = mix(i16v.x, i16v.y, true);
i16v = mix(i16vec2(i16), i16vec2(-i16), bvec2(false));
u16 = mix(u16v.x, u16v.y, true);
u16v = mix(u16vec3(u16), u16vec3(-u16), bvec3(false));
// frexp()
i16vec3 exp;
f16v = frexp(f16v, exp);
// ldexp()
f16v = ldexp(f16v, exp);
// float16BitsToInt16()
i16v = float16BitsToInt16(f16v.xy);
// float16BitsToUint16()
u16v.x = float16BitsToUint16(f16v.z);
// int16BitsToFloat16()
f16v.xy = int16BitsToFloat16(i16v);
// uint16BitsToFloat16()
f16v = uint16BitsToFloat16(u16v);
// packInt2x16()
int packi = packInt2x16(i16v);
// unpackInt2x16()
i16v = unpackInt2x16(packi);
// packUint2x16()
uint packu = packUint2x16(u16v.xy);
// unpackUint2x16()
u16v.xy = unpackUint2x16(packu);
// packInt4x16()
int64_t packi64 = packInt4x16(i16vec4(i16));
// unpackInt4x16()
i16v = unpackInt4x16(packi64).xy;
// packUint4x16()
uint64_t packu64 = packUint4x16(u16vec4(u16));
// unpackUint4x16()
u16v = unpackUint4x16(packu64).xyz;
// lessThan()
bv = lessThan(u16v, u16vec3(u16));
bv.xy = lessThan(i16v, i16vec2(i16));
// lessThanEqual()
bv = lessThanEqual(u16v, u16vec3(u16));
bv.xy = lessThanEqual(i16v, i16vec2(i16));
// greaterThan()
bv = greaterThan(u16v, u16vec3(u16));
bv.xy = greaterThan(i16v, i16vec2(i16));
// greaterThanEqual()
bv = greaterThanEqual(u16v, u16vec3(u16));
bv.xy = greaterThanEqual(i16v, i16vec2(i16));
// equal()
bv = equal(u16v, u16vec3(u16));
bv.xy = equal(i16v, i16vec2(i16));
// notEqual()
bv = notEqual(u16v, u16vec3(u16));
bv.xy = notEqual(i16v, i16vec2(i16));
}
// Type conversion for specialization constant
layout(constant_id = 100) const int64_t si64 = -10L;
layout(constant_id = 101) const uint64_t su64 = 20UL;
layout(constant_id = 102) const int si = -5;
layout(constant_id = 103) const uint su = 4;
layout(constant_id = 104) const bool sb = true;
layout(constant_id = 105) const int16_t si16 = -5S;
layout(constant_id = 106) const uint16_t su16 = 4US;
// bool <-> int16/uint16
const bool i16_to_b = bool(si16);
const bool u16_to_b = bool(su16);
const int16_t b_to_i16 = int16_t(sb);
const uint16_t b_to_u16 = uint16_t(sb);
// int <-> int16/uint16
const int i16_to_i = int(si16);
const int u16_to_i = int(su16);
const int16_t i_to_i16 = int16_t(si);
const uint16_t i_to_u16 = uint16_t(si);
// uint <-> int16/uint16
const uint i16_to_u = uint(si16);
const uint u16_to_u = uint(su16);
const int16_t u_to_i16 = int16_t(su);
const uint16_t u_to_u16 = uint16_t(su);
// int64 <-> int16/uint16
const int64_t i16_to_i64 = int64_t(si16);
const int64_t u16_to_i64 = int64_t(su16);
const int16_t i64_to_i16 = int16_t(si64);
const uint16_t i64_to_u16 = uint16_t(si64);
// uint64 <-> int16/uint16
const uint64_t i16_to_u64 = uint64_t(si16);
const uint64_t u16_to_u64 = uint64_t(su16);
const int16_t u64_to_i16 = int16_t(su64);
const uint16_t u64_to_u16 = uint16_t(su64);
// int16 <-> uint16
const uint16_t i16_to_u16 = uint16_t(si16);
const int16_t u16_to_i16 = int16_t(su16);
void main()
{
literal();
operators();
typeCast();
builtinFuncs();
}
glslang/Include/BaseTypes.h
......@@ -53,6 +53,10 @@ enum TBasicType {
EbtUint,
EbtInt64,
EbtUint64,
#ifdef AMD_EXTENSIONS
EbtInt16,
EbtUint16,
#endif
EbtBool,
EbtAtomicUint,
EbtSampler,
......
glslang/Include/Types.h
......@@ -190,8 +190,6 @@ struct TSampler { // misnomer now; includes images, textures without sampler,
case EbtFloat: break;
case EbtInt: s.append("i"); break;
case EbtUint: s.append("u"); break;
case EbtInt64: s.append("i64"); break;
case EbtUint64: s.append("u64"); break;
default: break; // some compilers want this
}
if (image) {
......@@ -1448,6 +1446,10 @@ public:
case EbtUint:
case EbtInt64:
case EbtUint64:
#ifdef AMD_EXTENSIONS
case EbtInt16:
case EbtUint16:
#endif
case EbtBool:
return true;
default:
......@@ -1531,6 +1533,10 @@ public:
case EbtUint: return "uint";
case EbtInt64: return "int64_t";
case EbtUint64: return "uint64_t";
#ifdef AMD_EXTENSIONS
case EbtInt16: return "int16_t";
case EbtUint16: return "uint16_t";
#endif
case EbtBool: return "bool";
case EbtAtomicUint: return "atomic_uint";
case EbtSampler: return "sampler/image";
......
glslang/Include/intermediate.h
......@@ -141,6 +141,42 @@ enum TOperator {
EOpConvFloat16ToDouble,
EOpConvFloat16ToInt64,
EOpConvFloat16ToUint64,
EOpConvBoolToInt16,
EOpConvIntToInt16,
EOpConvUintToInt16,
EOpConvFloatToInt16,
EOpConvDoubleToInt16,
EOpConvFloat16ToInt16,
EOpConvInt64ToInt16,
EOpConvUint64ToInt16,
EOpConvUint16ToInt16,
EOpConvInt16ToBool,
EOpConvInt16ToInt,
EOpConvInt16ToUint,
EOpConvInt16ToFloat,
EOpConvInt16ToDouble,
EOpConvInt16ToFloat16,
EOpConvInt16ToInt64,
EOpConvInt16ToUint64,
EOpConvBoolToUint16,
EOpConvIntToUint16,
EOpConvUintToUint16,
EOpConvFloatToUint16,
EOpConvDoubleToUint16,
EOpConvFloat16ToUint16,
EOpConvInt64ToUint16,
EOpConvUint64ToUint16,
EOpConvInt16ToUint16,
EOpConvUint16ToBool,
EOpConvUint16ToInt,
EOpConvUint16ToUint,
EOpConvUint16ToFloat,
EOpConvUint16ToDouble,
EOpConvUint16ToFloat16,
EOpConvUint16ToInt64,
EOpConvUint16ToUint64,
#endif
//
......@@ -244,6 +280,12 @@ enum TOperator {
EOpDoubleBitsToUint64,
EOpInt64BitsToDouble,
EOpUint64BitsToDouble,
#ifdef AMD_EXTENSIONS
EOpFloat16BitsToInt16,
EOpFloat16BitsToUint16,
EOpInt16BitsToFloat16,
EOpUint16BitsToFloat16,
#endif
EOpPackSnorm2x16,
EOpUnpackSnorm2x16,
EOpPackUnorm2x16,
......@@ -263,6 +305,14 @@ enum TOperator {
#ifdef AMD_EXTENSIONS
EOpPackFloat2x16,
EOpUnpackFloat2x16,
EOpPackInt2x16,
EOpUnpackInt2x16,
EOpPackUint2x16,
EOpUnpackUint2x16,
EOpPackInt4x16,
EOpUnpackInt4x16,
EOpPackUint4x16,
EOpUnpackUint4x16,
#endif
EOpLength,
......@@ -394,15 +444,27 @@ enum TOperator {
EOpConstructUint,
EOpConstructInt64,
EOpConstructUint64,
#ifdef AMD_EXTENSIONS
EOpConstructInt16,
EOpConstructUint16,
#endif
EOpConstructBool,
EOpConstructFloat,
EOpConstructDouble,
#ifdef AMD_EXTENSIONS
EOpConstructFloat16,
#endif
EOpConstructVec2,
EOpConstructVec3,
EOpConstructVec4,
EOpConstructDVec2,
EOpConstructDVec3,
EOpConstructDVec4,
#ifdef AMD_EXTENSIONS
EOpConstructF16Vec2,
EOpConstructF16Vec3,
EOpConstructF16Vec4,
#endif
EOpConstructBVec2,
EOpConstructBVec3,
EOpConstructBVec4,
......@@ -418,6 +480,14 @@ enum TOperator {
EOpConstructU64Vec2,
EOpConstructU64Vec3,
EOpConstructU64Vec4,
#ifdef AMD_EXTENSIONS
EOpConstructI16Vec2,
EOpConstructI16Vec3,
EOpConstructI16Vec4,
EOpConstructU16Vec2,
EOpConstructU16Vec3,
EOpConstructU16Vec4,
#endif
EOpConstructMat2x2,
EOpConstructMat2x3,
EOpConstructMat2x4,
......@@ -464,10 +534,6 @@ enum TOperator {
EOpConstructBMat4x3,
EOpConstructBMat4x4,
#ifdef AMD_EXTENSIONS
EOpConstructFloat16,
EOpConstructF16Vec2,
EOpConstructF16Vec3,
EOpConstructF16Vec4,
EOpConstructF16Mat2x2,
EOpConstructF16Mat2x3,
EOpConstructF16Mat2x4,
......
glslang/MachineIndependent/Constant.cpp
......@@ -193,7 +193,7 @@ TIntermTyped* TIntermConstantUnion::fold(TOperator op, const TIntermTyped* right
case EbtUint:
if (rightUnionArray[i] == 0) {
newConstArray[i].setUConst(0xFFFFFFFF);
newConstArray[i].setUConst(0xFFFFFFFFu);
} else
newConstArray[i].setUConst(leftUnionArray[i].getUConst() / rightUnionArray[i].getUConst());
break;
......@@ -213,6 +213,23 @@ TIntermTyped* TIntermConstantUnion::fold(TOperator op, const TIntermTyped* right
} else
newConstArray[i].setU64Const(leftUnionArray[i].getU64Const() / rightUnionArray[i].getU64Const());
break;
#ifdef AMD_EXTENSIONS
case EbtInt16:
if (rightUnionArray[i] == 0)
newConstArray[i].setIConst(0x7FFF);
else if (rightUnionArray[i].getIConst() == -1 && leftUnionArray[i].getIConst() == (int)0x8000)
newConstArray[i].setIConst(0x8000);
else
newConstArray[i].setIConst(leftUnionArray[i].getIConst() / rightUnionArray[i].getIConst());
break;
case EbtUint16:
if (rightUnionArray[i] == 0) {
newConstArray[i].setUConst(0xFFFFu);
} else
newConstArray[i].setUConst(leftUnionArray[i].getUConst() / rightUnionArray[i].getUConst());
break;
#endif
default:
return 0;
}
......@@ -457,10 +474,16 @@ TIntermTyped* TIntermConstantUnion::fold(TOperator op, const TType& returnType)
case EbtFloat16:
#endif
case EbtFloat: newConstArray[i].setDConst(-unionArray[i].getDConst()); break;
#ifdef AMD_EXTENSIONS
case EbtInt16:
#endif
case EbtInt: newConstArray[i].setIConst(-unionArray[i].getIConst()); break;
#ifdef AMD_EXTENSIONS
case EbtUint16:
#endif
case EbtUint: newConstArray[i].setUConst(static_cast<unsigned int>(-static_cast<int>(unionArray[i].getUConst()))); break;
case EbtInt64: newConstArray[i].setI64Const(-unionArray[i].getI64Const()); break;
case EbtUint64: newConstArray[i].setU64Const(static_cast<unsigned int>(-static_cast<int>(unionArray[i].getU64Const()))); break;
case EbtUint64: newConstArray[i].setU64Const(static_cast<unsigned long long>(-static_cast<long long>(unionArray[i].getU64Const()))); break;
default:
return 0;
}
......@@ -610,6 +633,14 @@ TIntermTyped* TIntermConstantUnion::fold(TOperator op, const TType& returnType)
case EOpUintBitsToFloat:
case EOpDoubleBitsToInt64:
case EOpDoubleBitsToUint64:
case EOpInt64BitsToDouble:
case EOpUint64BitsToDouble:
#ifdef AMD_EXTENSIONS
case EOpFloat16BitsToInt16:
case EOpFloat16BitsToUint16:
case EOpInt16BitsToFloat16:
case EOpUint16BitsToFloat16:
#endif
default:
return 0;
......@@ -702,6 +733,9 @@ TIntermTyped* TIntermediate::fold(TIntermAggregate* aggrNode)
#endif
children[0]->getAsTyped()->getBasicType() == EbtDouble;
bool isSigned = children[0]->getAsTyped()->getBasicType() == EbtInt ||
#ifdef AMD_EXTENSIONS
children[0]->getAsTyped()->getBasicType() == EbtInt16 ||
#endif
children[0]->getAsTyped()->getBasicType() == EbtInt64;
bool isInt64 = children[0]->getAsTyped()->getBasicType() == EbtInt64 ||
children[0]->getAsTyped()->getBasicType() == EbtUint64;
......
glslang/MachineIndependent/Initialize.cpp
......@@ -85,8 +85,6 @@ TBuiltIns::TBuiltIns()
prefixes[EbtFloat] = "";
prefixes[EbtInt] = "i";
prefixes[EbtUint] = "u";
prefixes[EbtInt64] = "i64";
prefixes[EbtUint64] = "u64";
postfixes[2] = "2";
postfixes[3] = "3";
postfixes[4] = "4";
......@@ -2612,6 +2610,157 @@ void TBuiltIns::initialize(int version, EProfile profile, const SpvVersion& spvV
"\n");
}
// GL_AMD_gpu_shader_int16
if (profile != EEsProfile && version >= 450) {
commonBuiltins.append(
"int16_t abs(int16_t);"
"i16vec2 abs(i16vec2);"
"i16vec3 abs(i16vec3);"
"i16vec4 abs(i16vec4);"
"int16_t sign(int16_t);"
"i16vec2 sign(i16vec2);"
"i16vec3 sign(i16vec3);"
"i16vec4 sign(i16vec4);"
"int16_t min(int16_t, int16_t);"
"i16vec2 min(i16vec2, int16_t);"
"i16vec3 min(i16vec3, int16_t);"
"i16vec4 min(i16vec4, int16_t);"
"i16vec2 min(i16vec2, i16vec2);"
"i16vec3 min(i16vec3, i16vec3);"
"i16vec4 min(i16vec4, i16vec4);"
"uint16_t min(uint16_t, uint16_t);"
"u16vec2 min(u16vec2, uint16_t);"
"u16vec3 min(u16vec3, uint16_t);"
"u16vec4 min(u16vec4, uint16_t);"
"u16vec2 min(u16vec2, u16vec2);"
"u16vec3 min(u16vec3, u16vec3);"
"u16vec4 min(u16vec4, u16vec4);"
"int16_t max(int16_t, int16_t);"
"i16vec2 max(i16vec2, int16_t);"
"i16vec3 max(i16vec3, int16_t);"
"i16vec4 max(i16vec4, int16_t);"
"i16vec2 max(i16vec2, i16vec2);"
"i16vec3 max(i16vec3, i16vec3);"
"i16vec4 max(i16vec4, i16vec4);"
"uint16_t max(uint16_t, uint16_t);"
"u16vec2 max(u16vec2, uint16_t);"
"u16vec3 max(u16vec3, uint16_t);"
"u16vec4 max(u16vec4, uint16_t);"
"u16vec2 max(u16vec2, u16vec2);"
"u16vec3 max(u16vec3, u16vec3);"
"u16vec4 max(u16vec4, u16vec4);"
"int16_t clamp(int16_t, int16_t, int16_t);"
"i16vec2 clamp(i16vec2, int16_t, int16_t);"
"i16vec3 clamp(i16vec3, int16_t, int16_t);"
"i16vec4 clamp(i16vec4, int16_t, int16_t);"
"i16vec2 clamp(i16vec2, i16vec2, i16vec2);"
"i16vec3 clamp(i16vec3, i16vec3, i16vec3);"
"i16vec4 clamp(i16vec4, i16vec4, i16vec4);"
"uint16_t clamp(uint16_t, uint16_t, uint16_t);"
"u16vec2 clamp(u16vec2, uint16_t, uint16_t);"
"u16vec3 clamp(u16vec3, uint16_t, uint16_t);"
"u16vec4 clamp(u16vec4, uint16_t, uint16_t);"
"u16vec2 clamp(u16vec2, u16vec2, u16vec2);"
"u16vec3 clamp(u16vec3, u16vec3, u16vec3);"
"u16vec4 clamp(u16vec4, u16vec4, u16vec4);"
"int16_t mix(int16_t, int16_t, bool);"
"i16vec2 mix(i16vec2, i16vec2, bvec2);"
"i16vec3 mix(i16vec3, i16vec3, bvec3);"
"i16vec4 mix(i16vec4, i16vec4, bvec4);"
"uint16_t mix(uint16_t, uint16_t, bool);"
"u16vec2 mix(u16vec2, u16vec2, bvec2);"
"u16vec3 mix(u16vec3, u16vec3, bvec3);"
"u16vec4 mix(u16vec4, u16vec4, bvec4);"
"float16_t frexp(float16_t, out int16_t);"
"f16vec2 frexp(f16vec2, out i16vec2);"
"f16vec3 frexp(f16vec3, out i16vec3);"
"f16vec4 frexp(f16vec4, out i16vec4);"
"float16_t ldexp(float16_t, int16_t);"
"f16vec2 ldexp(f16vec2, i16vec2);"
"f16vec3 ldexp(f16vec3, i16vec3);"
"f16vec4 ldexp(f16vec4, i16vec4);"
"int16_t float16BitsToInt16(float16_t);"
"i16vec2 float16BitsToInt16(f16vec2);"
"i16vec3 float16BitsToInt16(f16vec3);"
"i16vec4 float16BitsToInt16(f16vec4);"
"uint16_t float16BitsToUint16(float16_t);"
"u16vec2 float16BitsToUint16(f16vec2);"
"u16vec3 float16BitsToUint16(f16vec3);"
"u16vec4 float16BitsToUint16(f16vec4);"
"float16_t int16BitsToFloat16(int16_t);"
"f16vec2 int16BitsToFloat16(i16vec2);"
"f16vec3 int16BitsToFloat16(i16vec3);"
"f16vec4 int16BitsToFloat16(i16vec4);"
"float16_t uint16BitsToFloat16(uint16_t);"
"f16vec2 uint16BitsToFloat16(u16vec2);"
"f16vec3 uint16BitsToFloat16(u16vec3);"
"f16vec4 uint16BitsToFloat16(u16vec4);"
"int packInt2x16(i16vec2);"
"uint packUint2x16(u16vec2);"
"int64_t packInt4x16(i16vec4);"
"uint64_t packUint4x16(u16vec4);"
"i16vec2 unpackInt2x16(int);"
"u16vec2 unpackUint2x16(uint);"
"i16vec4 unpackInt4x16(int64_t);"
"u16vec4 unpackUint4x16(uint64_t);"
"bvec2 lessThan(i16vec2, i16vec2);"
"bvec3 lessThan(i16vec3, i16vec3);"
"bvec4 lessThan(i16vec4, i16vec4);"
"bvec2 lessThan(u16vec2, u16vec2);"
"bvec3 lessThan(u16vec3, u16vec3);"
"bvec4 lessThan(u16vec4, u16vec4);"
"bvec2 lessThanEqual(i16vec2, i16vec2);"
"bvec3 lessThanEqual(i16vec3, i16vec3);"
"bvec4 lessThanEqual(i16vec4, i16vec4);"
"bvec2 lessThanEqual(u16vec2, u16vec2);"
"bvec3 lessThanEqual(u16vec3, u16vec3);"
"bvec4 lessThanEqual(u16vec4, u16vec4);"
"bvec2 greaterThan(i16vec2, i16vec2);"
"bvec3 greaterThan(i16vec3, i16vec3);"
"bvec4 greaterThan(i16vec4, i16vec4);"
"bvec2 greaterThan(u16vec2, u16vec2);"
"bvec3 greaterThan(u16vec3, u16vec3);"
"bvec4 greaterThan(u16vec4, u16vec4);"
"bvec2 greaterThanEqual(i16vec2, i16vec2);"
"bvec3 greaterThanEqual(i16vec3, i16vec3);"
"bvec4 greaterThanEqual(i16vec4, i16vec4);"
"bvec2 greaterThanEqual(u16vec2, u16vec2);"
"bvec3 greaterThanEqual(u16vec3, u16vec3);"
"bvec4 greaterThanEqual(u16vec4, u16vec4);"
"bvec2 equal(i16vec2, i16vec2);"
"bvec3 equal(i16vec3, i16vec3);"
"bvec4 equal(i16vec4, i16vec4);"
"bvec2 equal(u16vec2, u16vec2);"
"bvec3 equal(u16vec3, u16vec3);"
"bvec4 equal(u16vec4, u16vec4);"
"bvec2 notEqual(i16vec2, i16vec2);"
"bvec3 notEqual(i16vec3, i16vec3);"
"bvec4 notEqual(i16vec4, i16vec4);"
"bvec2 notEqual(u16vec2, u16vec2);"
"bvec3 notEqual(u16vec3, u16vec3);"
"bvec4 notEqual(u16vec4, u16vec4);"
"\n");
}
#endif
//============================================================================
......@@ -5639,6 +5788,12 @@ void TBuiltIns::identifyBuiltIns(int version, EProfile profile, const SpvVersion
symbolTable.relateToOperator("doubleBitsToUint64", EOpDoubleBitsToUint64);
symbolTable.relateToOperator("int64BitsToDouble", EOpInt64BitsToDouble);
symbolTable.relateToOperator("uint64BitsToDouble", EOpUint64BitsToDouble);
#ifdef AMD_EXTENSIONS
symbolTable.relateToOperator("float16BitsToInt16", EOpFloat16BitsToInt16);
symbolTable.relateToOperator("float16BitsToUint16", EOpFloat16BitsToUint16);
symbolTable.relateToOperator("int16BitsToFloat16", EOpInt16BitsToFloat16);
symbolTable.relateToOperator("uint16BitsToFloat16", EOpUint16BitsToFloat16);
#endif
symbolTable.relateToOperator("packSnorm2x16", EOpPackSnorm2x16);
symbolTable.relateToOperator("unpackSnorm2x16", EOpUnpackSnorm2x16);
......@@ -5662,6 +5817,16 @@ void TBuiltIns::identifyBuiltIns(int version, EProfile profile, const SpvVersion
symbolTable.relateToOperator("unpackUint2x32", EOpUnpackUint2x32);
#ifdef AMD_EXTENSIONS
symbolTable.relateToOperator("packInt2x16", EOpPackInt2x16);
symbolTable.relateToOperator("unpackInt2x16", EOpUnpackInt2x16);
symbolTable.relateToOperator("packUint2x16", EOpPackUint2x16);
symbolTable.relateToOperator("unpackUint2x16", EOpUnpackUint2x16);
symbolTable.relateToOperator("packInt4x16", EOpPackInt4x16);
symbolTable.relateToOperator("unpackInt4x16", EOpUnpackInt4x16);
symbolTable.relateToOperator("packUint4x16", EOpPackUint4x16);
symbolTable.relateToOperator("unpackUint4x16", EOpUnpackUint4x16);
symbolTable.relateToOperator("packFloat2x16", EOpPackFloat2x16);
symbolTable.relateToOperator("unpackFloat2x16", EOpUnpackFloat2x16);
#endif
......
glslang/MachineIndependent/ParseHelper.cpp
......@@ -2220,6 +2220,10 @@ bool TParseContext::constructorError(const TSourceLoc& loc, TIntermNode* node, T
case EOpConstructUint:
case EOpConstructInt64:
case EOpConstructUint64:
#ifdef AMD_EXTENSIONS
case EOpConstructInt16:
case EOpConstructUint16:
#endif
case EOpConstructBool:
case EOpConstructBVec2:
case EOpConstructBVec3:
......@@ -2236,6 +2240,14 @@ bool TParseContext::constructorError(const TSourceLoc& loc, TIntermNode* node, T
case EOpConstructU64Vec2:
case EOpConstructU64Vec3:
case EOpConstructU64Vec4:
#ifdef AMD_EXTENSIONS
case EOpConstructI16Vec2:
case EOpConstructI16Vec3:
case EOpConstructI16Vec4:
case EOpConstructU16Vec2:
case EOpConstructU16Vec3:
case EOpConstructU16Vec4:
#endif
// This was the list of valid ones, if they aren't converting from float
// and aren't making an array.
makeSpecConst = ! floatArgument && ! type.isArray();
......@@ -2528,6 +2540,9 @@ void TParseContext::globalQualifierTypeCheck(const TSourceLoc& loc, const TQuali
}
if (publicType.basicType == EbtInt || publicType.basicType == EbtUint ||
#ifdef AMD_EXTENSIONS
publicType.basicType == EbtInt16 || publicType.basicType == EbtUint16 ||
#endif
publicType.basicType == EbtInt64 || publicType.basicType == EbtUint64 ||
publicType.basicType == EbtDouble)
profileRequires(loc, EEsProfile, 300, nullptr, "shader input/output");
......@@ -2538,6 +2553,9 @@ void TParseContext::globalQualifierTypeCheck(const TSourceLoc& loc, const TQuali
if (!qualifier.flat) {
#endif
if (publicType.basicType == EbtInt || publicType.basicType == EbtUint ||
#ifdef AMD_EXTENSIONS
publicType.basicType == EbtInt16 || publicType.basicType == EbtUint16 ||
#endif
publicType.basicType == EbtInt64 || publicType.basicType == EbtUint64 ||
publicType.basicType == EbtDouble ||
(publicType.userDef && (publicType.userDef->containsBasicType(EbtInt) ||
......@@ -4634,6 +4652,10 @@ void TParseContext::layoutTypeCheck(const TSourceLoc& loc, const TType& type)
case EbtUint:
case EbtInt64:
case EbtUint64:
#ifdef AMD_EXTENSIONS
case EbtInt16:
case EbtUint16:
#endif
case EbtBool:
case EbtFloat:
case EbtDouble:
......@@ -5580,6 +5602,22 @@ TIntermTyped* TParseContext::constructBuiltIn(const TType& type, TOperator op, T
basicOp = EOpConstructUint64;
break;
#ifdef AMD_EXTENSIONS
case EOpConstructI16Vec2:
case EOpConstructI16Vec3:
case EOpConstructI16Vec4:
case EOpConstructInt16:
basicOp = EOpConstructInt16;
break;
case EOpConstructU16Vec2:
case EOpConstructU16Vec3:
case EOpConstructU16Vec4:
case EOpConstructUint16:
basicOp = EOpConstructUint16;
break;
#endif
case EOpConstructBVec2:
case EOpConstructBVec3:
case EOpConstructBVec4:
......
glslang/MachineIndependent/Scan.cpp
......@@ -464,6 +464,15 @@ void TScanContext::fillInKeywordMap()
(*KeywordMap)["u64vec4"] = U64VEC4;
#ifdef AMD_EXTENSIONS
(*KeywordMap)["int16_t"] = INT16_T;
(*KeywordMap)["uint16_t"] = UINT16_T;
(*KeywordMap)["i16vec2"] = I16VEC2;
(*KeywordMap)["i16vec3"] = I16VEC3;
(*KeywordMap)["i16vec4"] = I16VEC4;
(*KeywordMap)["u16vec2"] = U16VEC2;
(*KeywordMap)["u16vec3"] = U16VEC3;
(*KeywordMap)["u16vec4"] = U16VEC4;
(*KeywordMap)["float16_t"] = FLOAT16_T;
(*KeywordMap)["f16vec2"] = F16VEC2;
(*KeywordMap)["f16vec3"] = F16VEC3;
......@@ -709,6 +718,10 @@ int TScanContext::tokenize(TPpContext* pp, TParserToken& token)
case PpAtomConstUint: parserToken->sType.lex.i = ppToken.ival; return UINTCONSTANT;
case PpAtomConstInt64: parserToken->sType.lex.i64 = ppToken.i64val; return INT64CONSTANT;
case PpAtomConstUint64: parserToken->sType.lex.i64 = ppToken.i64val; return UINT64CONSTANT;
#ifdef AMD_EXTENSIONS
case PpAtomConstInt16: parserToken->sType.lex.i = ppToken.ival; return INT16CONSTANT;
case PpAtomConstUint16: parserToken->sType.lex.i = ppToken.ival; return UINT16CONSTANT;
#endif
case PpAtomConstFloat: parserToken->sType.lex.d = ppToken.dval; return FLOATCONSTANT;
case PpAtomConstDouble: parserToken->sType.lex.d = ppToken.dval; return DOUBLECONSTANT;
#ifdef AMD_EXTENSIONS
......@@ -973,6 +986,21 @@ int TScanContext::tokenizeIdentifier()
return identifierOrType();
#ifdef AMD_EXTENSIONS
case INT16_T:
case UINT16_T:
case I16VEC2:
case I16VEC3:
case I16VEC4:
case U16VEC2:
case U16VEC3:
case U16VEC4:
afterType = true;
if (parseContext.symbolTable.atBuiltInLevel() ||
(parseContext.extensionTurnedOn(E_GL_AMD_gpu_shader_int16) &&
parseContext.profile != EEsProfile && parseContext.version >= 450))
return keyword;
return identifierOrType();
case FLOAT16_T:
case F16VEC2:
case F16VEC3:
......
glslang/MachineIndependent/SymbolTable.cpp
......@@ -67,6 +67,10 @@ void TType::buildMangledName(TString& mangledName) const
case EbtUint: mangledName += 'u'; break;
case EbtInt64: mangledName += "i64"; break;
case EbtUint64: mangledName += "u64"; break;
#ifdef AMD_EXTENSIONS
case EbtInt16: mangledName += "i16"; break;
case EbtUint16: mangledName += "u16"; break;
#endif
case EbtBool: mangledName += 'b'; break;
case EbtAtomicUint: mangledName += "au"; break;
case EbtSampler:
......
glslang/MachineIndependent/Versions.cpp
......@@ -195,6 +195,7 @@ void TParseVersions::initializeExtensionBehavior()
extensionBehavior[E_GL_AMD_gcn_shader] = EBhDisable;
extensionBehavior[E_GL_AMD_gpu_shader_half_float] = EBhDisable;
extensionBehavior[E_GL_AMD_texture_gather_bias_lod] = EBhDisable;
extensionBehavior[E_GL_AMD_gpu_shader_int16] = EBhDisable;
#endif
#ifdef NV_EXTENSIONS
......@@ -318,6 +319,7 @@ void TParseVersions::getPreamble(std::string& preamble)
"#define GL_AMD_gcn_shader 1\n"
"#define GL_AMD_gpu_shader_half_float 1\n"
"#define GL_AMD_texture_gather_bias_lod 1\n"
"#define GL_AMD_gpu_shader_int16 1\n"
#endif
#ifdef NV_EXTENSIONS
......@@ -710,10 +712,21 @@ void TParseVersions::doubleCheck(const TSourceLoc& loc, const char* op)
}
#ifdef AMD_EXTENSIONS
// Call for any operation needing float16 data-type support.
// Call for any operation needing GLSL 16-bit integer data-type support.
void TParseVersions::int16Check(const TSourceLoc& loc, const char* op, bool builtIn)
{
if (! builtIn) {
requireExtensions(loc, 1, &E_GL_AMD_gpu_shader_int16, "shader int16");
requireProfile(loc, ECoreProfile | ECompatibilityProfile, op);
profileRequires(loc, ECoreProfile, 450, nullptr, op);
profileRequires(loc, ECompatibilityProfile, 450, nullptr, op);
}
}
// Call for any operation needing GLSL float16 data-type support.
void TParseVersions::float16Check(const TSourceLoc& loc, const char* op, bool builtIn)
{
if (!builtIn) {
if (! builtIn) {
requireExtensions(loc, 1, &E_GL_AMD_gpu_shader_half_float, "shader half float");
requireProfile(loc, ECoreProfile | ECompatibilityProfile, op);
profileRequires(loc, ECoreProfile, 450, nullptr, op);
......
glslang/MachineIndependent/Versions.h
......@@ -147,7 +147,9 @@ const char* const E_GL_AMD_shader_explicit_vertex_parameter = "GL_AMD_shader
const char* const E_GL_AMD_gcn_shader = "GL_AMD_gcn_shader";
const char* const E_GL_AMD_gpu_shader_half_float = "GL_AMD_gpu_shader_half_float";
const char* const E_GL_AMD_texture_gather_bias_lod = "GL_AMD_texture_gather_bias_lod";
const char* const E_GL_AMD_gpu_shader_int16 = "GL_AMD_gpu_shader_int16";
#endif
#ifdef NV_EXTENSIONS
const char* const E_GL_NV_sample_mask_override_coverage = "GL_NV_sample_mask_override_coverage";
......
glslang/MachineIndependent/glslang.y
......@@ -121,7 +121,7 @@ extern int yylex(YYSTYPE*, TParseContext&);
%expect 1 // One shift reduce conflict because of if | else
%token <lex> ATTRIBUTE VARYING
%token <lex> CONST BOOL FLOAT DOUBLE INT UINT INT64_T UINT64_T FLOAT16_T
%token <lex> CONST BOOL FLOAT DOUBLE INT UINT INT64_T UINT64_T INT16_T UINT16_T FLOAT16_T
%token <lex> BREAK CONTINUE DO ELSE FOR IF DISCARD RETURN SWITCH CASE DEFAULT SUBROUTINE
%token <lex> BVEC2 BVEC3 BVEC4 IVEC2 IVEC3 IVEC4 I64VEC2 I64VEC3 I64VEC4 UVEC2 UVEC3 UVEC4 U64VEC2 U64VEC3 U64VEC4 VEC2 VEC3 VEC4
%token <lex> MAT2 MAT3 MAT4 CENTROID IN OUT INOUT
......@@ -129,6 +129,7 @@ extern int yylex(YYSTYPE*, TParseContext&);
%token <lex> COHERENT VOLATILE RESTRICT READONLY WRITEONLY
%token <lex> DVEC2 DVEC3 DVEC4 DMAT2 DMAT3 DMAT4
%token <lex> F16VEC2 F16VEC3 F16VEC4 F16MAT2 F16MAT3 F16MAT4
%token <lex> I16VEC2 I16VEC3 I16VEC4 U16VEC2 U16VEC3 U16VEC4
%token <lex> NOPERSPECTIVE FLAT SMOOTH LAYOUT __EXPLICITINTERPAMD
%token <lex> MAT2X2 MAT2X3 MAT2X4
......@@ -188,7 +189,7 @@ extern int yylex(YYSTYPE*, TParseContext&);
%token <lex> STRUCT VOID WHILE
%token <lex> IDENTIFIER TYPE_NAME
%token <lex> FLOATCONSTANT DOUBLECONSTANT INTCONSTANT UINTCONSTANT INT64CONSTANT UINT64CONSTANT BOOLCONSTANT FLOAT16CONSTANT
%token <lex> FLOATCONSTANT DOUBLECONSTANT INTCONSTANT UINTCONSTANT INT64CONSTANT UINT64CONSTANT INT16CONSTANT UINT16CONSTANT BOOLCONSTANT FLOAT16CONSTANT
%token <lex> LEFT_OP RIGHT_OP
%token <lex> INC_OP DEC_OP LE_OP GE_OP EQ_OP NE_OP
%token <lex> AND_OP OR_OP XOR_OP MUL_ASSIGN DIV_ASSIGN ADD_ASSIGN
......@@ -273,6 +274,18 @@ primary_expression
parseContext.int64Check($1.loc, "64-bit unsigned integer literal");
$$ = parseContext.intermediate.addConstantUnion($1.u64, $1.loc, true);
}
| INT16CONSTANT {
#ifdef AMD_EXTENSIONS
parseContext.int16Check($1.loc, "16-bit integer literal");
$$ = parseContext.intermediate.addConstantUnion((short)$1.i, $1.loc, true);
#endif
}
| UINT16CONSTANT {
#ifdef AMD_EXTENSIONS
parseContext.int16Check($1.loc, "16-bit unsigned integer literal");
$$ = parseContext.intermediate.addConstantUnion((unsigned short)$1.u, $1.loc, true);
#endif
}
| FLOATCONSTANT {
$$ = parseContext.intermediate.addConstantUnion($1.d, EbtFloat, $1.loc, true);
}
......@@ -1363,6 +1376,20 @@ type_specifier_nonarray
$$.init($1.loc, parseContext.symbolTable.atGlobalLevel());
$$.basicType = EbtUint64;
}
| INT16_T {
#ifdef AMD_EXTENSIONS
parseContext.int16Check($1.loc, "16-bit integer", parseContext.symbolTable.atBuiltInLevel());
$$.init($1.loc, parseContext.symbolTable.atGlobalLevel());
$$.basicType = EbtInt16;
#endif
}
| UINT16_T {
#ifdef AMD_EXTENSIONS
parseContext.int16Check($1.loc, "16-bit unsigned integer", parseContext.symbolTable.atBuiltInLevel());
$$.init($1.loc, parseContext.symbolTable.atGlobalLevel());
$$.basicType = EbtUint16;
#endif
}
| BOOL {
$$.init($1.loc, parseContext.symbolTable.atGlobalLevel());
$$.basicType = EbtBool;
......@@ -1472,6 +1499,30 @@ type_specifier_nonarray
$$.basicType = EbtInt64;
$$.setVector(4);
}
| I16VEC2 {
#ifdef AMD_EXTENSIONS
parseContext.int16Check($1.loc, "16-bit integer vector", parseContext.symbolTable.atBuiltInLevel());
$$.init($1.loc, parseContext.symbolTable.atGlobalLevel());
$$.basicType = EbtInt16;
$$.setVector(2);
#endif
}
| I16VEC3 {
#ifdef AMD_EXTENSIONS
parseContext.int16Check($1.loc, "16-bit integer vector", parseContext.symbolTable.atBuiltInLevel());
$$.init($1.loc, parseContext.symbolTable.atGlobalLevel());
$$.basicType = EbtInt16;
$$.setVector(3);
#endif
}
| I16VEC4 {
#ifdef AMD_EXTENSIONS
parseContext.int16Check($1.loc, "16-bit integer vector", parseContext.symbolTable.atBuiltInLevel());
$$.init($1.loc, parseContext.symbolTable.atGlobalLevel());
$$.basicType = EbtInt16;
$$.setVector(4);
#endif
}
| UVEC2 {
parseContext.fullIntegerCheck($1.loc, "unsigned integer vector");
$$.init($1.loc, parseContext.symbolTable.atGlobalLevel());
......@@ -1508,6 +1559,30 @@ type_specifier_nonarray
$$.basicType = EbtUint64;
$$.setVector(4);
}
| U16VEC2 {
#ifdef AMD_EXTENSIONS
parseContext.int16Check($1.loc, "16-bit unsigned integer vector", parseContext.symbolTable.atBuiltInLevel());
$$.init($1.loc, parseContext.symbolTable.atGlobalLevel());
$$.basicType = EbtUint16;
$$.setVector(2);
#endif
}
| U16VEC3 {
#ifdef AMD_EXTENSIONS
parseContext.int16Check($1.loc, "16-bit unsigned integer vector", parseContext.symbolTable.atBuiltInLevel());
$$.init($1.loc, parseContext.symbolTable.atGlobalLevel());
$$.basicType = EbtUint16;
$$.setVector(3);
#endif
}
| U16VEC4 {
#ifdef AMD_EXTENSIONS
parseContext.int16Check($1.loc, "16-bit unsigned integer vector", parseContext.symbolTable.atBuiltInLevel());
$$.init($1.loc, parseContext.symbolTable.atGlobalLevel());
$$.basicType = EbtUint16;
$$.setVector(4);
#endif
}
| MAT2 {
$$.init($1.loc, parseContext.symbolTable.atGlobalLevel());
$$.basicType = EbtFloat;
......
glslang/MachineIndependent/glslang_tab.cpp
This source diff could not be displayed because it is too large. You can view the blob instead.
glslang/MachineIndependent/intermOut.cpp
......@@ -308,6 +308,13 @@ bool TOutputTraverser::visitUnary(TVisit /* visit */, TIntermUnary* node)
case EOpDoubleBitsToUint64: out.debug << "doubleBitsToUint64"; break;
case EOpInt64BitsToDouble: out.debug << "int64BitsToDouble"; break;
case EOpUint64BitsToDouble: out.debug << "uint64BitsToDouble"; break;
#ifdef AMD_EXTENSIONS
case EOpFloat16BitsToInt16: out.debug << "float16BitsToInt16"; break;
case EOpFloat16BitsToUint16: out.debug << "float16BitsToUint16"; break;
case EOpInt16BitsToFloat16: out.debug << "int16BitsToFloat16"; break;
case EOpUint16BitsToFloat16: out.debug << "uint16BitsToFloat16"; break;
#endif
case EOpPackSnorm2x16: out.debug << "packSnorm2x16"; break;
case EOpUnpackSnorm2x16:out.debug << "unpackSnorm2x16"; break;
case EOpPackUnorm2x16: out.debug << "packUnorm2x16"; break;
......@@ -328,6 +335,16 @@ bool TOutputTraverser::visitUnary(TVisit /* visit */, TIntermUnary* node)
case EOpUnpackUint2x32: out.debug << "unpackUint2x32"; break;
#ifdef AMD_EXTENSIONS
case EOpPackInt2x16: out.debug << "packInt2x16"; break;
case EOpUnpackInt2x16: out.debug << "unpackInt2x16"; break;
case EOpPackUint2x16: out.debug << "packUint2x16"; break;
case EOpUnpackUint2x16: out.debug << "unpackUint2x16"; break;
case EOpPackInt4x16: out.debug << "packInt4x16"; break;
case EOpUnpackInt4x16: out.debug << "unpackInt4x16"; break;
case EOpPackUint4x16: out.debug << "packUint4x16"; break;
case EOpUnpackUint4x16: out.debug << "unpackUint4x16"; break;
case EOpPackFloat2x16: out.debug << "packFloat2x16"; break;
case EOpUnpackFloat2x16: out.debug << "unpackFloat2x16"; break;
#endif
......@@ -433,6 +450,42 @@ bool TOutputTraverser::visitUnary(TVisit /* visit */, TIntermUnary* node)
case EOpConvFloat16ToDouble: out.debug << "Convert float16 to double"; break;
case EOpConvFloat16ToInt64: out.debug << "Convert float16 to int64"; break;
case EOpConvFloat16ToUint64: out.debug << "Convert float16 to uint64"; break;
case EOpConvBoolToInt16: out.debug << "Convert bool to int16"; break;
case EOpConvIntToInt16: out.debug << "Convert int to int16"; break;
case EOpConvUintToInt16: out.debug << "Convert uint to int16"; break;
case EOpConvFloatToInt16: out.debug << "Convert float to int16"; break;
case EOpConvDoubleToInt16: out.debug << "Convert double to int16"; break;
case EOpConvFloat16ToInt16: out.debug << "Convert float16 to int16"; break;
case EOpConvInt64ToInt16: out.debug << "Convert int64 to int16"; break;
case EOpConvUint64ToInt16: out.debug << "Convert uint64 to int16"; break;
case EOpConvUint16ToInt16: out.debug << "Convert uint16 to int16"; break;
case EOpConvInt16ToBool: out.debug << "Convert int16 to bool"; break;
case EOpConvInt16ToInt: out.debug << "Convert int16 to int"; break;
case EOpConvInt16ToUint: out.debug << "Convert int16 to uint"; break;
case EOpConvInt16ToFloat: out.debug << "Convert int16 to float"; break;
case EOpConvInt16ToDouble: out.debug << "Convert int16 to double"; break;
case EOpConvInt16ToFloat16: out.debug << "Convert int16 to float16"; break;
case EOpConvInt16ToInt64: out.debug << "Convert int16 to int64"; break;
case EOpConvInt16ToUint64: out.debug << "Convert int16 to uint64"; break;
case EOpConvBoolToUint16: out.debug << "Convert bool to uint16"; break;
case EOpConvIntToUint16: out.debug << "Convert int to uint16"; break;
case EOpConvUintToUint16: out.debug << "Convert uint to uint16"; break;
case EOpConvFloatToUint16: out.debug << "Convert float to uint16"; break;
case EOpConvDoubleToUint16: out.debug << "Convert double to uint16"; break;
case EOpConvFloat16ToUint16: out.debug << "Convert float16 to uint16"; break;
case EOpConvInt64ToUint16: out.debug << "Convert int64 to uint16"; break;
case EOpConvUint64ToUint16: out.debug << "Convert uint64 to uint16"; break;
case EOpConvInt16ToUint16: out.debug << "Convert int16 to uint16"; break;
case EOpConvUint16ToBool: out.debug << "Convert uint16 to bool"; break;
case EOpConvUint16ToInt: out.debug << "Convert uint16 to int"; break;
case EOpConvUint16ToUint: out.debug << "Convert uint16 to uint"; break;
case EOpConvUint16ToFloat: out.debug << "Convert uint16 to float"; break;
case EOpConvUint16ToDouble: out.debug << "Convert uint16 to double"; break;
case EOpConvUint16ToFloat16: out.debug << "Convert uint16 to float16"; break;
case EOpConvUint16ToInt64: out.debug << "Convert uint16 to int64"; break;
case EOpConvUint16ToUint64: out.debug << "Convert uint16 to uint64"; break;
#endif
default: out.debug.message(EPrefixError, "Bad unary op");
......@@ -466,9 +519,13 @@ bool TOutputTraverser::visitAggregate(TVisit /* visit */, TIntermAggregate* node
case EOpConstructFloat: out.debug << "Construct float"; break;
case EOpConstructDouble:out.debug << "Construct double"; break;
case EOpConstructVec2: out.debug << "Construct vec2"; break;
case EOpConstructVec3: out.debug << "Construct vec3"; break;
case EOpConstructVec4: out.debug << "Construct vec4"; break;
case EOpConstructDVec2: out.debug << "Construct dvec2"; break;
case EOpConstructDVec3: out.debug << "Construct dvec3"; break;
case EOpConstructDVec4: out.debug << "Construct dvec4"; break;
case EOpConstructBool: out.debug << "Construct bool"; break;
case EOpConstructBVec2: out.debug << "Construct bvec2"; break;
case EOpConstructBVec3: out.debug << "Construct bvec3"; break;
......@@ -489,6 +546,16 @@ bool TOutputTraverser::visitAggregate(TVisit /* visit */, TIntermAggregate* node
case EOpConstructU64Vec2: out.debug << "Construct u64vec2"; break;
case EOpConstructU64Vec3: out.debug << "Construct u64vec3"; break;
case EOpConstructU64Vec4: out.debug << "Construct u64vec4"; break;
#ifdef AMD_EXTENSIONS
case EOpConstructInt16: out.debug << "Construct int16_t"; break;
case EOpConstructI16Vec2: out.debug << "Construct i16vec2"; break;
case EOpConstructI16Vec3: out.debug << "Construct i16vec3"; break;
case EOpConstructI16Vec4: out.debug << "Construct i16vec4"; break;
case EOpConstructUint16: out.debug << "Construct uint16_t"; break;
case EOpConstructU16Vec2: out.debug << "Construct u16vec2"; break;
case EOpConstructU16Vec3: out.debug << "Construct u16vec3"; break;
case EOpConstructU16Vec4: out.debug << "Construct u16vec4"; break;
#endif
case EOpConstructMat2x2: out.debug << "Construct mat2"; break;
case EOpConstructMat2x3: out.debug << "Construct mat2x3"; break;
case EOpConstructMat2x4: out.debug << "Construct mat2x4"; break;
......@@ -827,6 +894,26 @@ static void OutputConstantUnion(TInfoSink& out, const TIntermTyped* node, const
out.debug << buf << "\n";
}
break;
#ifdef AMD_EXTENSIONS
case EbtInt16:
{
const int maxSize = 300;
char buf[maxSize];
snprintf(buf, maxSize, "%d (%s)", constUnion[i].getIConst(), "const int16_t");
out.debug << buf << "\n";
}
break;
case EbtUint16:
{
const int maxSize = 300;
char buf[maxSize];
snprintf(buf, maxSize, "%u (%s)", constUnion[i].getUConst(), "const uint16_t");
out.debug << buf << "\n";
}
break;
#endif
default:
out.info.message(EPrefixInternalError, "Unknown constant", node->getLoc());
break;
......
glslang/MachineIndependent/linkValidate.cpp
......@@ -1057,6 +1057,8 @@ int TIntermediate::getBaseAlignmentScalar(const TType& type, int& size)
case EbtUint64:
case EbtDouble: size = 8; return 8;
#ifdef AMD_EXTENSIONS
case EbtInt16:
case EbtUint16:
case EbtFloat16: size = 2; return 2;
#endif
default: size = 4; return 4;
......
glslang/MachineIndependent/localintermediate.h
......@@ -285,6 +285,11 @@ public:
TIntermConstantUnion* addConstantUnion(unsigned int, const TSourceLoc&, bool literal = false) const;
TIntermConstantUnion* addConstantUnion(long long, const TSourceLoc&, bool literal = false) const;
TIntermConstantUnion* addConstantUnion(unsigned long long, const TSourceLoc&, bool literal = false) const;
#ifdef AMD_EXTENSIONS
TIntermConstantUnion* addConstantUnion(short, const TSourceLoc&, bool literal = false) const;
TIntermConstantUnion* addConstantUnion(unsigned short, const TSourceLoc&, bool literal = false) const;
#endif
TIntermConstantUnion* addConstantUnion(bool, const TSourceLoc&, bool literal = false) const;
TIntermConstantUnion* addConstantUnion(double, TBasicType, const TSourceLoc&, bool literal = false) const;
TIntermConstantUnion* addConstantUnion(const TString*, const TSourceLoc&, bool literal = false) const;
......
glslang/MachineIndependent/parseVersions.h
......@@ -78,6 +78,7 @@ public:
virtual void fullIntegerCheck(const TSourceLoc&, const char* op);
virtual void doubleCheck(const TSourceLoc&, const char* op);
#ifdef AMD_EXTENSIONS
virtual void int16Check(const TSourceLoc& loc, const char* op, bool builtIn = false);
virtual void float16Check(const TSourceLoc&, const char* op, bool builtIn = false);
#endif
virtual void int64Check(const TSourceLoc&, const char* op, bool builtIn = false);
......
glslang/MachineIndependent/preprocessor/Pp.cpp
......@@ -724,6 +724,7 @@ int TPpContext::CPPerror(TPpToken* ppToken)
if (token == PpAtomConstInt || token == PpAtomConstUint ||
token == PpAtomConstInt64 || token == PpAtomConstUint64 ||
#ifdef AMD_EXTENSIONS
token == PpAtomConstInt16 || token == PpAtomConstUint16 ||
token == PpAtomConstFloat16 ||
#endif
token == PpAtomConstFloat || token == PpAtomConstDouble) {
......@@ -758,6 +759,10 @@ int TPpContext::CPPpragma(TPpToken* ppToken)
case PpAtomConstUint:
case PpAtomConstInt64:
case PpAtomConstUint64:
#ifdef AMD_EXTENSIONS
case PpAtomConstInt16:
case PpAtomConstUint16:
#endif
case PpAtomConstFloat:
case PpAtomConstDouble:
#ifdef AMD_EXTENSIONS
......
glslang/MachineIndependent/preprocessor/PpTokens.cpp
......@@ -141,6 +141,10 @@ void TPpContext::TokenStream::putToken(int token, TPpToken* ppToken)
case PpAtomConstUint:
case PpAtomConstInt64:
case PpAtomConstUint64:
#ifdef AMD_EXTENSIONS
case PpAtomConstInt16:
case PpAtomConstUint16:
#endif
case PpAtomConstFloat:
case PpAtomConstDouble:
#ifdef AMD_EXTENSIONS
......@@ -190,6 +194,10 @@ int TPpContext::TokenStream::getToken(TParseContextBase& parseContext, TPpToken
case PpAtomConstUint:
case PpAtomConstInt64:
case PpAtomConstUint64:
#ifdef AMD_EXTENSIONS
case PpAtomConstInt16:
case PpAtomConstUint16:
#endif
len = 0;
ch = getSubtoken();
while (ch != 0 && ch != EndOfInput) {
......@@ -217,6 +225,9 @@ int TPpContext::TokenStream::getToken(TParseContextBase& parseContext, TPpToken
ppToken->dval = atof(ppToken->name);
break;
case PpAtomConstInt:
#ifdef AMD_EXTENSIONS
case PpAtomConstInt16:
#endif
if (len > 0 && ppToken->name[0] == '0') {
if (len > 1 && (ppToken->name[1] == 'x' || ppToken->name[1] == 'X'))
ppToken->ival = (int)strtol(ppToken->name, 0, 16);
......@@ -226,6 +237,9 @@ int TPpContext::TokenStream::getToken(TParseContextBase& parseContext, TPpToken
ppToken->ival = atoi(ppToken->name);
break;
case PpAtomConstUint:
#ifdef AMD_EXTENSIONS
case PpAtomConstUint16:
#endif
if (len > 0 && ppToken->name[0] == '0') {
if (len > 1 && (ppToken->name[1] == 'x' || ppToken->name[1] == 'X'))
ppToken->ival = (int)strtoul(ppToken->name, 0, 16);
......
glslang/MachineIndependent/preprocessor/PpTokens.h
......@@ -127,6 +127,10 @@ enum EFixedAtoms {
PpAtomConstUint,
PpAtomConstInt64,
PpAtomConstUint64,
#ifdef AMD_EXTENSIONS
PpAtomConstInt16,
PpAtomConstUint16,
#endif
PpAtomConstFloat,
PpAtomConstDouble,
PpAtomConstFloat16,
......
gtests/Spv.FromFile.cpp
......@@ -396,6 +396,7 @@ INSTANTIATE_TEST_CASE_P(
Glsl, CompileVulkanToSpirvTestAMD,
::testing::ValuesIn(std::vector<std::string>({
"spv.float16.frag",
"spv.int16.frag",
"spv.shaderBallotAMD.comp",
"spv.textureGatherBiasLod.frag"
})),
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment