diff options
author | Andrzej Janik <[email protected]> | 2020-11-06 00:56:45 +0100 |
---|---|---|
committer | Andrzej Janik <[email protected]> | 2020-11-06 00:56:45 +0100 |
commit | ac6265f257654180f6661c406a025313190448c4 (patch) | |
tree | df84117141e484c0c9da03a94aae4b5018c24607 | |
parent | d7bf1acf84faa8f6cb1d5edb6c4d9eb0f05a5ae0 (diff) | |
download | ZLUDA-ac6265f257654180f6661c406a025313190448c4.tar.gz ZLUDA-ac6265f257654180f6661c406a025313190448c4.zip |
Implement instructions bfe, rem, xor
-rw-r--r-- | ptx/lib/notcuda_ptx_impl.cl | 22 | ||||
-rw-r--r-- | ptx/lib/notcuda_ptx_impl.spv | bin | 48348 -> 49396 bytes | |||
-rw-r--r-- | ptx/src/ast.rs | 19 | ||||
-rw-r--r-- | ptx/src/ptx.lalrpop | 63 | ||||
-rw-r--r-- | ptx/src/test/spirv_run/bfe.ptx | 23 | ||||
-rw-r--r-- | ptx/src/test/spirv_run/bfe.spvtxt | 70 | ||||
-rw-r--r-- | ptx/src/test/spirv_run/mod.rs | 14 | ||||
-rw-r--r-- | ptx/src/test/spirv_run/rem.ptx | 23 | ||||
-rw-r--r-- | ptx/src/test/spirv_run/rem.spvtxt | 55 | ||||
-rw-r--r-- | ptx/src/test/spirv_run/xor.ptx | 23 | ||||
-rw-r--r-- | ptx/src/test/spirv_run/xor.spvtxt | 55 | ||||
-rw-r--r-- | ptx/src/translate.rs | 249 |
12 files changed, 576 insertions, 40 deletions
diff --git a/ptx/lib/notcuda_ptx_impl.cl b/ptx/lib/notcuda_ptx_impl.cl index a0d487b..4249f2b 100644 --- a/ptx/lib/notcuda_ptx_impl.cl +++ b/ptx/lib/notcuda_ptx_impl.cl @@ -1,5 +1,5 @@ // Every time this file changes it must te rebuilt:
-// ocloc -file notcuda_ptx_impl.cl -64 -options "-cl-std=CL2.0" -out_dir . -device kbl -output_no_suffix -spv_only
+// ocloc -file notcuda_ptx_impl.cl -64 -options "-cl-std=CL2.0 -Dcl_intel_bit_instructions" -out_dir . -device kbl -output_no_suffix -spv_only
// Additionally you should strip names:
// spirv-opt --strip-debug notcuda_ptx_impl.spv -o notcuda_ptx_impl.spv
@@ -119,3 +119,23 @@ atomic_dec(atom_relaxed_sys_shared_dec, memory_order_relaxed, memory_order_relax atomic_dec(atom_acquire_sys_shared_dec, memory_order_acquire, memory_order_acquire, memory_scope_device, __local);
atomic_dec(atom_release_sys_shared_dec, memory_order_release, memory_order_acquire, memory_scope_device, __local);
atomic_dec(atom_acq_rel_sys_shared_dec, memory_order_acq_rel, memory_order_acquire, memory_scope_device, __local);
+
+uint FUNC(bfe_u32)(uint base, uint pos, uint len)
+{
+ return intel_ubfe(base, pos, len);
+}
+
+ulong FUNC(bfe_u64)(ulong base, uint pos, uint len)
+{
+ return intel_ubfe(base, pos, len);
+}
+
+int FUNC(bfe_s32)(int base, uint pos, uint len)
+{
+ return intel_sbfe(base, pos, len);
+}
+
+long FUNC(bfe_s64)(long base, uint pos, uint len)
+{
+ return intel_sbfe(base, pos, len);
+}
\ No newline at end of file diff --git a/ptx/lib/notcuda_ptx_impl.spv b/ptx/lib/notcuda_ptx_impl.spv Binary files differindex 36f37bb..1ef470f 100644 --- a/ptx/lib/notcuda_ptx_impl.spv +++ b/ptx/lib/notcuda_ptx_impl.spv diff --git a/ptx/src/ast.rs b/ptx/src/ast.rs index b6ac3db..5a5f6be 100644 --- a/ptx/src/ast.rs +++ b/ptx/src/ast.rs @@ -558,7 +558,7 @@ pub enum Instruction<P: ArgParams> { Add(ArithDetails, Arg3<P>), Setp(SetpData, Arg4Setp<P>), SetpBool(SetpBoolData, Arg5<P>), - Not(NotType, Arg2<P>), + Not(BooleanType, Arg2<P>), Bra(BraData, Arg1<P>), Cvt(CvtDetails, Arg2<P>), Cvta(CvtaDetails, Arg2<P>), @@ -569,12 +569,12 @@ pub enum Instruction<P: ArgParams> { Call(CallInst<P>), Abs(AbsDetails, Arg2<P>), Mad(MulDetails, Arg4<P>), - Or(OrAndType, Arg3<P>), + Or(BooleanType, Arg3<P>), Sub(ArithDetails, Arg3<P>), Min(MinMaxDetails, Arg3<P>), Max(MinMaxDetails, Arg3<P>), Rcp(RcpDetails, Arg2<P>), - And(OrAndType, Arg3<P>), + And(BooleanType, Arg3<P>), Selp(SelpType, Arg4<P>), Bar(BarDetails, Arg1Bar<P>), Atom(AtomDetails, Arg3<P>), @@ -590,6 +590,9 @@ pub enum Instruction<P: ArgParams> { Clz { typ: BitType, arg: Arg2<P> }, Brev { typ: BitType, arg: Arg2<P> }, Popc { typ: BitType, arg: Arg2<P> }, + Xor { typ: BooleanType, arg: Arg3<P> }, + Bfe { typ: IntType, arg: Arg4<P> }, + Rem { typ: IntType, arg: Arg3<P> }, } #[derive(Copy, Clone)] @@ -896,14 +899,6 @@ pub struct SetpBoolData { pub bool_op: SetpBoolPostOp, } -#[derive(PartialEq, Eq, Copy, Clone)] -pub enum NotType { - Pred, - B16, - B32, - B64, -} - pub struct BraData { pub uniform: bool, } @@ -1058,7 +1053,7 @@ pub struct RetData { pub uniform: bool, } -sub_enum!(OrAndType { +sub_enum!(BooleanType { Pred, B16, B32, diff --git a/ptx/src/ptx.lalrpop b/ptx/src/ptx.lalrpop index cd1c642..6c231b2 100644 --- a/ptx/src/ptx.lalrpop +++ b/ptx/src/ptx.lalrpop @@ -142,6 +142,7 @@ match { "atom", "bar", "barrier", + "bfe", "bra", "brev", "call", @@ -166,6 +167,7 @@ match { "or", "popc", "rcp", + "rem", "ret", "rsqrt", "selp", @@ -179,6 +181,7 @@ match { "sub", "texmode_independent", "texmode_unified", + "xor", } else { // https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#identifiers r"[a-zA-Z][a-zA-Z0-9_$]*|[_$%][a-zA-Z0-9_$]+" => ID, @@ -192,6 +195,7 @@ ExtendedID : &'input str = { "atom", "bar", "barrier", + "bfe", "bra", "brev", "call", @@ -216,6 +220,7 @@ ExtendedID : &'input str = { "or", "popc", "rcp", + "rem", "ret", "rsqrt", "selp", @@ -229,6 +234,7 @@ ExtendedID : &'input str = { "sub", "texmode_independent", "texmode_unified", + "xor", ID } @@ -708,6 +714,9 @@ Instruction: ast::Instruction<ast::ParsedArgParams<'input>> = { InstClz, InstBrev, InstPopc, + InstXor, + InstRem, + InstBfe, }; // https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#data-movement-and-conversion-instructions-ld @@ -874,6 +883,13 @@ IntType : ast::IntType = { ".s64" => ast::IntType::S64, }; +IntType3264: ast::IntType = { + ".u32" => ast::IntType::U32, + ".u64" => ast::IntType::U64, + ".s32" => ast::IntType::S32, + ".s64" => ast::IntType::S64, +} + UIntType: ast::UIntType = { ".u16" => ast::UIntType::U16, ".u32" => ast::UIntType::U32, @@ -979,14 +995,14 @@ SetpTypeNoF32: ast::ScalarType = { // https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#logic-and-shift-instructions-not InstNot: ast::Instruction<ast::ParsedArgParams<'input>> = { - "not" <t:NotType> <a:Arg2> => ast::Instruction::Not(t, a) + "not" <t:BooleanType> <a:Arg2> => ast::Instruction::Not(t, a) }; -NotType: ast::NotType = { - ".pred" => ast::NotType::Pred, - ".b16" => ast::NotType::B16, - ".b32" => ast::NotType::B32, - ".b64" => ast::NotType::B64, +BooleanType: ast::BooleanType = { + ".pred" => ast::BooleanType::Pred, + ".b16" => ast::BooleanType::B16, + ".b32" => ast::BooleanType::B32, + ".b64" => ast::BooleanType::B64, }; // https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#control-flow-instructions-at @@ -1294,19 +1310,12 @@ SignedIntType: ast::ScalarType = { // https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#logic-and-shift-instructions-or InstOr: ast::Instruction<ast::ParsedArgParams<'input>> = { - "or" <d:OrAndType> <a:Arg3> => ast::Instruction::Or(d, a), + "or" <d:BooleanType> <a:Arg3> => ast::Instruction::Or(d, a), }; -OrAndType: ast::OrAndType = { - ".pred" => ast::OrAndType::Pred, - ".b16" => ast::OrAndType::B16, - ".b32" => ast::OrAndType::B32, - ".b64" => ast::OrAndType::B64, -} - // https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#logic-and-shift-instructions-and InstAnd: ast::Instruction<ast::ParsedArgParams<'input>> = { - "and" <d:OrAndType> <a:Arg3> => ast::Instruction::And(d, a), + "and" <d:BooleanType> <a:Arg3> => ast::Instruction::And(d, a), }; // https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#floating-point-instructions-rcp @@ -1447,7 +1456,7 @@ InstAtom: ast::Instruction<ast::ParsedArgParams<'input>> = { }; ast::Instruction::Atom(details,a) }, - "atom" <sema:AtomSemantics?> <scope:MemScope?> <space:AtomSpace?> <op: AtomUIntOp> <typ:AtomUIntType> <a:Arg3Atom> => { + "atom" <sema:AtomSemantics?> <scope:MemScope?> <space:AtomSpace?> <op: AtomUIntOp> <typ:UIntType3264> <a:Arg3Atom> => { let details = ast::AtomDetails { semantics: sema.unwrap_or(ast::AtomSemantics::Relaxed), scope: scope.unwrap_or(ast::MemScope::Gpu), @@ -1456,7 +1465,7 @@ InstAtom: ast::Instruction<ast::ParsedArgParams<'input>> = { }; ast::Instruction::Atom(details,a) }, - "atom" <sema:AtomSemantics?> <scope:MemScope?> <space:AtomSpace?> <op: AtomSIntOp> <typ:AtomSIntType> <a:Arg3Atom> => { + "atom" <sema:AtomSemantics?> <scope:MemScope?> <space:AtomSpace?> <op: AtomSIntOp> <typ:SIntType3264> <a:Arg3Atom> => { let details = ast::AtomDetails { semantics: sema.unwrap_or(ast::AtomSemantics::Relaxed), scope: scope.unwrap_or(ast::MemScope::Gpu), @@ -1515,12 +1524,12 @@ BitType: ast::BitType = { ".b64" => ast::BitType::B64, } -AtomUIntType: ast::UIntType = { +UIntType3264: ast::UIntType = { ".u32" => ast::UIntType::U32, ".u64" => ast::UIntType::U64, } -AtomSIntType: ast::SIntType = { +SIntType3264: ast::SIntType = { ".s32" => ast::SIntType::S32, ".s64" => ast::SIntType::S64, } @@ -1664,6 +1673,22 @@ InstPopc: ast::Instruction<ast::ParsedArgParams<'input>> = { "popc" <typ:BitType> <arg:Arg2> => ast::Instruction::Popc{ <> } } +// https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#logic-and-shift-instructions-xor +InstXor: ast::Instruction<ast::ParsedArgParams<'input>> = { + "xor" <typ:BooleanType> <arg:Arg3> => ast::Instruction::Xor{ <> } +} + +// https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#integer-arithmetic-instructions-bfe +InstBfe: ast::Instruction<ast::ParsedArgParams<'input>> = { + "bfe" <typ:IntType3264> <arg:Arg4> => ast::Instruction::Bfe{ <> } +} + +// https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#integer-arithmetic-instructions-rem +InstRem: ast::Instruction<ast::ParsedArgParams<'input>> = { + "rem" <typ:IntType> <arg:Arg3> => ast::Instruction::Rem{ <> } +} + + NegTypeFtz: ast::ScalarType = { ".f16" => ast::ScalarType::F16, ".f16x2" => ast::ScalarType::F16x2, diff --git a/ptx/src/test/spirv_run/bfe.ptx b/ptx/src/test/spirv_run/bfe.ptx new file mode 100644 index 0000000..60ee8a6 --- /dev/null +++ b/ptx/src/test/spirv_run/bfe.ptx @@ -0,0 +1,23 @@ +.version 6.5
+.target sm_30
+.address_size 64
+
+.visible .entry bfe(
+ .param .u64 input,
+ .param .u64 output
+)
+{
+ .reg .u64 in_addr;
+ .reg .u64 out_addr;
+ .reg .u32 temp<3>;
+
+ ld.param.u64 in_addr, [input];
+ ld.param.u64 out_addr, [output];
+
+ ld.u32 temp0, [in_addr];
+ ld.u32 temp1, [in_addr+4];
+ ld.u32 temp2, [in_addr+8];
+ bfe.u32 temp0, temp0, temp1, temp2;
+ st.u32 [out_addr], temp0;
+ ret;
+}
diff --git a/ptx/src/test/spirv_run/bfe.spvtxt b/ptx/src/test/spirv_run/bfe.spvtxt new file mode 100644 index 0000000..edcf138 --- /dev/null +++ b/ptx/src/test/spirv_run/bfe.spvtxt @@ -0,0 +1,70 @@ + OpCapability GenericPointer + OpCapability Linkage + OpCapability Addresses + OpCapability Kernel + OpCapability Int8 + OpCapability Int16 + OpCapability Int64 + OpCapability Float16 + OpCapability Float64 + %40 = OpExtInstImport "OpenCL.std" + OpMemoryModel Physical64 OpenCL + OpEntryPoint Kernel %1 "bfe" + OpDecorate %34 LinkageAttributes "__notcuda_ptx_impl__bfe_u32" Import + %void = OpTypeVoid + %uint = OpTypeInt 32 0 + %43 = OpTypeFunction %uint %uint %uint %uint + %ulong = OpTypeInt 64 0 + %45 = OpTypeFunction %void %ulong %ulong +%_ptr_Function_ulong = OpTypePointer Function %ulong +%_ptr_Function_uint = OpTypePointer Function %uint +%_ptr_Generic_uint = OpTypePointer Generic %uint + %ulong_4 = OpConstant %ulong 4 + %ulong_8 = OpConstant %ulong 8 + %34 = OpFunction %uint None %43 + %36 = OpFunctionParameter %uint + %37 = OpFunctionParameter %uint + %38 = OpFunctionParameter %uint + OpFunctionEnd + %1 = OpFunction %void None %45 + %9 = OpFunctionParameter %ulong + %10 = OpFunctionParameter %ulong + %33 = OpLabel + %2 = OpVariable %_ptr_Function_ulong Function + %3 = OpVariable %_ptr_Function_ulong Function + %4 = OpVariable %_ptr_Function_ulong Function + %5 = OpVariable %_ptr_Function_ulong Function + %6 = OpVariable %_ptr_Function_uint Function + %7 = OpVariable %_ptr_Function_uint Function + %8 = OpVariable %_ptr_Function_uint Function + OpStore %2 %9 + OpStore %3 %10 + %11 = OpLoad %ulong %2 + OpStore %4 %11 + %12 = OpLoad %ulong %3 + OpStore %5 %12 + %14 = OpLoad %ulong %4 + %29 = OpConvertUToPtr %_ptr_Generic_uint %14 + %13 = OpLoad %uint %29 + OpStore %6 %13 + %16 = OpLoad %ulong %4 + %26 = OpIAdd %ulong %16 %ulong_4 + %30 = OpConvertUToPtr %_ptr_Generic_uint %26 + %15 = OpLoad %uint %30 + OpStore %7 %15 + %18 = OpLoad %ulong %4 + %28 = OpIAdd %ulong %18 %ulong_8 + %31 = OpConvertUToPtr %_ptr_Generic_uint %28 + %17 = OpLoad %uint %31 + OpStore %8 %17 + %20 = OpLoad %uint %6 + %21 = OpLoad %uint %7 + %22 = OpLoad %uint %8 + %19 = OpFunctionCall %uint %34 %20 %21 %22 + OpStore %6 %19 + %23 = OpLoad %ulong %5 + %24 = OpLoad %uint %6 + %32 = OpConvertUToPtr %_ptr_Generic_uint %23 + OpStore %32 %24 + OpReturn + OpFunctionEnd diff --git a/ptx/src/test/spirv_run/mod.rs b/ptx/src/test/spirv_run/mod.rs index a7ef75b..5bbe45a 100644 --- a/ptx/src/test/spirv_run/mod.rs +++ b/ptx/src/test/spirv_run/mod.rs @@ -116,6 +116,20 @@ test_ptx!( [0b11000111_01011100_10101110_11111011u32],
[0b11011111_01110101_00111010_11100011u32]
);
+test_ptx!(
+ xor,
+ [
+ 0b01010010_00011010_01000000_00001101u32,
+ 0b11100110_10011011_00001100_00100011u32
+ ],
+ [0b10110100100000010100110000101110u32]
+);
+test_ptx!(rem, [21692i32, 13i32], [8i32]);
+test_ptx!(
+ bfe,
+ [0b11111000_11000001_00100010_10100000u32, 16u32, 8u32],
+ [0b11000001u32]
+);
struct DisplayError<T: Debug> {
err: T,
diff --git a/ptx/src/test/spirv_run/rem.ptx b/ptx/src/test/spirv_run/rem.ptx new file mode 100644 index 0000000..2ac482d --- /dev/null +++ b/ptx/src/test/spirv_run/rem.ptx @@ -0,0 +1,23 @@ +.version 6.5
+.target sm_30
+.address_size 64
+
+.visible .entry rem(
+ .param .u64 input,
+ .param .u64 output
+)
+{
+ .reg .u64 in_addr;
+ .reg .u64 out_addr;
+ .reg .s32 temp1;
+ .reg .s32 temp2;
+
+ ld.param.u64 in_addr, [input];
+ ld.param.u64 out_addr, [output];
+
+ ld.s32 temp1, [in_addr];
+ ld.s32 temp2, [in_addr+4];
+ rem.s32 temp1, temp1, temp2;
+ st.s32 [out_addr], temp1;
+ ret;
+}
diff --git a/ptx/src/test/spirv_run/rem.spvtxt b/ptx/src/test/spirv_run/rem.spvtxt new file mode 100644 index 0000000..72d0965 --- /dev/null +++ b/ptx/src/test/spirv_run/rem.spvtxt @@ -0,0 +1,55 @@ + OpCapability GenericPointer + OpCapability Linkage + OpCapability Addresses + OpCapability Kernel + OpCapability Int8 + OpCapability Int16 + OpCapability Int64 + OpCapability Float16 + OpCapability Float64 + %28 = OpExtInstImport "OpenCL.std" + OpMemoryModel Physical64 OpenCL + OpEntryPoint Kernel %1 "rem" + %void = OpTypeVoid + %ulong = OpTypeInt 64 0 + %31 = OpTypeFunction %void %ulong %ulong +%_ptr_Function_ulong = OpTypePointer Function %ulong + %uint = OpTypeInt 32 0 +%_ptr_Function_uint = OpTypePointer Function %uint +%_ptr_Generic_uint = OpTypePointer Generic %uint + %ulong_4 = OpConstant %ulong 4 + %1 = OpFunction %void None %31 + %8 = OpFunctionParameter %ulong + %9 = OpFunctionParameter %ulong + %26 = OpLabel + %2 = OpVariable %_ptr_Function_ulong Function + %3 = OpVariable %_ptr_Function_ulong Function + %4 = OpVariable %_ptr_Function_ulong Function + %5 = OpVariable %_ptr_Function_ulong Function + %6 = OpVariable %_ptr_Function_uint Function + %7 = OpVariable %_ptr_Function_uint Function + OpStore %2 %8 + OpStore %3 %9 + %10 = OpLoad %ulong %2 + OpStore %4 %10 + %11 = OpLoad %ulong %3 + OpStore %5 %11 + %13 = OpLoad %ulong %4 + %23 = OpConvertUToPtr %_ptr_Generic_uint %13 + %12 = OpLoad %uint %23 + OpStore %6 %12 + %15 = OpLoad %ulong %4 + %22 = OpIAdd %ulong %15 %ulong_4 + %24 = OpConvertUToPtr %_ptr_Generic_uint %22 + %14 = OpLoad %uint %24 + OpStore %7 %14 + %17 = OpLoad %uint %6 + %18 = OpLoad %uint %7 + %16 = OpSMod %uint %17 %18 + OpStore %6 %16 + %19 = OpLoad %ulong %5 + %20 = OpLoad %uint %6 + %25 = OpConvertUToPtr %_ptr_Generic_uint %19 + OpStore %25 %20 + OpReturn + OpFunctionEnd diff --git a/ptx/src/test/spirv_run/xor.ptx b/ptx/src/test/spirv_run/xor.ptx new file mode 100644 index 0000000..a28b321 --- /dev/null +++ b/ptx/src/test/spirv_run/xor.ptx @@ -0,0 +1,23 @@ +.version 6.5
+.target sm_30
+.address_size 64
+
+.visible .entry xor(
+ .param .u64 input,
+ .param .u64 output
+)
+{
+ .reg .u64 in_addr;
+ .reg .u64 out_addr;
+ .reg .b32 temp1;
+ .reg .b32 temp2;
+
+ ld.param.u64 in_addr, [input];
+ ld.param.u64 out_addr, [output];
+
+ ld.b32 temp1, [in_addr];
+ ld.b32 temp2, [in_addr+4];
+ xor.b32 temp1, temp1, temp2;
+ st.b32 [out_addr], temp1;
+ ret;
+}
diff --git a/ptx/src/test/spirv_run/xor.spvtxt b/ptx/src/test/spirv_run/xor.spvtxt new file mode 100644 index 0000000..ee09898 --- /dev/null +++ b/ptx/src/test/spirv_run/xor.spvtxt @@ -0,0 +1,55 @@ + OpCapability GenericPointer + OpCapability Linkage + OpCapability Addresses + OpCapability Kernel + OpCapability Int8 + OpCapability Int16 + OpCapability Int64 + OpCapability Float16 + OpCapability Float64 + %28 = OpExtInstImport "OpenCL.std" + OpMemoryModel Physical64 OpenCL + OpEntryPoint Kernel %1 "xor" + %void = OpTypeVoid + %ulong = OpTypeInt 64 0 + %31 = OpTypeFunction %void %ulong %ulong +%_ptr_Function_ulong = OpTypePointer Function %ulong + %uint = OpTypeInt 32 0 +%_ptr_Function_uint = OpTypePointer Function %uint +%_ptr_Generic_uint = OpTypePointer Generic %uint + %ulong_4 = OpConstant %ulong 4 + %1 = OpFunction %void None %31 + %8 = OpFunctionParameter %ulong + %9 = OpFunctionParameter %ulong + %26 = OpLabel + %2 = OpVariable %_ptr_Function_ulong Function + %3 = OpVariable %_ptr_Function_ulong Function + %4 = OpVariable %_ptr_Function_ulong Function + %5 = OpVariable %_ptr_Function_ulong Function + %6 = OpVariable %_ptr_Function_uint Function + %7 = OpVariable %_ptr_Function_uint Function + OpStore %2 %8 + OpStore %3 %9 + %10 = OpLoad %ulong %2 + OpStore %4 %10 + %11 = OpLoad %ulong %3 + OpStore %5 %11 + %13 = OpLoad %ulong %4 + %23 = OpConvertUToPtr %_ptr_Generic_uint %13 + %12 = OpLoad %uint %23 + OpStore %6 %12 + %15 = OpLoad %ulong %4 + %22 = OpIAdd %ulong %15 %ulong_4 + %24 = OpConvertUToPtr %_ptr_Generic_uint %22 + %14 = OpLoad %uint %24 + OpStore %7 %14 + %17 = OpLoad %uint %6 + %18 = OpLoad %uint %7 + %16 = OpBitwiseXor %uint %17 %18 + OpStore %6 %16 + %19 = OpLoad %ulong %5 + %20 = OpLoad %uint %6 + %25 = OpConvertUToPtr %_ptr_Generic_uint %19 + OpStore %25 %20 + OpReturn + OpFunctionEnd diff --git a/ptx/src/translate.rs b/ptx/src/translate.rs index 23a63be..365d1e8 100644 --- a/ptx/src/translate.rs +++ b/ptx/src/translate.rs @@ -1289,6 +1289,9 @@ fn extract_globals<'input, 'b>( ..
},
) => global.push(var),
+ Statement::Instruction(ast::Instruction::Bfe { typ, arg }) => {
+ local.push(to_ptx_impl_bfe_call(id_def, ptx_impl_imports, typ, arg));
+ }
Statement::Instruction(ast::Instruction::Atom(
d
@
@@ -1591,6 +1594,24 @@ fn convert_to_typed_statements( arg: arg.cast(),
}))
}
+ ast::Instruction::Xor { typ, arg } => {
+ result.push(Statement::Instruction(ast::Instruction::Xor {
+ typ,
+ arg: arg.cast(),
+ }))
+ }
+ ast::Instruction::Bfe { typ, arg } => {
+ result.push(Statement::Instruction(ast::Instruction::Bfe {
+ typ,
+ arg: arg.cast(),
+ }))
+ }
+ ast::Instruction::Rem { typ, arg } => {
+ result.push(Statement::Instruction(ast::Instruction::Rem {
+ typ,
+ arg: arg.cast(),
+ }))
+ }
},
Statement::Label(i) => result.push(Statement::Label(i)),
Statement::Variable(v) => result.push(Statement::Variable(v)),
@@ -1610,6 +1631,7 @@ fn convert_to_typed_statements( Ok(result)
}
+//TODO: share common code between this and to_ptx_impl_bfe_call
fn to_ptx_impl_atomic_call(
id_defs: &mut NumericIdResolver,
ptx_impl_imports: &mut HashMap<String, Directive>,
@@ -1705,6 +1727,100 @@ fn to_ptx_impl_atomic_call( })
}
+fn to_ptx_impl_bfe_call(
+ id_defs: &mut NumericIdResolver,
+ ptx_impl_imports: &mut HashMap<String, Directive>,
+ typ: ast::IntType,
+ arg: ast::Arg4<ExpandedArgParams>,
+) -> ExpandedStatement {
+ let prefix = "__notcuda_ptx_impl__";
+ let suffix = match typ {
+ ast::IntType::U32 => "bfe_u32",
+ ast::IntType::U64 => "bfe_u64",
+ ast::IntType::S32 => "bfe_s32",
+ ast::IntType::S64 => "bfe_s64",
+ _ => unreachable!(),
+ };
+ let fn_name = format!("{}{}", prefix, suffix);
+ let fn_id = match ptx_impl_imports.entry(fn_name) {
+ hash_map::Entry::Vacant(entry) => {
+ let fn_id = id_defs.new_id(None);
+ let func_decl = ast::MethodDecl::Func::<spirv::Word>(
+ vec![ast::FnArgument {
+ align: None,
+ v_type: ast::FnArgumentType::Reg(ast::VariableRegType::Scalar(typ.into())),
+ name: id_defs.new_id(None),
+ array_init: Vec::new(),
+ }],
+ fn_id,
+ vec![
+ ast::FnArgument {
+ align: None,
+ v_type: ast::FnArgumentType::Reg(ast::VariableRegType::Scalar(typ.into())),
+ name: id_defs.new_id(None),
+ array_init: Vec::new(),
+ },
+ ast::FnArgument {
+ align: None,
+ v_type: ast::FnArgumentType::Reg(ast::VariableRegType::Scalar(
+ ast::ScalarType::U32,
+ )),
+ name: id_defs.new_id(None),
+ array_init: Vec::new(),
+ },
+ ast::FnArgument {
+ align: None,
+ v_type: ast::FnArgumentType::Reg(ast::VariableRegType::Scalar(
+ ast::ScalarType::U32,
+ )),
+ name: id_defs.new_id(None),
+ array_init: Vec::new(),
+ },
+ ],
+ );
+ let spirv_decl = SpirvMethodDecl::new(&func_decl);
+ let func = Function {
+ func_decl,
+ globals: Vec::new(),
+ body: None,
+ import_as: Some(entry.key().clone()),
+ spirv_decl,
+ };
+ entry.insert(Directive::Method(func));
+ fn_id
+ }
+ hash_map::Entry::Occupied(entry) => match entry.get() {
+ Directive::Method(Function {
+ func_decl: ast::MethodDecl::Func(_, name, _),
+ ..
+ }) => *name,
+ _ => unreachable!(),
+ },
+ };
+ Statement::Call(ResolvedCall {
+ uniform: false,
+ func: fn_id,
+ ret_params: vec![(
+ arg.dst,
+ ast::FnArgumentType::Reg(ast::VariableRegType::Scalar(typ.into())),
+ )],
+ param_list: vec![
+ (
+ arg.src1,
+ ast::FnArgumentType::Reg(ast::VariableRegType::Scalar(typ.into())),
+ ),
+ (
+ arg.src2,
+ ast::FnArgumentType::Reg(ast::VariableRegType::Scalar(ast::ScalarType::U32)),
+ ),
+ (
+ arg.src3,
+ ast::FnArgumentType::Reg(ast::VariableRegType::Scalar(ast::ScalarType::U32)),
+ ),
+ ],
+ })
+}
+
fn to_resolved_fn_args<T>(
params: Vec<T>,
params_decl: &[ast::FnArgumentType],
@@ -2803,7 +2919,7 @@ fn emit_function_body_ops( let result_id = Some(a.dst);
let operand = a.src;
match t {
- ast::NotType::Pred => {
+ ast::BooleanType::Pred => {
// HACK ALERT
// Temporary workaround until IGC gets its shit together
// Currently IGC carries two copies of SPIRV-LLVM translator
@@ -2854,7 +2970,7 @@ fn emit_function_body_ops( },
ast::Instruction::Or(t, a) => {
let result_type = map.get_or_add_scalar(builder, ast::ScalarType::from(*t));
- if *t == ast::OrAndType::Pred {
+ if *t == ast::BooleanType::Pred {
builder.logical_or(result_type, Some(a.dst), a.src1, a.src2)?;
} else {
builder.bitwise_or(result_type, Some(a.dst), a.src1, a.src2)?;
@@ -2882,7 +2998,7 @@ fn emit_function_body_ops( }
ast::Instruction::And(t, a) => {
let result_type = map.get_or_add_scalar(builder, ast::ScalarType::from(*t));
- if *t == ast::OrAndType::Pred {
+ if *t == ast::BooleanType::Pred {
builder.logical_and(result_type, Some(a.dst), a.src1, a.src2)?;
} else {
builder.bitwise_and(result_type, Some(a.dst), a.src1, a.src2)?;
@@ -3033,6 +3149,39 @@ fn emit_function_body_ops( let result_type = map.get_or_add_scalar(builder, (*typ).into());
builder.bit_count(result_type, Some(arg.dst), arg.src)?;
}
+ ast::Instruction::Xor { typ, arg } => {
+ let builder_fn = match typ {
+ ast::BooleanType::Pred => emit_logical_xor_spirv,
+ _ => dr::Builder::bitwise_xor,
+ };
+ let result_type = map.get_or_add_scalar(builder, (*typ).into());
+ builder_fn(builder, result_type, Some(arg.dst), arg.src1, arg.src2)?;
+ }
+ ast::Instruction::Bfe { typ, arg } => {
+ let builder_fn = if typ.is_signed() {
+ dr::Builder::bit_field_s_extract
+ } else {
+ dr::Builder::bit_field_u_extract
+ };
+ let result_type = map.get_or_add_scalar(builder, (*typ).into());
+ builder_fn(
+ builder,
+ result_type,
+ Some(arg.dst),
+ arg.src1,
+ arg.src2,
+ arg.src3,
+ )?;
+ }
+ ast::Instruction::Rem { typ, arg } => {
+ let builder_fn = if typ.is_signed() {
+ dr::Builder::s_mod
+ } else {
+ dr::Builder::u_mod
+ };
+ let result_type = map.get_or_add_scalar(builder, (*typ).into());
+ builder_fn(builder, result_type, Some(arg.dst), arg.src1, arg.src2)?;
+ }
},
Statement::LoadVar(arg, typ) => {
let type_id = map.get_or_add(builder, SpirvType::from(typ.clone()));
@@ -3079,6 +3228,20 @@ fn emit_function_body_ops( Ok(())
}
+// TODO: check what kind of assembly do we emit
+fn emit_logical_xor_spirv(
+ builder: &mut dr::Builder,
+ result_type: spirv::Word,
+ result_id: Option<spirv::Word>,
+ op1: spirv::Word,
+ op2: spirv::Word,
+) -> Result<spirv::Word, dr::Error> {
+ let temp_or = builder.logical_or(result_type, None, op1, op2)?;
+ let temp_and = builder.logical_and(result_type, None, op1, op2)?;
+ let temp_neg = builder.logical_not(result_type, None, temp_and)?;
+ builder.logical_and(result_type, result_id, temp_or, temp_neg)
+}
+
fn emit_sqrt(
builder: &mut dr::Builder,
map: &mut TypeWordMap,
@@ -5039,6 +5202,27 @@ impl<T: ArgParamsEx> ast::Instruction<T> { arg: arg.map_different_types(visitor, &dst_type, &src_type)?,
}
}
+ ast::Instruction::Xor { typ, arg } => {
+ let full_type = ast::Type::Scalar(typ.into());
+ ast::Instruction::Xor {
+ typ,
+ arg: arg.map_non_shift(visitor, &full_type, false)?,
+ }
+ }
+ ast::Instruction::Bfe { typ, arg } => {
+ let full_type = ast::Type::Scalar(typ.into());
+ ast::Instruction::Bfe {
+ typ,
+ arg: arg.map_bfe(visitor, &full_type)?,
+ }
+ }
+ ast::Instruction::Rem { typ, arg } => {
+ let full_type = ast::Type::Scalar(typ.into());
+ ast::Instruction::Rem {
+ typ,
+ arg: arg.map_non_shift(visitor, &full_type, false)?,
+ }
+ }
})
}
}
@@ -5351,6 +5535,9 @@ impl ast::Instruction<ExpandedArgParams> { ast::Instruction::Clz { .. } => None,
ast::Instruction::Brev { .. } => None,
ast::Instruction::Popc { .. } => None,
+ ast::Instruction::Xor { .. } => None,
+ ast::Instruction::Bfe { .. } => None,
+ ast::Instruction::Rem { .. } => None,
ast::Instruction::Sub(ast::ArithDetails::Float(float_control), _)
| ast::Instruction::Add(ast::ArithDetails::Float(float_control), _)
| ast::Instruction::Mul(ast::MulDetails::Float(float_control), _)
@@ -6192,6 +6379,52 @@ impl<T: ArgParamsEx> ast::Arg4<T> { src3,
})
}
+
+ fn map_bfe<U: ArgParamsEx, V: ArgumentMapVisitor<T, U>>(
+ self,
+ visitor: &mut V,
+ typ: &ast::Type,
+ ) -> Result<ast::Arg4<U>, TranslateError> {
+ let dst = visitor.id(
+ ArgumentDescriptor {
+ op: self.dst,
+ is_dst: true,
+ sema: ArgumentSemantics::Default,
+ },
+ Some(typ),
+ )?;
+ let src1 = visitor.operand(
+ ArgumentDescriptor {
+ op: self.src1,
+ is_dst: false,
+ sema: ArgumentSemantics::Default,
+ },
+ typ,
+ )?;
+ let u32_type = ast::Type::Scalar(ast::ScalarType::U32);
+ let src2 = visitor.operand(
+ ArgumentDescriptor {
+ op: self.src2,
+ is_dst: false,
+ sema: ArgumentSemantics::Default,
+ },
+ &u32_type,
+ )?;
+ let src3 = visitor.operand(
+ ArgumentDescriptor {
+ op: self.src3,
+ is_dst: false,
+ sema: ArgumentSemantics::Default,
+ },
+ &u32_type,
+ )?;
+ Ok(ast::Arg4 {
+ dst,
+ src1,
+ src2,
+ src3,
+ })
+ }
}
impl<T: ArgParamsEx> ast::Arg4Setp<T> {
@@ -6437,13 +6670,13 @@ impl ast::ScalarType { }
}
-impl ast::NotType {
+impl ast::BooleanType {
fn to_type(self) -> ast::Type {
match self {
- ast::NotType::Pred => ast::Type::Scalar(ast::ScalarType::Pred),
- ast::NotType::B16 => ast::Type::Scalar(ast::ScalarType::B16),
- ast::NotType::B32 => ast::Type::Scalar(ast::ScalarType::B32),
- ast::NotType::B64 => ast::Type::Scalar(ast::ScalarType::B64),
+ ast::BooleanType::Pred => ast::Type::Scalar(ast::ScalarType::Pred),
+ ast::BooleanType::B16 => ast::Type::Scalar(ast::ScalarType::B16),
+ ast::BooleanType::B32 => ast::Type::Scalar(ast::ScalarType::B32),
+ ast::BooleanType::B64 => ast::Type::Scalar(ast::ScalarType::B64),
}
}
}
|