Skip to content

Commit e2c5d99

Browse files
a74nha74nh
committed
JIT ARM64-SVE: Add CreateWhileLessThan*
1 parent d1747a7 commit e2c5d99

File tree

9 files changed

+767
-12
lines changed

9 files changed

+767
-12
lines changed

‎src/coreclr/jit/hwintrinsicarm64.cpp

Lines changed: 38 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -2194,6 +2194,44 @@ GenTree* Compiler::impSpecialIntrinsic(NamedIntrinsic intrinsic,
21942194
break;
21952195
}
21962196

2197+
case NI_Sve_CreateWhileLessThanMask8Bit:
2198+
case NI_Sve_CreateWhileLessThanMask16Bit:
2199+
case NI_Sve_CreateWhileLessThanMask32Bit:
2200+
case NI_Sve_CreateWhileLessThanMask64Bit:
2201+
case NI_Sve_CreateWhileLessThanOrEqualMask8Bit:
2202+
case NI_Sve_CreateWhileLessThanOrEqualMask16Bit:
2203+
case NI_Sve_CreateWhileLessThanOrEqualMask32Bit:
2204+
case NI_Sve_CreateWhileLessThanOrEqualMask64Bit:
2205+
{
2206+
// Target instruction is dependent on whether the inputs are signed or unsigned.
2207+
// This information is lost when the type is converted from CorInfoType to var_type.
2208+
// Ensure this is marked using GTF_UNSIGNED.
2209+
2210+
CORINFO_ARG_LIST_HANDLE arg1 = sig->args;
2211+
CORINFO_ARG_LIST_HANDLE arg2 = info.compCompHnd->getArgNext(arg1);
2212+
var_types argType = TYP_UNKNOWN;
2213+
CORINFO_CLASS_HANDLE argClass = NO_CLASS_HANDLE;
2214+
CorInfoType argCoreInfoType = strip(info.compCompHnd->getArgType(sig, arg2, &argClass));
2215+
2216+
assert(sig->numArgs == 2);
2217+
argType = JITtype2varType(argCoreInfoType);
2218+
op2 = getArgForHWIntrinsic(argType, argClass);
2219+
argType = JITtype2varType(strip(info.compCompHnd->getArgType(sig, arg1, &argClass)));
2220+
op1 = impPopStack().val;
2221+
2222+
retNode = gtNewSimdHWIntrinsicNode(retType, op1, op2, intrinsic, simdBaseJitType, simdSize);
2223+
2224+
if (argCoreInfoType == CORINFO_TYPE_ULONG || argCoreInfoType == CORINFO_TYPE_UINT)
2225+
{
2226+
retNode->gtFlags |= GTF_UNSIGNED;
2227+
}
2228+
else
2229+
{
2230+
assert(argCoreInfoType == CORINFO_TYPE_LONG || argCoreInfoType == CORINFO_TYPE_INT);
2231+
}
2232+
}
2233+
break;
2234+
21972235
default:
21982236
{
21992237
return nullptr;

‎src/coreclr/jit/hwintrinsiccodegenarm64.cpp

Lines changed: 22 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1300,6 +1300,28 @@ void CodeGen::genHWIntrinsic(GenTreeHWIntrinsic* node)
13001300
GetEmitter()->emitIns_R_PATTERN(ins, emitSize, targetReg, opt, SVE_PATTERN_ALL);
13011301
break;
13021302

1303+
case NI_Sve_CreateWhileLessThanMask8Bit:
1304+
case NI_Sve_CreateWhileLessThanMask16Bit:
1305+
case NI_Sve_CreateWhileLessThanMask32Bit:
1306+
case NI_Sve_CreateWhileLessThanMask64Bit:
1307+
// Emit size is the size of the scalar operands.
1308+
emitSize = emitActualTypeSize(intrin.op1->TypeGet());
1309+
// Instruction is dependent on whether the inputs are signed or unsigned.
1310+
ins = ((node->gtFlags & GTF_UNSIGNED) != 0) ? INS_sve_whilelo : INS_sve_whilelt;
1311+
GetEmitter()->emitIns_R_R_R(ins, emitSize, targetReg, op1Reg, op2Reg, opt);
1312+
break;
1313+
1314+
case NI_Sve_CreateWhileLessThanOrEqualMask8Bit:
1315+
case NI_Sve_CreateWhileLessThanOrEqualMask16Bit:
1316+
case NI_Sve_CreateWhileLessThanOrEqualMask32Bit:
1317+
case NI_Sve_CreateWhileLessThanOrEqualMask64Bit:
1318+
// Emit size is the size of the scalar operands.
1319+
emitSize = emitActualTypeSize(intrin.op1->TypeGet());
1320+
// Instruction is dependent on whether the inputs are signed or unsigned.
1321+
ins = ((node->gtFlags & GTF_UNSIGNED) != 0) ? INS_sve_whilels : INS_sve_whilele;
1322+
GetEmitter()->emitIns_R_R_R(ins, emitSize, targetReg, op1Reg, op2Reg, opt);
1323+
break;
1324+
13031325
default:
13041326
unreached();
13051327
}

‎src/coreclr/jit/hwintrinsiclistarm64sve.h

Lines changed: 8 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -27,6 +27,14 @@ HARDWARE_INTRINSIC(Sve, CreateTrueMaskSingle,
2727
HARDWARE_INTRINSIC(Sve, CreateTrueMaskUInt16, -1, 1, false, {INS_invalid, INS_invalid, INS_invalid, INS_sve_ptrue, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_EnumPattern, HW_Flag_Scalable|HW_Flag_HasImmediateOperand|HW_Flag_ReturnsPerElementMask)
2828
HARDWARE_INTRINSIC(Sve, CreateTrueMaskUInt32, -1, 1, false, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_sve_ptrue, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_EnumPattern, HW_Flag_Scalable|HW_Flag_HasImmediateOperand|HW_Flag_ReturnsPerElementMask)
2929
HARDWARE_INTRINSIC(Sve, CreateTrueMaskUInt64, -1, 1, false, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_sve_ptrue, INS_invalid, INS_invalid}, HW_Category_EnumPattern, HW_Flag_Scalable|HW_Flag_HasImmediateOperand|HW_Flag_ReturnsPerElementMask)
30+
HARDWARE_INTRINSIC(Sve, CreateWhileLessThanMask16Bit, -1, 2, false, {INS_invalid, INS_invalid, INS_invalid, INS_sve_whilelt, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_SIMD, HW_Flag_Scalable|HW_Flag_SpecialImport|HW_Flag_SpecialCodeGen|HW_Flag_ReturnsPerElementMask)
31+
HARDWARE_INTRINSIC(Sve, CreateWhileLessThanMask32Bit, -1, 2, false, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_sve_whilelt, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_SIMD, HW_Flag_Scalable|HW_Flag_SpecialImport|HW_Flag_SpecialCodeGen|HW_Flag_ReturnsPerElementMask)
32+
HARDWARE_INTRINSIC(Sve, CreateWhileLessThanMask64Bit, -1, 2, false, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_sve_whilelt, INS_invalid, INS_invalid}, HW_Category_SIMD, HW_Flag_Scalable|HW_Flag_SpecialImport|HW_Flag_SpecialCodeGen|HW_Flag_ReturnsPerElementMask)
33+
HARDWARE_INTRINSIC(Sve, CreateWhileLessThanMask8Bit, -1, 2, false, {INS_invalid, INS_sve_whilelt, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_SIMD, HW_Flag_Scalable|HW_Flag_SpecialImport|HW_Flag_SpecialCodeGen|HW_Flag_ReturnsPerElementMask)
34+
HARDWARE_INTRINSIC(Sve, CreateWhileLessThanOrEqualMask16Bit, -1, 2, false, {INS_invalid, INS_invalid, INS_invalid, INS_sve_whilele, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_SIMD, HW_Flag_Scalable|HW_Flag_SpecialImport|HW_Flag_SpecialCodeGen|HW_Flag_ReturnsPerElementMask)
35+
HARDWARE_INTRINSIC(Sve, CreateWhileLessThanOrEqualMask32Bit, -1, 2, false, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_sve_whilele, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_SIMD, HW_Flag_Scalable|HW_Flag_SpecialImport|HW_Flag_SpecialCodeGen|HW_Flag_ReturnsPerElementMask)
36+
HARDWARE_INTRINSIC(Sve, CreateWhileLessThanOrEqualMask64Bit, -1, 2, false, {INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_sve_whilele, INS_invalid, INS_invalid}, HW_Category_SIMD, HW_Flag_Scalable|HW_Flag_SpecialImport|HW_Flag_SpecialCodeGen|HW_Flag_ReturnsPerElementMask)
37+
HARDWARE_INTRINSIC(Sve, CreateWhileLessThanOrEqualMask8Bit, -1, 2, false, {INS_invalid, INS_sve_whilele, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid, INS_invalid}, HW_Category_SIMD, HW_Flag_Scalable|HW_Flag_SpecialImport|HW_Flag_SpecialCodeGen|HW_Flag_ReturnsPerElementMask)
3038

3139
HARDWARE_INTRINSIC(Sve, LoadVector, -1, 2, true, {INS_sve_ld1b, INS_sve_ld1b, INS_sve_ld1h, INS_sve_ld1h, INS_sve_ld1w, INS_sve_ld1w, INS_sve_ld1d, INS_sve_ld1d, INS_sve_ld1w, INS_sve_ld1d}, HW_Category_MemoryLoad, HW_Flag_Scalable|HW_Flag_LowMaskedOperation)
3240

‎src/libraries/System.Private.CoreLib/src/System/Runtime/Intrinsics/Arm/Sve.PlatformNotSupported.cs

Lines changed: 183 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -121,6 +121,189 @@ internal Arm64() { }
121121
public static unsafe Vector<ulong> CreateTrueMaskUInt64([ConstantExpected] SveMaskPattern pattern = SveMaskPattern.All) { throw new PlatformNotSupportedException(); }
122122

123123

124+
/// CreateWhileLessThanMask16Bit : While incrementing scalar is less than
125+
126+
/// <summary>
127+
/// svbool_t svwhilelt_b16[_s32](int32_t op1, int32_t op2)
128+
/// </summary>
129+
public static unsafe Vector<ushort> CreateWhileLessThanMask16Bit(int left, int right) { throw new PlatformNotSupportedException(); }
130+
131+
/// <summary>
132+
/// svbool_t svwhilelt_b16[_s64](int64_t op1, int64_t op2)
133+
/// </summary>
134+
public static unsafe Vector<ushort> CreateWhileLessThanMask16Bit(long left, long right) { throw new PlatformNotSupportedException(); }
135+
136+
/// <summary>
137+
/// svbool_t svwhilelt_b16[_u32](uint32_t op1, uint32_t op2)
138+
/// </summary>
139+
public static unsafe Vector<ushort> CreateWhileLessThanMask16Bit(uint left, uint right) { throw new PlatformNotSupportedException(); }
140+
141+
/// <summary>
142+
/// svbool_t svwhilelt_b16[_u64](uint64_t op1, uint64_t op2)
143+
/// </summary>
144+
public static unsafe Vector<ushort> CreateWhileLessThanMask16Bit(ulong left, ulong right) { throw new PlatformNotSupportedException(); }
145+
146+
147+
/// CreateWhileLessThanMask32Bit : While incrementing scalar is less than
148+
149+
/// <summary>
150+
/// svbool_t svwhilelt_b32[_s32](int32_t op1, int32_t op2)
151+
/// </summary>
152+
public static unsafe Vector<uint> CreateWhileLessThanMask32Bit(int left, int right) { throw new PlatformNotSupportedException(); }
153+
154+
/// <summary>
155+
/// svbool_t svwhilelt_b32[_s64](int64_t op1, int64_t op2)
156+
/// </summary>
157+
public static unsafe Vector<uint> CreateWhileLessThanMask32Bit(long left, long right) { throw new PlatformNotSupportedException(); }
158+
159+
/// <summary>
160+
/// svbool_t svwhilelt_b32[_u32](uint32_t op1, uint32_t op2)
161+
/// </summary>
162+
public static unsafe Vector<uint> CreateWhileLessThanMask32Bit(uint left, uint right) { throw new PlatformNotSupportedException(); }
163+
164+
/// <summary>
165+
/// svbool_t svwhilelt_b32[_u64](uint64_t op1, uint64_t op2)
166+
/// </summary>
167+
public static unsafe Vector<uint> CreateWhileLessThanMask32Bit(ulong left, ulong right) { throw new PlatformNotSupportedException(); }
168+
169+
170+
/// CreateWhileLessThanMask64Bit : While incrementing scalar is less than
171+
172+
/// <summary>
173+
/// svbool_t svwhilelt_b64[_s32](int32_t op1, int32_t op2)
174+
/// </summary>
175+
public static unsafe Vector<ulong> CreateWhileLessThanMask64Bit(int left, int right) { throw new PlatformNotSupportedException(); }
176+
177+
/// <summary>
178+
/// svbool_t svwhilelt_b64[_s64](int64_t op1, int64_t op2)
179+
/// </summary>
180+
public static unsafe Vector<ulong> CreateWhileLessThanMask64Bit(long left, long right) { throw new PlatformNotSupportedException(); }
181+
182+
/// <summary>
183+
/// svbool_t svwhilelt_b64[_u32](uint32_t op1, uint32_t op2)
184+
/// </summary>
185+
public static unsafe Vector<ulong> CreateWhileLessThanMask64Bit(uint left, uint right) { throw new PlatformNotSupportedException(); }
186+
187+
/// <summary>
188+
/// svbool_t svwhilelt_b64[_u64](uint64_t op1, uint64_t op2)
189+
/// </summary>
190+
public static unsafe Vector<ulong> CreateWhileLessThanMask64Bit(ulong left, ulong right) { throw new PlatformNotSupportedException(); }
191+
192+
193+
/// CreateWhileLessThanMask8Bit : While incrementing scalar is less than
194+
195+
/// <summary>
196+
/// svbool_t svwhilelt_b8[_s32](int32_t op1, int32_t op2)
197+
/// </summary>
198+
public static unsafe Vector<byte> CreateWhileLessThanMask8Bit(int left, int right) { throw new PlatformNotSupportedException(); }
199+
200+
/// <summary>
201+
/// svbool_t svwhilelt_b8[_s64](int64_t op1, int64_t op2)
202+
/// </summary>
203+
public static unsafe Vector<byte> CreateWhileLessThanMask8Bit(long left, long right) { throw new PlatformNotSupportedException(); }
204+
205+
/// <summary>
206+
/// svbool_t svwhilelt_b8[_u32](uint32_t op1, uint32_t op2)
207+
/// </summary>
208+
public static unsafe Vector<byte> CreateWhileLessThanMask8Bit(uint left, uint right) { throw new PlatformNotSupportedException(); }
209+
210+
/// <summary>
211+
/// svbool_t svwhilelt_b8[_u64](uint64_t op1, uint64_t op2)
212+
/// </summary>
213+
public static unsafe Vector<byte> CreateWhileLessThanMask8Bit(ulong left, ulong right) { throw new PlatformNotSupportedException(); }
214+
215+
216+
/// CreateWhileLessThanOrEqualMask16Bit : While incrementing scalar is less than or equal to
217+
218+
/// <summary>
219+
/// svbool_t svwhilele_b16[_s32](int32_t op1, int32_t op2)
220+
/// </summary>
221+
public static unsafe Vector<ushort> CreateWhileLessThanOrEqualMask16Bit(int left, int right) { throw new PlatformNotSupportedException(); }
222+
223+
/// <summary>
224+
/// svbool_t svwhilele_b16[_s64](int64_t op1, int64_t op2)
225+
/// </summary>
226+
public static unsafe Vector<ushort> CreateWhileLessThanOrEqualMask16Bit(long left, long right) { throw new PlatformNotSupportedException(); }
227+
228+
/// <summary>
229+
/// svbool_t svwhilele_b16[_u32](uint32_t op1, uint32_t op2)
230+
/// </summary>
231+
public static unsafe Vector<ushort> CreateWhileLessThanOrEqualMask16Bit(uint left, uint right) { throw new PlatformNotSupportedException(); }
232+
233+
/// <summary>
234+
/// svbool_t svwhilele_b16[_u64](uint64_t op1, uint64_t op2)
235+
/// </summary>
236+
public static unsafe Vector<ushort> CreateWhileLessThanOrEqualMask16Bit(ulong left, ulong right) { throw new PlatformNotSupportedException(); }
237+
238+
239+
/// CreateWhileLessThanOrEqualMask32Bit : While incrementing scalar is less than or equal to
240+
241+
/// <summary>
242+
/// svbool_t svwhilele_b32[_s32](int32_t op1, int32_t op2)
243+
/// </summary>
244+
public static unsafe Vector<uint> CreateWhileLessThanOrEqualMask32Bit(int left, int right) { throw new PlatformNotSupportedException(); }
245+
246+
/// <summary>
247+
/// svbool_t svwhilele_b32[_s64](int64_t op1, int64_t op2)
248+
/// </summary>
249+
public static unsafe Vector<uint> CreateWhileLessThanOrEqualMask32Bit(long left, long right) { throw new PlatformNotSupportedException(); }
250+
251+
/// <summary>
252+
/// svbool_t svwhilele_b32[_u32](uint32_t op1, uint32_t op2)
253+
/// </summary>
254+
public static unsafe Vector<uint> CreateWhileLessThanOrEqualMask32Bit(uint left, uint right) { throw new PlatformNotSupportedException(); }
255+
256+
/// <summary>
257+
/// svbool_t svwhilele_b32[_u64](uint64_t op1, uint64_t op2)
258+
/// </summary>
259+
public static unsafe Vector<uint> CreateWhileLessThanOrEqualMask32Bit(ulong left, ulong right) { throw new PlatformNotSupportedException(); }
260+
261+
262+
/// CreateWhileLessThanOrEqualMask64Bit : While incrementing scalar is less than or equal to
263+
264+
/// <summary>
265+
/// svbool_t svwhilele_b64[_s32](int32_t op1, int32_t op2)
266+
/// </summary>
267+
public static unsafe Vector<ulong> CreateWhileLessThanOrEqualMask64Bit(int left, int right) { throw new PlatformNotSupportedException(); }
268+
269+
/// <summary>
270+
/// svbool_t svwhilele_b64[_s64](int64_t op1, int64_t op2)
271+
/// </summary>
272+
public static unsafe Vector<ulong> CreateWhileLessThanOrEqualMask64Bit(long left, long right) { throw new PlatformNotSupportedException(); }
273+
274+
/// <summary>
275+
/// svbool_t svwhilele_b64[_u32](uint32_t op1, uint32_t op2)
276+
/// </summary>
277+
public static unsafe Vector<ulong> CreateWhileLessThanOrEqualMask64Bit(uint left, uint right) { throw new PlatformNotSupportedException(); }
278+
279+
/// <summary>
280+
/// svbool_t svwhilele_b64[_u64](uint64_t op1, uint64_t op2)
281+
/// </summary>
282+
public static unsafe Vector<ulong> CreateWhileLessThanOrEqualMask64Bit(ulong left, ulong right) { throw new PlatformNotSupportedException(); }
283+
284+
285+
/// CreateWhileLessThanOrEqualMask8Bit : While incrementing scalar is less than or equal to
286+
287+
/// <summary>
288+
/// svbool_t svwhilele_b8[_s32](int32_t op1, int32_t op2)
289+
/// </summary>
290+
public static unsafe Vector<byte> CreateWhileLessThanOrEqualMask8Bit(int left, int right) { throw new PlatformNotSupportedException(); }
291+
292+
/// <summary>
293+
/// svbool_t svwhilele_b8[_s64](int64_t op1, int64_t op2)
294+
/// </summary>
295+
public static unsafe Vector<byte> CreateWhileLessThanOrEqualMask8Bit(long left, long right) { throw new PlatformNotSupportedException(); }
296+
297+
/// <summary>
298+
/// svbool_t svwhilele_b8[_u32](uint32_t op1, uint32_t op2)
299+
/// </summary>
300+
public static unsafe Vector<byte> CreateWhileLessThanOrEqualMask8Bit(uint left, uint right) { throw new PlatformNotSupportedException(); }
301+
302+
/// <summary>
303+
/// svbool_t svwhilele_b8[_u64](uint64_t op1, uint64_t op2)
304+
/// </summary>
305+
public static unsafe Vector<byte> CreateWhileLessThanOrEqualMask8Bit(ulong left, ulong right) { throw new PlatformNotSupportedException(); }
306+
124307

125308
/// LoadVector : Unextended load
126309

0 commit comments

Comments
 (0)