Add support of CreateScalarUnsafe() for arm64 intrinsic in JIT #34579

kunalspathak · 2020-04-06T08:13:17Z

Add support for Vector64.CreateUnsafe() and Vector128.CreateUnsafe() inside JIT that maps to one of fmov, ins or movi/movni instruction.

For float/double, if operand is immediate and can be encoded in fmov, use it. Else, use fmov on register.
For byte/sbyte/short/ushort/int/uint/long/ulong, if operand is immediate and can be encoded in movi/movi, use it. Otherwise, load the value in register and use ins to save in 0th lane.

Thanks @echesakovMSFT for the suggestion in #34485 (comment)

static void Vector64Test(Single value1, Single value2)
{
	var data0 = Vector64.CreateScalarUnsafe((float)5.9f);
	Console.WriteLine(data0);
	var data1 = Vector64.CreateScalarUnsafe((float)2.0f);
	Console.WriteLine(data1);
	var data2 = Vector64.CreateScalarUnsafe(4353);
	Console.WriteLine(data2);	
	var data3 = Vector64.CreateScalarUnsafe(32);
	Console.WriteLine(data3);
}

Output:

<5.9, 0>
<2, 0>
<4353, 0>
<32, 32>
<1.0696732, 0>
<43, 0>

Assembly output

G_M707_IG02:
       D28AE400          movz    x0, #0x5720
       F2A70740          movk    x0, #0x383a LSL #16
       F2CFFFA0          movk    x0, #0x7ffd LSL #32
       97FF15C2          bl      CORINFO_HELP_NEWSFAST
       1C0005B0          ldr     s16, [@RWD00]
       FD000410          str     d16, [x0,#8]
       97FFADE3          bl      System.Console:WriteLine(System.Object)
       D28AE400          movz    x0, #0x5720
       F2A70740          movk    x0, #0x383a LSL #16
       F2CFFFA0          movk    x0, #0x7ffd LSL #32
       97FF15BB          bl      CORINFO_HELP_NEWSFAST
       1E201010          fmov    s16, #2.0000
       FD000410          str     d16, [x0,#8]
       97FFADDC          bl      System.Console:WriteLine(System.Object)
       D28BA800          movz    x0, #0x5d40
       F2A70740          movk    x0, #0x383a LSL #16
       F2CFFFA0          movk    x0, #0x7ffd LSL #32
       97FF15B4          bl      CORINFO_HELP_NEWSFAST
       52822021          mov     w1, #0x1101
       4E041C30          ins     v16.s[0], w1
       FD000410          str     d16, [x0,#8]
       97FFADD4          bl      System.Console:WriteLine(System.Object)
       D28BA800          movz    x0, #0x5d40
       F2A70740          movk    x0, #0x383a LSL #16
       F2CFFFA0          movk    x0, #0x7ffd LSL #32
       97FF15AC          bl      CORINFO_HELP_NEWSFAST
       0F010410          movi    v16.2s, #0x20
       FD000410          str     d16, [x0,#8]
       97FFADCD          bl      System.Console:WriteLine(System.Object)
       D28AE400          movz    x0, #0x5720
       F2A70740          movk    x0, #0x383a LSL #16
       F2CFFFA0          movk    x0, #0x7ffd LSL #32
       97FF15A5          bl      CORINFO_HELP_NEWSFAST
       BD401FA0          ldr     s0, [fp,#28]  // [V00 arg0]
       BD401BA1          ldr     s1, [fp,#24]  // [V01 arg1]
       1E212810          fadd    s16, s0, s1
       FD000410          str     d16, [x0,#8]
       97FFADC4          bl      System.Console:WriteLine(System.Object)
       D28BA800          movz    x0, #0x5d40
       F2A70740          movk    x0, #0x383a LSL #16
       F2CFFFA0          movk    x0, #0x7ffd LSL #32
       97FF159C          bl      CORINFO_HELP_NEWSFAST
       0B140261          add     w1, w19, w20
       4E041C30          ins     v16.s[0], w1
       FD000410          str     d16, [x0,#8]
       97FFADBC          bl      System.Console:WriteLine(System.Object)

RWD00  dd       40BCCCCDh

static void Vector128Test(double value1, double value2, long value3, long value4)
{
	var data0 = Vector128.CreateScalarUnsafe((float)5.9f);
	Console.WriteLine(data0);
	var data1 = Vector128.CreateScalarUnsafe((float)2.0f);
	Console.WriteLine(data1);
	var data2 = Vector128.CreateScalarUnsafe((long)4353);
	Console.WriteLine(data2);	
	var data3 = Vector128.CreateScalarUnsafe(32);
	Console.WriteLine(data3);
	var data4 = Vector128.CreateScalarUnsafe(value1 + value2);
	Console.WriteLine(data4);
	var data5 = Vector128.CreateScalarUnsafe(value3 + value4);
	Console.WriteLine(data5);
}

Output:

<5.9, 0, 0, 0>
<2, 0, 0, 0>
<4353, 0>
<32, 32, 32, 32>
<0.30312630808659924, 0>
<12884901882, 0>

Assembly output

G_M3610_IG02:
       D28C2E00          movz    x0, #0x6170
       F2A70720          movk    x0, #0x3839 LSL #16
       F2CFFFA0          movk    x0, #0x7ffd LSL #32
       97FF13D2          bl      CORINFO_HELP_NEWSFAST
       1C0005B0          ldr     s16, [@RWD00]
       3C808010          str     q16, [x0,#8]
       97FFAC09          bl      System.Console:WriteLine(System.Object)
       D28C2E00          movz    x0, #0x6170
       F2A70720          movk    x0, #0x3839 LSL #16
       F2CFFFA0          movk    x0, #0x7ffd LSL #32
       97FF13CB          bl      CORINFO_HELP_NEWSFAST
       1E201010          fmov    s16, #2.0000
       3C808010          str     q16, [x0,#8]
       97FFAC02          bl      System.Console:WriteLine(System.Object)
       D28CFD00          movz    x0, #0x67e8
       F2A70720          movk    x0, #0x3839 LSL #16
       F2CFFFA0          movk    x0, #0x7ffd LSL #32
       97FF13C4          bl      CORINFO_HELP_NEWSFAST
       D2822021          mov     x1, #0x1101
       4E081C30          ins     v16.d[0], x1
       3C808010          str     q16, [x0,#8]
       97FFABFA          bl      System.Console:WriteLine(System.Object)
       D28DCC00          movz    x0, #0x6e60
       F2A70720          movk    x0, #0x3839 LSL #16
       F2CFFFA0          movk    x0, #0x7ffd LSL #32
       97FF13BC          bl      CORINFO_HELP_NEWSFAST
       4F010410          movi    v16.4s, #0x20
       3C808010          str     q16, [x0,#8]
       97FFABF3          bl      System.Console:WriteLine(System.Object)
       D28E9B00          movz    x0, #0x74d8
       F2A70720          movk    x0, #0x3839 LSL #16
       F2CFFFA0          movk    x0, #0x7ffd LSL #32
       97FF13B5          bl      CORINFO_HELP_NEWSFAST
       FD400FA0          ldr     d0, [fp,#24]  // [V00 arg0]
       FD400BA1          ldr     d1, [fp,#16]  // [V01 arg1]
       1E612810          fadd    d16, d0, d1
       3C808010          str     q16, [x0,#8]
       97FFABEA          bl      System.Console:WriteLine(System.Object)
       D28CFD00          movz    x0, #0x67e8
       F2A70720          movk    x0, #0x3839 LSL #16
       F2CFFFA0          movk    x0, #0x7ffd LSL #32
       97FF13AC          bl      CORINFO_HELP_NEWSFAST
       8B140261          add     x1, x19, x20
       4E081C30          ins     v16.d[0], x1
       3C808010          str     q16, [x0,#8]
       97FFABE2          bl      System.Console:WriteLine(System.Object)

RWD00  dd       40BCCCCDh

Fixes: #34485
Contributes to #33308

kunalspathak · 2020-04-25T06:16:36Z

/cc: @dotnet/jit-contrib , @tannergooding , @TamarChristinaArm

…rt, int, uint, long, ulong

src/coreclr/src/jit/hwintrinsiccodegenarm64.cpp

src/coreclr/src/jit/hwintrinsiclistarm64.h

src/coreclr/src/jit/lowerarmarch.cpp

echesakov

LGTM (minus nit)

src/coreclr/src/jit/hwintrinsiccodegenarm64.cpp

kunalspathak · 2020-04-27T23:11:01Z

Failure related to #35502

Dotnet-GitSync-Bot added the area-CodeGen-coreclr CLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI label Apr 6, 2020

kunalspathak force-pushed the create-scalar-unsafe branch from 825f28f to ef15294 Compare April 25, 2020 05:32

kunalspathak marked this pull request as ready for review April 25, 2020 06:16

jaredpar mentioned this pull request Apr 27, 2020

Test timeline issue jaredpar/runfo#8

Closed

kunalspathak added 5 commits April 27, 2020 09:39

Intrinsify Vector64.CreateScalarUnsafe() for byte, sbyte, short, usho…

4e47301

…rt, int, uint, long, ulong

Intrinsify Vector64.CreateScalarUnsafe() for float

a41856f

Fixed minor things and code format

f8ff810

Intrinsify Vector128.CreateScalarUnsafe() for all

0c280c8

fix formatting

b55e6d4

kunalspathak force-pushed the create-scalar-unsafe branch from 5e41e37 to b55e6d4 Compare April 27, 2020 16:41

echesakov suggested changes Apr 27, 2020

View reviewed changes

Review feedback

75a1a3a

jaredpar mentioned this pull request Apr 27, 2020

OSX machines are de-provisioned during CI / PR runs leading to failures #34472

Closed

echesakov approved these changes Apr 27, 2020

View reviewed changes

src/coreclr/src/jit/hwintrinsiccodegenarm64.cpp Show resolved Hide resolved

kunalspathak merged commit d68342c into dotnet:master Apr 27, 2020

kunalspathak deleted the create-scalar-unsafe branch April 27, 2020 23:14

ghost locked as resolved and limited conversation to collaborators Dec 9, 2020

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add support of CreateScalarUnsafe() for arm64 intrinsic in JIT #34579

Add support of CreateScalarUnsafe() for arm64 intrinsic in JIT #34579

kunalspathak commented Apr 6, 2020 •

edited

Loading

kunalspathak commented Apr 25, 2020 •

edited

Loading

echesakov left a comment

kunalspathak commented Apr 27, 2020

Add support of CreateScalarUnsafe() for arm64 intrinsic in JIT #34579

Add support of CreateScalarUnsafe() for arm64 intrinsic in JIT #34579

Conversation

kunalspathak commented Apr 6, 2020 • edited Loading

kunalspathak commented Apr 25, 2020 • edited Loading

echesakov left a comment

Choose a reason for hiding this comment

kunalspathak commented Apr 27, 2020

kunalspathak commented Apr 6, 2020 •

edited

Loading

kunalspathak commented Apr 25, 2020 •

edited

Loading