>>...The performance will be

CLi37 · ‎01-29-2013

I do not think there are constant registers in x86. When I define a const array, x86 access these constants from a memory but not a direct constant in instruction. Any instructions can assign a 128bit/256bit constant to a SSE/AVX register?

TimP · ‎01-30-2013

Are you talking about C programming? About facilities of some specific compiler?

How about a short example to make this specific?

Bernard · ‎01-30-2013

chang-li wrote:

I do not think there are constant registers in x86. When I define a const array, x86 access these constants from a memory but not a direct constant in instruction. Any instructions can assign a 128bit/256bit constant to a SSE/AVX register?

You can access XMMn registers with the help of inline assembly.This is my preffered method of SSE -aware programming.In order to load XMM register I use align 16 directive on my typedef structure which holds single precision fp and double precision fp scalar values arranged in 1D array and I use movaps instruction to directly load XMMn registers.

CLi37 · ‎01-30-2013

In C language

const unsigned char data[16] = {0xFF, 0x00, 0x01, 0xFF,0xFF, 0x00, 0x01, 0xFF,0xFF, 0x00, 0x01, 0xFF,0xFF, 0x00, 0x01, 0xFF};
__m128i xmm0;

xmm0 = _mm_loadu_si128((__m128i *)data);

In ASM it becomes

movdqa xmm0, PQDWORD PTR [esi+4]

What I expected is

movdqa xmm0, 0xFF0001FFFF0001FFFF0001FFFF0001FF

I could not find this form in assembly.

Bernard · ‎01-30-2013

>>>What I expected is

movdqa xmm0, 0xFF0001FFFF0001FFFF0001FFFF0001FF

I could not find this form in assembly>>>

In MASM I can load XMM register directly by using declared primitive type with DUP directive.When using inline assembly you can load directly xmm register by using array name only and without using pointer dereference operator.

CLi37 · ‎01-30-2013

Do you have the right expression of inline assembly below?

movdqa xmm0, 0xFF0001FFFF0001FFFF0001FFFF0001FF

SergeyKostrov · ‎01-30-2013

>>Do you have the right expression of inline assembly below? >> >>movdqa xmm0, 0xFF0001FFFF0001FFFF0001FFFF0001FF I checked the Instruction Set Reference ( Order Number: 325383-044US / August 2012 ) and I see that movdqa can not be used with constants. Please take a look at a page 572 of the manual. Here is a quote: ... This instruction can be used to load an XMM register from a 128-bit memory location, to store the contents of an XMM register into a 128-bit memory location, or to move data between two XMM registers. ...

SergeyKostrov · ‎01-30-2013

Iliya, >>...In MASM I can load XMM register directly by using declared primitive type with DUP... We really would like to see how you do it, please. Thanks in advance.

Bernard · ‎01-30-2013

Sergey Kostrov wrote:

Iliya,

>>...In MASM I can load XMM register directly by using declared primitive type with DUP...

We really would like to see how you do it, please. Thanks in advance.

Here is the code which calculates cosine function by Taylor series expansion, please bear in mind that this code is not optimized and run slowly because of stack accesses which are not needed. I was able to load directly XMM register without dereferencing pointer.While reading the code please look at "coef" variables which are initialized to cosine series factorial denominators.

.XMM
.STACK 4096

.DATA

argument REAL4 0.0,0.0,0.0,0.0
step REAL4 0.01,0.01,0.01,0.01
hi_bound REAL4 1.0,1.0,1.0,1.0
lo_bound REAL4 0.0,0.0,0.0,0.0
up_range REAL4 1.0
lo_range REAL4 0.0
one REAL4 1.0,1.0,1.0,1.0
counter BYTE 147
coef1 REAL4 2.0,2.0,2.0,2.0 ;2!
coef2 REAL4 24.0,24.0,24.0,24.0 ;4!
coef3 REAL4 720.0,720.0,720.0,720.0 ;6!
coef4 REAL4 40320.0,40320.0,40320.0,40320.0 ;8!
coef5 REAL4 3628800.0,3628800.0,3628800.0,3628800.0 ;10!
coef6 REAL4 479001600.0,479001600.0,479001600.0,479001600.0 ;12!
coef7 REAL4 87178291200.0,87178291200.0,87178291200.0,87178291200.0 ;14!
coef8 REAL4 20922789888000.0,20922789888000.0,20922789888000.0,20922789888000.0 ;16!
coef9 REAL4 6402373705728000.0,6402373705728000.0,6402373705728000.0,6402373705728000.0 ;18!
coef10 REAL4 2432902008176640000.0,2432902008176640000.0,2432902008176640000.0,2432902008176640000.0 ;20!
coef11 REAL4 1124000727777607680000.0,1124000727777607680000.0,1124000727777607680000.0,1124000727777607680000.0 ;22!
coef12 REAL4 620448401733239439360000.0,620448401733239439360000.0,620448401733239439360000.0,620448401733239439360000.0;24!
coef13 REAL4 403291461126605635584000000.0,403291461126605635584000000.0,403291461126605635584000000.0,403291461126605635584000000.0 ;26!
loop_counter BYTE 50
loop_counter2 BYTE 25
loop_compare REAL4 0.5,0.5,0.5,0.5

.DATA?
result REAL4 147 DUP(?)
com_lo REAL4 4 DUP(?)
com_hi REAL4 4 DUP(?)
start_time DWORD ?
end_time DWORD ?
value REAL4 ?
upper REAL4 1.0
lower REAL4 0.0
counter_compare REAL4 4 DUP(?)

.CODE
main PROC

push ebp
mov esp,ebp
sub esp,224
mov cl,counter
xor eax,eax
xor ebx,ebx
xorps xmm2,xmm2
xorps xmm0,xmm0
xorps xmm1,xmm1
movups xmm5,argument
movups xmm0,argument

finit
mWrite "Please enter a starting value for cosine calculation"
call ReadFloat
fst value
call Crlf

fld upper
fcom value
fnstsw ax
sahf
jnb error

movss xmm5,value
movss xmm5,value
movss xmm5,value
movss xmm5,value
L1:
   movups xmm4,loop_compare
   movups xmm3,xmm5
   cmpps xmm3,xmm4,6
   mov eax,OFFSET counter_compare
   movups [eax],xmm3
   mov ebx,[eax]
   cmp ebx,0
   jne L2
   mov ebx,[eax+4]
   cmp ebx,0
   jne L2
   mov ebx,[eax+8]
   cmp ebx,0
   jne L2
   mov ebx,[eax+12]
   cmp ebx,0
   jne L2
   movups xmm4,loop_compare
   movups xmm3,xmm5
   cmpps xmm3,xmm4,1
   mov eax,OFFSET counter_compare
   movups [eax],xmm3
   mov ebx,[eax]
   cmp ebx,11111111111111111111111111111111b
   je L4
   mov ebx,[eax+4]
   cmp ebx,11111111111111111111111111111111b
   je L4
   mov ebx,[eax+8]
   cmp ebx,11111111111111111111111111111111b
   je L4
   mov ebx,[eax+12]
   cmp ebx,11111111111111111111111111111111b
   je L4
   xor eax,eax
   jz L3
L2:
mov cl,loop_counter
xor eax,eax
jz L3
L4:
mov cl,loop_counter2
xor eax,eax
jz L3

L3:
mov edx,OFFSET step
movups xmm4,[edx]
addps xmm5,xmm4

movups xmm7,xmm5
movups xmm0,one
mulps xmm7,xmm7 ;x^2
movups xmm6,xmm7
movups [ebp-16],xmm7 ;store x^7
movups xmm2,coef1
rcpps xmm1,xmm2 ;1/coef1
mulps xmm1,xmm7
subps xmm0,xmm1 ;1-x^2/2! xmm0 accumulator
movups xmm7,[ebp-16]
mulps xmm7,xmm6 ;x^4
movups [ebp-32],xmm7 ;store x^4
movups xmm2,coef2
rcpps xmm1,xmm2 ;1/coef2
mulps xmm1,xmm7
addps xmm0,xmm1 ;1-1x^2/2!+x^4/4!
movups xmm7,[ebp-32]
mulps xmm7,xmm6 ;x^6
movups [ebp-48],xmm7 ;store x^6
movups xmm2,coef3
rcpps xmm1,xmm2 ;1/coef3
mulps xmm1,xmm7
subps xmm0,xmm1 ;1-x^2/2!+x^4/4!-x^6/6!
movups xmm7,[ebp-48]
mulps xmm7,xmm6 ;x^8
movups [ebp-64],xmm7 ;store x^8
movups xmm2,coef4
rcpps xmm1,xmm2 ;1/coef4
mulps xmm1,xmm7
addps xmm0,xmm1 ;1-x^2/2!+x^4/4!-x^6/6!+x^8/8!
movups xmm7,[ebp-64]
mulps xmm7,xmm6 ;x^10
movups [ebp-80],xmm7 ;store x^10
movups xmm2,coef5
rcpps xmm1,xmm2 ;1/coef5
mulps xmm1,xmm7
subps xmm0,xmm1 ;1-x^2/2!+x^4/4!-x^6/6!+x^8/8!-x^10/10!
movups xmm7,[ebp-80]
mulps xmm7,xmm6 ;x^12
movups [ebp-96],xmm7 ;store x^12
movups xmm2,coef6      <---    XMM REGISTER IS DIRECTLY LOADED BY INITIALIZED COEF ARGUMENT
rcpps xmm1,xmm2 ;1/coef6
mulps xmm1,xmm7
addps xmm0,xmm1 ;1-x^2/2!+x^4/4!-x^6/6!+x^8/8!-x^10/10!+x^12/12!
movups xmm7,[ebp-96]
mulps xmm7,xmm6;x^14
movups [ebp-112],xmm7 ;store x^14
movups xmm2,coef7        <---    XMM REGISTER IS DIRECTLY LOADED BY INITIALIZED COEF ARGUMENT
rcpps xmm1,xmm2 ;1/coef7
mulps xmm1,xmm7
subps xmm0,xmm1 ;1-x^2/2!+x^4/4!-x^6/6!+x^8/8!-x^10/10!+x^12/12!-x^14/14!
movups xmm7,[ebp-112]
mulps xmm7,xmm6 ;x^16
movups [ebp-128],xmm7 ;store x^16
movups xmm2,coef8
rcpps xmm1,xmm2 ;1/coef8
mulps xmm1,xmm7
addps xmm0,xmm1 ;1-x^2/2!+x^4/4!-x^6/6!+x^8/8!-x^10/10!+x^12/12!-x^14/14!+x^16/16!
movups xmm7,[ebp-128]
mulps xmm7,xmm6 ;x^18
movups [ebp-144],xmm7;store x^18
movups xmm2,coef9
rcpps xmm1,xmm2 ;1/coef9
mulps xmm1,xmm7
subps xmm0,xmm1 ;1-x^2/2!+x^4/4!-x^6/6!+x^8/8!-x^10/10!+x^12/12!-x^14/14!+x^16/16!-x^18/18!
movups xmm7,[ebp-144]
mulps xmm7,xmm6 ;x^20
movups [ebp-160],xmm7 ;store x^20
movups xmm2,coef10
rcpps xmm1,xmm2 ;1/coef10
mulps xmm1,xmm7
addps xmm0,xmm1 ;1-x^2/2!+x^4/4!-x^6/6!+x^8/8!-x^10/10!+x^12/12!-x^14/14!+x^16/16!-x^18/18!+x^20/20!
movups xmm7,[ebp-160]
mulps xmm7,xmm6 ;x^22
movups [ebp-176],xmm7 ;store x^22
movups xmm2,coef11
rcpps xmm1,xmm2 ;1/coef11
mulps xmm1,xmm7
subps xmm0,xmm1;1-x^2/2!+x^4/4!-x^6/6!+x^8/8!-x^10/10!+x^12/12!-x^14/14!+x^16/16!-x^18/18!+x^20/20!-x^22/22!
movups xmm7,[ebp-176]
mulps xmm7,xmm6 ;x^24
movups [ebp-192],xmm7 ;store x^24
movups xmm2,coef12
rcpps xmm1,xmm2 ;1/coef12
mulps xmm1,xmm7
addps xmm0,xmm1 ; +x^24/24!
movups xmm7,[ebp-192]
mulps xmm7,xmm6 ;x^26
movups xmm2,coef13
rcpps xmm1,xmm2 ;1/coef13
mulps xmm1,xmm7
subps xmm0,xmm1

mov ebx,OFFSET result

movups [ebx],xmm0
fld DWORD PTR[ebx]
call WriteFloat
call Crlf
sub cl,1
jnz L3
xor eax,eax
jz L5

error:
movups xmm5,argument
xor eax,eax
jz L3

L5:
exit
main ENDP
END main

Bernard · ‎01-31-2013

My code works with movups instruction,but testing movdqa has not been ever done.

SergeyKostrov · ‎01-31-2013

Thank you. I see that you're using a different movups instruction ( actually it is OK / page 623 in a manual ) and this is how the instruction is used in your codes. >>... >>one REAL4 1.0,1.0,1.0,1.0......Note: memory is allocated here / It is Not a literal constant >>... >>movups xmm0, one >>... It means, that a set of values 1.0,1.0,1.0,1.0 is moved from a memory (!) to xmm0 register. Once again, both instructions, that is movdqa and movups, can not work with constants by design.

SergeyKostrov · ‎01-31-2013

>>...both instructions, that is movdqa and movups, can not work with constants by design... Here is a really small test case with inline assembler in a C/C++ test application: ... _asm MOVDQA xmm0, 0xFFFFFFFFFFFFFFFF ... [ Compilation output ] ... ..\prttests.cpp(8759) : error C2415: improper operand type ...

SergeyKostrov · ‎01-31-2013

Here is another small test case: ... __m128 mmValue = { 1.0L, 2.0L, 3.0L, 4.0L }; // _asm MOVDQA xmm0, 0xFFFFFFFFFFFFFFFF // error C2415: improper operand type // _asm MOVUPS xmm1, 0xFFFFFFFFFFFFFFFF // error C2415: improper operand type _asm MOVDQA xmm0, [ mmValue ] _asm MOVUPS xmm1, [ mmValue ] _asm MOVDQA [ mmValue ], xmm2 _asm MOVUPS [ mmValue ], xmm3 ...

Bernard · ‎01-31-2013

>>>It means, that a set of values 1.0,1.0,1.0,1.0 is moved from a memory (!) to xmm0 register. Once again, both instructions, that is movdqa and movups, can not work with constants by design>>>

I misunderstood the problem.In my code the load is coming from the memory and this was not the thread starter's question.The problem is "how to load XMM register with the immediate value".

SergeyKostrov · ‎01-31-2013

Exactly and I'd like to change my former statement to: >>...both instructions, that is movdqa and movups, can not work with literal constants by design...

Bernard · ‎01-31-2013

>>>// _asm MOVDQA xmm0, 0xFFFFFFFFFFFFFFFF // error C2415: improper operand type>>>

Now I remember I have the same situation when I tried to load directly(immediate value)XMM registers.The second test is my preffered method of loading 1D vector represented by the 2 or 4 elements array into XMM register.You can also load the registers when passing structure members.For this use unaligned movups instruction.

Bernard · ‎01-31-2013

Sergey Kostrov wrote:

Exactly and I'd like to change my former statement to:

>>...both instructions, that is movdqa and movups, can not work with literal constants by design...

Sadly Intel processor designers decided to not allow loading SSEn registers with the immediate values.I would like to know what is the cause of such a decision.

CLi37 · ‎01-31-2013

>>one REAL4 1.0,1.0,1.0,1.0......Note: memory is allocated here / It is Not a literal constant
>>movups xmm0, one

It also assing xmm0 to a value of a vector with 4 components.

"how to load XMM register with the immediate value"

yes, this is the exact question. Looks answer is no. So following code

movups xmm0,one

because the data one was loaded from memory, the algorithm is not all register based, in which the cache access may be related. The performance will be unexpected. The optimization on SSE/AVX may be collapsed.

SergeyKostrov · ‎01-31-2013

>>...The performance will be unexpected. The optimization on SSE/AVX may be collapsed... These are a different set of issues ( actually I don't see any problems here / this is how Intel designed these instructions ). The instruction formats are in the manual and please take a look. Here is another test case ( compiled with Intel C++ compiler ): ... __m128 mmValue0 = { 1.0L, 2.0L, 3.0L, 4.0L }; __m128 mmValue1 = { 5.0L, 6.0L, 7.0L, 8.0L }; _asm MOVDQA xmm0, 0xFFFFFFFFFFFFFFFF // Microsoft C++ compiler: Error C2415: improper operand type _asm MOVUPS xmm1, 0xFFFFFFFFFFFFFFFF // Microsoft C++ compiler: Error C2415: improper operand type _asm MOVDQA xmm0, xmmword ptr [ mmValue0 ] _asm MOVUPS xmm1, xmmword ptr [ mmValue1 ] _asm MOVDQA xmmword ptr [ mmValue0 ], xmm1 _asm MOVUPS xmmword ptr [ mmValue1 ], xmm0 ... [ Intel C++ compiler output ] ... ..\PrtTests.cpp(8762): (col. 8) error: Unsupported instruction form in asm instruction movdqa. ..\PrtTests.cpp(8763): (col. 8) error: Unsupported instruction form in asm instruction movups. (0): catastrophic error: fatal error: compilation terminated ...

Bernard · ‎01-31-2013

Because of the design of those SSEn load/store instructions there will be some performance penalty when the data needs to be loaded first time from the memory.

Loading of the XMM register with the pre-initialized structure member.

SinVector sinvec1 = {-0.1666666,-0.1666666,-0.1666666,-0.1666666},*sinvec1ptr; sinvec1ptr = &sinvec1 // structure initialization

Loading of the member

movups xmm1,sinvec1

By using custom typedef structure of array data type which is aligned on 16 bytes boundaries I can use movaps instructions.

Bernard · ‎01-31-2013

Because of the design of those SSEn load/store instructions there will be some performance penalty when the data needs to be loaded first time from the memory.

Loading of the XMM register with the pre-initialized structure member:

SinVector sinvec1 = {-0.1666666,-0.1666666,-0.1666666,-0.1666666},*sinvec1ptr; sinvec1ptr = &sinvec1 // structure initialization

Loading of the member:

movups xmm1,sinvec1

By using custom typedef structure of array data type which is aligned on 16 bytes boundaries I can use movaps instructions.

How to assign a constant?