_MM_SHUFFLE question

[swati24]

Hi,

This question is regarding the SSE macro function for shuffle operations. I don't understand it fully.

MSDN says that _MM_SHUFFLE(z,y,x,w) expands to :

(z << 6) | (y <<4) | (x << 2) | w

This macro is used in conjunction with SHUFPS or MMX instrinsic function _mm_shuffle_ps(m1, m2, int mask), where mask is replaced by the macro that acts on m1 and m2. m1 and m2 are 128 bit registers.

MSDN has the following example:

Let m1 be a : b : c : d
Let m2 be e : f : g : h

where each of the a,b,c,d,e,f,g,h are 32 bit single precision floating point values. a is the highest double word and d is the lowest double word. Similarly, e is the highest double word and h is the lowest double word.

Now, when the following function is performed on the m1 and m2 MMX registers:

_mm_shuffle_ps(m1,m2,_MM_SHUFFLE(1,0,3,2))

we get

m3 g : h : a : b

-------------
Working

I tried to work this example myself but couldn't arrive at the right answer.

1 << 6 gives 0100 0000
0 << 4 gives 0000 0000
3 << 2 gives 0000 1100
2 gives 0000 0010

Bitwise OR operation gives 0100 1110

If we apply this mask on m1 and m2 how do we end up with m3? What am I doing wrong here. Any insights will be most helpful.

Thank you
Swati

[Hobson]

According to Intel Instruction Reference Manual:


SHUFPS—Shuffle Single-FP operations
FP_SELECT = (imm8 >> 0) AND 0X3;
IF (FP_SELECT = 0) THEN
DEST[31-0] = DEST[31-0];
ELSE
IF (FP_SELECT = 1) THEN
DEST[31-0] = DEST[63-32];
ELSE
IF (FP_SELECT = 2) THEN
DEST[31-0] = DEST[95-64];
ELSE
DEST[31-0] = DEST[127-96];
FI
FI
FI

FP_SELECT = (imm8 >> 2) AND 0X3;
IF (FP_SELECT = 0) THEN
DEST[63-32] = DEST[31-0];
ELSE
IF (FP_SELECT = 1) THEN
DEST[63-32] = DEST[63-32];
ELSE
IF (FP_SELECT = 2) THEN
DEST[63-32] = DEST[95-64];
ELSE
DEST[63-32] = DEST[127-96];
FI
FI
FI

FP_SELECT = (imm8 >> 4) AND 0X3;
IF (FP_SELECT = 0) THEN
DEST[95-64] = SRC/m128[31-0];
ELSE
IF (FP_SELECT = 1) THEN
DEST[95-64] = SRC/m128 [63-32];
ELSE
IF (FP_SELECT = 2) THEN
DEST[95-64] = SRC/m128 [95-64];
ELSE
DEST[95-64] = SRC/m128 [127-96];
FI
FI
FI

FP_SELECT = (imm8 >> 6) AND 0X3;
IF (FP_SELECT = 0) THEN
DEST[127-96] = SRC/m128 [31-0];
ELSE
IF (FP_SELECT = 1) THEN
DEST[127-96] = SRC/m128 [63-32];
ELSE
IF (FP_SELECT = 2) THEN
DEST[127-96] = SRC/m128 [95-64];
ELSE
DEST[127-96] = SRC/m128 [127-96];
FI
FI
FI


This is not very complicated instruction, all it does is that it decomposites 8bit mask to 4 2bit numbers, and selects appropriate fragments from appropriate source operands and places them in appropriate places in dest operand. There is no shifting\xoring\anding or any other bit logic magic.

Hob

[swati24]

I should have included the URL for the MSDN explanation of _MM_SHUFFLE, here it is:
http://msdn.microsoft.com/library/default.asp?url=/library/en-us/vclang/html/vcrefmacrofunctionforshuffleusingstreamingsimdextensions.asp

[swati24]

Thank you Hobson. That was very useful. It is a simple selection operation of the operands m1 and m2. So, _MM_SHUFFLE(z,y,x,w) selects x&w 32 bit double words from m1 and z&y from m2. How simple!!.

Thanks
Swati

[Hobson]

Yes, it is simple, however, one little very formal suggestion:
_MM_SHUFFLE(z,y,x,w) does not select anything, this macro just creates a mask. SHUFPS instruction (or _mm_shuffle_ps wrapper function) performs selection, using mask created by _MM_SHUFFLE macro.

Hob