Vector MaxDiff

Overview

Vector maxdiff.

Performs the integer subtraction between the lanes of X/Y (selected using the scheme described in the Advanced Compare page) and returns the maximum between zero and the subtraction result as a lane in the output vector.

Functions
v16int32	maxdiff16 (v16int32 xbuff, int xstart, unsigned int xoffsets, unsigned int xoffsets_hi, int ystart, unsigned int yoffsets, unsigned int yoffsets_hi)
	Performs a maximum difference computation between lanes of xbuff.
v16int32	maxdiff16 (v16int32 xbuff, int xstart, unsigned int xoffsets, unsigned int xoffsets_hi, v16int32 ybuff, int ystart, unsigned int yoffsets, unsigned int yoffsets_hi)
	Performs a maximum difference computation between lanes of xbuff and ybuff.
v16int32	maxdiff16 (v32int32 xbuff, int xstart, unsigned int xoffsets, unsigned int xoffsets_hi, int ystart, unsigned int yoffsets, unsigned int yoffsets_hi)
	Performs a maximum difference computation between lanes of xbuff.
v32int16	maxdiff32 (v32int16 xbuff, int xstart, unsigned int xoffsets, unsigned int xoffsets_hi, unsigned int xsquare, int ystart, unsigned int yoffsets, unsigned int yoffsets_hi, unsigned int ysquare)
	Performs a maximum difference computation between lanes of xbuff.
v32int16	maxdiff32 (v32int16 xbuff, int xstart, unsigned int xoffsets, unsigned int xoffsets_hi, unsigned int xsquare, v32int16 ybuff, int ystart, unsigned int yoffsets, unsigned int yoffsets_hi, unsigned int ysquare)
	Performs a maximum difference computation between lanes of xbuff and ybuff.
v32int16	maxdiff32 (v64int16 xbuff, int xstart, unsigned int xoffsets, unsigned int xoffsets_hi, unsigned int xsquare, int ystart, unsigned int yoffsets, unsigned int yoffsets_hi, unsigned int ysquare)
	Performs a maximum difference computation between lanes of xbuff.
v16int32	maxdiffcmp16 (v16int32 xbuff, int xstart, unsigned int xoffsets, unsigned int xoffsets_hi, int ystart, unsigned int yoffsets, unsigned int yoffsets_hi, unsigned int &cmp)
	Performs a maximum difference computation between lanes of xbuff.
v16int32	maxdiffcmp16 (v16int32 xbuff, int xstart, unsigned int xoffsets, unsigned int xoffsets_hi, v16int32 ybuff, int ystart, unsigned int yoffsets, unsigned int yoffsets_hi, unsigned int &cmp)
	Performs a maximum difference computation between lanes of xbuff and ybuff.
v16int32	maxdiffcmp16 (v32int32 xbuff, int xstart, unsigned int xoffsets, unsigned int xoffsets_hi, int ystart, unsigned int yoffsets, unsigned int yoffsets_hi, unsigned int &cmp)
	Performs a maximum difference computation between lanes of xbuff.
v32int16	maxdiffcmp32 (v32int16 xbuff, int xstart, unsigned int xoffsets, unsigned int xoffsets_hi, unsigned int xsquare, int ystart, unsigned int yoffsets, unsigned int yoffsets_hi, unsigned int ysquare, unsigned int &cmp)
	Performs a maximum difference computation between lanes of xbuff.
v32int16	maxdiffcmp32 (v32int16 xbuff, int xstart, unsigned int xoffsets, unsigned int xoffsets_hi, unsigned int xsquare, v32int16 ybuff, int ystart, unsigned int yoffsets, unsigned int yoffsets_hi, unsigned int ysquare, unsigned int &cmp)
	Performs a maximum difference computation between lanes of xbuff and ybuff.
v32int16	maxdiffcmp32 (v64int16 xbuff, int xstart, unsigned int xoffsets, unsigned int xoffsets_hi, unsigned int xsquare, int ystart, unsigned int yoffsets, unsigned int yoffsets_hi, unsigned int ysquare, unsigned int &cmp)
	Performs a maximum difference computation between lanes of xbuff.

Function Documentation

maxdiff16() [1/3]

v16int32 maxdiff16	(	v16int32	xbuff,
		int	xstart,
		unsigned int	xoffsets,
		unsigned int	xoffsets_hi,
		int	ystart,
		unsigned int	yoffsets,
		unsigned int	yoffsets_hi
	)

Performs a maximum difference computation between lanes of xbuff.

maxdiff(a, b)
{
  d = a - b;
  return max(d, 0);
}

for (int i = 0; i < 16; i++)
    idx = f( xstart, xoffsets[i]);
    idy = f( ystart, yoffsets[i]);
    o[i] = maxdiff(x[idx], x[idy])
 
xoffsets, xoffsets_hi, yoffsets, yoffsets_hi have 8 offset values each. 4 bits per offset.
For Example: for v16int32 output type, idx for output_lane_0 = f(xstart,xoffsets[0])
For Example: for v16int32 output type, idx for output_lane_15 = f(xstart,xoffsets_hi[7])
In case of v32int16, 1 offset is used for 2 adjacent lanes.
For more information on how the function f() selects data from the buffers refer to Lane selection note below.

Returns

Value of each lane is the result of a maximum difference computation between lanes of xbuff where the result of lane 0 goes to lane 0 of the output.

Parameters

xbuff	Input buffer of 16 elements with 32-bit precision
xstart	Starting position offset applied to all lanes of input from X buffer
xoffsets	4b offset for each lane, applied to the xbuffer. LSB apply to first lane
xoffsets_hi	4b offset for each lane, applied to the xbuffer. LSB apply to 8th lane
ystart	Starting position offset applied to all lanes of input from xbuffer for the second input
yoffsets	4b offset for each lane, applied to the xbuffer. LSB apply to first lane
yoffsets_hi	4b offset for each lane, applied to the xbuffer. LSB apply to 8th lane

Note

• For more information on how the function f() selects data from the buffers go here.
• Data from xbuff using 'xstart','xoffsets(_hi)' params is the left operand and data from xbuff using 'ystart','yoffsets(_hi)' params is the right operand.

maxdiff16() [2/3]

v16int32 maxdiff16	(	v16int32	xbuff,
		int	xstart,
		unsigned int	xoffsets,
		unsigned int	xoffsets_hi,
		v16int32	ybuff,
		int	ystart,
		unsigned int	yoffsets,
		unsigned int	yoffsets_hi
	)

Performs a maximum difference computation between lanes of xbuff and ybuff.

maxdiff(a, b)
{
  d = a - b;
  return max(d, 0);
}

for (int i = 0; i < 16; i++)
    idx = f( xstart, xoffsets[i]);
    idy = f( ystart, yoffsets[i]);
    o[i] = maxdiff(x[idx], y[idy])
 
xoffsets, xoffsets_hi, yoffsets, yoffsets_hi have 8 offset values each. 4 bits per offset.
For Example: for v16int32 output type, idx for output_lane_0 = f(xstart,xoffsets[0])
For Example: for v16int32 output type, idx for output_lane_15 = f(xstart,xoffsets_hi[7])
In case of v32int16, 1 offset is used for 2 adjacent lanes.
For more information on how the function f() selects data from the buffers refer to Lane selection note below.

Returns

Value of each lane is the result of a maximum difference computation between lanes of xbuff where the result of lane 0 goes to lane 0 of the output.

Parameters

xbuff	Input buffer of 16 elements with 32-bit precision
xstart	Starting position offset applied to all lanes of input from X buffer
xoffsets	4b offset for each lane, applied to the xbuffer. LSB apply to first lane
xoffsets_hi	4b offset for each lane, applied to the xbuffer. LSB apply to 8th lane
ybuff	Input buffer of 16 elements with 32-bit precision
ystart	Starting position offset applied to all lanes of input from ybuffer for the second input
yoffsets	4b offset for each lane, applied to the ybuffer. LSB apply to first lane
yoffsets_hi	4b offset for each lane, applied to the ybuffer. LSB apply to 8th lane

Note

• For more information on how the function f() selects data from the buffers go here.
• Data from xbuff is the left operand and data from ybuff is the right operand.

maxdiff16() [3/3]

v16int32 maxdiff16	(	v32int32	xbuff,
		int	xstart,
		unsigned int	xoffsets,
		unsigned int	xoffsets_hi,
		int	ystart,
		unsigned int	yoffsets,
		unsigned int	yoffsets_hi
	)

Performs a maximum difference computation between lanes of xbuff.

maxdiff(a, b)
{
  d = a - b;
  return max(d, 0);
}

for (int i = 0; i < 16; i++)
    idx = f( xstart, xoffsets[i]);
    idy = f( ystart, yoffsets[i]);
    o[i] = maxdiff(x[idx], x[idy])
 
xoffsets, xoffsets_hi, yoffsets, yoffsets_hi have 8 offset values each. 4 bits per offset.
For Example: for v16int32 output type, idx for output_lane_0 = f(xstart,xoffsets[0])
For Example: for v16int32 output type, idx for output_lane_15 = f(xstart,xoffsets_hi[7])
In case of v32int16, 1 offset is used for 2 adjacent lanes.
For more information on how the function f() selects data from the buffers refer to Lane selection note below.

Returns

Value of each lane is the result of a maximum difference computation between lanes of xbuff where the result of lane 0 goes to lane 0 of the output.

Parameters

xbuff	Input buffer of 32 elements with 32-bit precision
xstart	Starting position offset applied to all lanes of input from X buffer
xoffsets	4b offset for each lane, applied to the xbuffer. LSB apply to first lane
xoffsets_hi	4b offset for each lane, applied to the xbuffer. LSB apply to 8th lane
ystart	Starting position offset applied to all lanes of input from xbuffer for the second input
yoffsets	4b offset for each lane, applied to the xbuffer. LSB apply to first lane
yoffsets_hi	4b offset for each lane, applied to the xbuffer. LSB apply to 8th lane

Note

• For more information on how the function f() selects data from the buffers go here.
• Data from xbuff using 'xstart','xoffsets(_hi)' params is the left operand and data from xbuff using 'ystart','yoffsets(_hi)' params is the right operand.

maxdiff32() [1/3]

v32int16 maxdiff32	(	v32int16	xbuff,
		int	xstart,
		unsigned int	xoffsets,
		unsigned int	xoffsets_hi,
		unsigned int	xsquare,
		int	ystart,
		unsigned int	yoffsets,
		unsigned int	yoffsets_hi,
		unsigned int	ysquare
	)

Performs a maximum difference computation between lanes of xbuff.

maxdiff(a, b)
{
  d = a - b;
  return max(d, 0);
}

for (int i = 0; i < 32; i++)
    idx = f( xstart, xoffsets[i],xsquare);
    idy = f( ystart, yoffsets[i],ysquare);
    o[i] = maxdiff(x[idx], x[idy])
 
xoffsets, xoffsets_hi, yoffsets, yoffsets_hi have 8 offset values each. 4 bits per offset.
For Example: for v16int32 output type, idx for output_lane_0 = f(xstart,xoffsets[0],xsquare)
For Example: for v16int32 output type, idx for output_lane_15 = f(xstart,xoffsets_hi[7],xsquare)
In case of v32int16, 1 offset is used for 2 adjacent lanes.
For more information on how the function f() selects data from the buffers refer to Lane selection note below.

Returns

Value of each lane is the result of a maximum difference computation between lanes of xbuff where the result of lane 0 goes to lane 0 of the output.

Parameters

xbuff	Input buffer of 32 elements with 16-bit precision
xstart	Starting position offset applied to all lanes of input from X buffer
xoffsets	4b offset for each lane, applied to the xbuffer. LSB apply to first lane
xoffsets_hi	4b offset for each lane, applied to the xbuffer. LSB apply to 16th lane
xsquare	Select order of the mini-permute square (default=0x3210). LSB apply to first element. Value per lane needs to be less than 4. max value for this field is (0x3333)
ystart	Starting position offset applied to all lanes of input from xbuffer for the second input
yoffsets	4b offset for each lane, applied to the xbuffer. LSB apply to first lane
yoffsets_hi	4b offset for each lane, applied to the xbuffer. LSB apply to 16th lane
ysquare	Select order of the mini-permute square (default=0x3210). LSB apply to first element. Value per lane needs be less than 4. max value for this field is (0x3333)

Note

• This intrinsic uses the 'square' parameter, to have more information on how to use this please go here
• For more information on how the function f() selects data from the buffers go here.
• Data from xbuff using 'xstart','xoffsets(_hi)' params is the left operand and data from xbuff using 'ystart','yoffsets(_hi)' params is the right operand.

maxdiff32() [2/3]

v32int16 maxdiff32	(	v32int16	xbuff,
		int	xstart,
		unsigned int	xoffsets,
		unsigned int	xoffsets_hi,
		unsigned int	xsquare,
		v32int16	ybuff,
		int	ystart,
		unsigned int	yoffsets,
		unsigned int	yoffsets_hi,
		unsigned int	ysquare
	)

Performs a maximum difference computation between lanes of xbuff and ybuff.

maxdiff(a, b)
{
  d = a - b;
  return max(d, 0);
}

for (int i = 0; i < 32; i++)
    idx = f( xstart, xoffsets[i],xsquare);
    idy = f( ystart, yoffsets[i],ysquare);
    o[i] = maxdiff(x[idx], y[idy])
 
xoffsets, xoffsets_hi, yoffsets, yoffsets_hi have 8 offset values each. 4 bits per offset.
For Example: for v16int32 output type, idx for output_lane_0 = f(xstart,xoffsets[0],xsquare)
For Example: for v16int32 output type, idx for output_lane_15 = f(xstart,xoffsets_hi[7],xsquare)
In case of v32int16, 1 offset is used for 2 adjacent lanes.
For more information on how the function f() selects data from the buffers refer to Lane selection note below.

Returns

Value of each lane is the result of a maximum difference computation between lanes of xbuff where the result of lane 0 goes to lane 0 of the output.

Parameters

xbuff	Input buffer of 32 elements with 16-bit precision
xstart	Starting position offset applied to all lanes of input from X buffer
xoffsets	4b offset for each lane, applied to the xbuffer. LSB apply to first lane
xoffsets_hi	4b offset for each lane, applied to the xbuffer. LSB apply to 16th lane
xsquare	Select order of the mini-permute square (default=0x3210). LSB apply to first element. Value per lane needs to be less than 4. max value for this field is (0x3333)
ybuff	Input buffer of 32 elements with 16-bit precision
ystart	Starting position offset applied to all lanes of input from ybuffer for the second input
yoffsets	4b offset for each lane, applied to the ybuffer. LSB apply to first lane
yoffsets_hi	4b offset for each lane, applied to the ybuffer. LSB apply to 16th lane
ysquare	Select order of the mini-permute square (default=0x3210). LSB apply to first element. Value per lane needs be less than 4. max value for this field is (0x3333)

Note

• This intrinsic uses the 'square' parameter, to have more information on how to use this please go here
• For more information on how the function f() selects data from the buffers go here.
• Data from xbuff is the left operand and data from ybuff is the right operand.

maxdiff32() [3/3]

v32int16 maxdiff32	(	v64int16	xbuff,
		int	xstart,
		unsigned int	xoffsets,
		unsigned int	xoffsets_hi,
		unsigned int	xsquare,
		int	ystart,
		unsigned int	yoffsets,
		unsigned int	yoffsets_hi,
		unsigned int	ysquare
	)

Performs a maximum difference computation between lanes of xbuff.

maxdiff(a, b)
{
  d = a - b;
  return max(d, 0);
}

for (int i = 0; i < 32; i++)
    idx = f( xstart, xoffsets[i],xsquare);
    idy = f( ystart, yoffsets[i],ysquare);
    o[i] = maxdiff(x[idx], x[idy])
 
xoffsets, xoffsets_hi, yoffsets, yoffsets_hi have 8 offset values each. 4 bits per offset.
For Example: for v16int32 output type, idx for output_lane_0 = f(xstart,xoffsets[0],xsquare)
For Example: for v16int32 output type, idx for output_lane_15 = f(xstart,xoffsets_hi[7],xsquare)
In case of v32int16, 1 offset is used for 2 adjacent lanes.
For more information on how the function f() selects data from the buffers refer to Lane selection note below.

Returns

Value of each lane is the result of a maximum difference computation between lanes of xbuff where the result of lane 0 goes to lane 0 of the output.

Parameters

xbuff	Input buffer of 64 elements with 16-bit precision
xstart	Starting position offset applied to all lanes of input from X buffer
xoffsets	4b offset for each lane, applied to the xbuffer. LSB apply to first lane
xoffsets_hi	4b offset for each lane, applied to the xbuffer. LSB apply to 16th lane
xsquare	Select order of the mini-permute square (default=0x3210). LSB apply to first element. Value per lane needs to be less than 4. max value for this field is (0x3333)
ystart	Starting position offset applied to all lanes of input from xbuffer for the second input
yoffsets	4b offset for each lane, applied to the xbuffer. LSB apply to first lane
yoffsets_hi	4b offset for each lane, applied to the xbuffer. LSB apply to 16th lane
ysquare	Select order of the mini-permute square (default=0x3210). LSB apply to first element. Value per lane needs be less than 4. max value for this field is (0x3333)

Note

• This intrinsic uses the 'square' parameter, to have more information on how to use this please go here
• For more information on how the function f() selects data from the buffers go here.
• Data from xbuff using 'xstart','xoffsets(_hi)' params is the left operand and data from xbuff using 'ystart','yoffsets(_hi)' params is the right operand.

maxdiffcmp16() [1/3]

v16int32 maxdiffcmp16	(	v16int32	xbuff,
		int	xstart,
		unsigned int	xoffsets,
		unsigned int	xoffsets_hi,
		int	ystart,
		unsigned int	yoffsets,
		unsigned int	yoffsets_hi,
		unsigned int &	cmp
	)

Performs a maximum difference computation between lanes of xbuff.

for (int i = 0; i < 16; i++)
    idx = f( xstart, xoffsets[i]);
    idy = f( ystart, yoffsets[i]);
    o[i] = maxdiff(x[idx], x[idy])
 
xoffsets, xoffsets_hi, yoffsets, yoffsets_hi have 8 offset values each. 4 bits per offset.
For Example: for v16int32 output type, idx for output_lane_0 = f(xstart,xoffsets[0])
For Example: for v16int32 output type, idx for output_lane_15 = f(xstart,xoffsets_hi[7])
In case of v32int16, 1 offset is used for 2 adjacent lanes.
For more information on how the function f() selects data from the buffers refer to Lane selection note below.
    cmp[i-th bit] = ( x[idx] > x[idy] ? 1 : 0 );

Returns

Value of each lane is the result of a maximum difference computation between lanes of xbuff where the result of lane 0 goes to lane 0 of the output.

Parameters

xbuff	Input buffer of 16 elements with 32-bit precision
xstart	Starting position offset applied to all lanes of input from X buffer
xoffsets	4b offset for each lane, applied to the xbuffer. LSB apply to first lane
xoffsets_hi	4b offset for each lane, applied to the xbuffer. LSB apply to 8th lane
ystart	Starting position offset applied to all lanes of input from xbuffer for the second input
yoffsets	4b offset for each lane, applied to the xbuffer. LSB apply to first lane
yoffsets_hi	4b offset for each lane, applied to the xbuffer. LSB apply to 8th lane
cmp	32bit value where each bit is the results of the comparison lane by lane, referred to the output lanes

Note

• For more information on how the function f() selects data from the buffers go here.

maxdiffcmp16() [2/3]

v16int32 maxdiffcmp16	(	v16int32	xbuff,
		int	xstart,
		unsigned int	xoffsets,
		unsigned int	xoffsets_hi,
		v16int32	ybuff,
		int	ystart,
		unsigned int	yoffsets,
		unsigned int	yoffsets_hi,
		unsigned int &	cmp
	)

Performs a maximum difference computation between lanes of xbuff and ybuff.

for (int i = 0; i < 16; i++)
    idx = f( xstart, xoffsets[i]);
    idy = f( ystart, yoffsets[i]);
    o[i] = maxdiff(x[idx], y[idy])
 
xoffsets, xoffsets_hi, yoffsets, yoffsets_hi have 8 offset values each. 4 bits per offset.
For Example: for v16int32 output type, idx for output_lane_0 = f(xstart,xoffsets[0])
For Example: for v16int32 output type, idx for output_lane_15 = f(xstart,xoffsets_hi[7])
In case of v32int16, 1 offset is used for 2 adjacent lanes.
For more information on how the function f() selects data from the buffers refer to Lane selection note below.
    cmp[i-th bit] = ( x[idx] > y[idy] ? 1 : 0 );

Returns

Value of each lane is the result of a maximum difference computation between lanes of xbuff where the result of lane 0 goes to lane 0 of the output.

Parameters

xbuff	Input buffer of 16 elements with 32-bit precision
xstart	Starting position offset applied to all lanes of input from X buffer
xoffsets	4b offset for each lane, applied to the xbuffer. LSB apply to first lane
xoffsets_hi	4b offset for each lane, applied to the xbuffer. LSB apply to 8th lane
ybuff	Input buffer of 16 elements with 32-bit precision
ystart	Starting position offset applied to all lanes of input from ybuffer for the second input
yoffsets	4b offset for each lane, applied to the ybuffer. LSB apply to first lane
yoffsets_hi	4b offset for each lane, applied to the ybuffer. LSB apply to 8th lane
cmp	32bit value where each bit is the results of the comparison lane by lane, referred to the output lanes

Note

• For more information on how the function f() selects data from the buffers go here.

maxdiffcmp16() [3/3]

v16int32 maxdiffcmp16	(	v32int32	xbuff,
		int	xstart,
		unsigned int	xoffsets,
		unsigned int	xoffsets_hi,
		int	ystart,
		unsigned int	yoffsets,
		unsigned int	yoffsets_hi,
		unsigned int &	cmp
	)

Performs a maximum difference computation between lanes of xbuff.

for (int i = 0; i < 16; i++)
    idx = f( xstart, xoffsets[i]);
    idy = f( ystart, yoffsets[i]);
    o[i] = maxdiff(x[idx], x[idy])
 
xoffsets, xoffsets_hi, yoffsets, yoffsets_hi have 8 offset values each. 4 bits per offset.
For Example: for v16int32 output type, idx for output_lane_0 = f(xstart,xoffsets[0])
For Example: for v16int32 output type, idx for output_lane_15 = f(xstart,xoffsets_hi[7])
In case of v32int16, 1 offset is used for 2 adjacent lanes.
For more information on how the function f() selects data from the buffers refer to Lane selection note below.
    cmp[i-th bit] = ( x[idx] > x[idy] ? 1 : 0 );

Returns

Value of each lane is the result of a maximum difference computation between lanes of xbuff where the result of lane 0 goes to lane 0 of the output.

Parameters

xbuff	Input buffer of 32 elements with 32-bit precision
xstart	Starting position offset applied to all lanes of input from X buffer
xoffsets	4b offset for each lane, applied to the xbuffer. LSB apply to first lane
xoffsets_hi	4b offset for each lane, applied to the xbuffer. LSB apply to 8th lane
ystart	Starting position offset applied to all lanes of input from xbuffer for the second input
yoffsets	4b offset for each lane, applied to the xbuffer. LSB apply to first lane
yoffsets_hi	4b offset for each lane, applied to the xbuffer. LSB apply to 8th lane
cmp	32bit value where each bit is the results of the comparison lane by lane, referred to the output lanes

Note

• For more information on how the function f() selects data from the buffers go here.

maxdiffcmp32() [1/3]

v32int16 maxdiffcmp32	(	v32int16	xbuff,
		int	xstart,
		unsigned int	xoffsets,
		unsigned int	xoffsets_hi,
		unsigned int	xsquare,
		int	ystart,
		unsigned int	yoffsets,
		unsigned int	yoffsets_hi,
		unsigned int	ysquare,
		unsigned int &	cmp
	)

Performs a maximum difference computation between lanes of xbuff.

for (int i = 0; i < 32; i++)
    idx = f( xstart, xoffsets[i],xsquare);
    idy = f( ystart, yoffsets[i],ysquare);
    o[i] = maxdiff(x[idx], x[idy])
 
xoffsets, xoffsets_hi, yoffsets, yoffsets_hi have 8 offset values each. 4 bits per offset.
For Example: for v16int32 output type, idx for output_lane_0 = f(xstart,xoffsets[0],xsquare)
For Example: for v16int32 output type, idx for output_lane_15 = f(xstart,xoffsets_hi[7],xsquare)
In case of v32int16, 1 offset is used for 2 adjacent lanes.
For more information on how the function f() selects data from the buffers refer to Lane selection note below.
    cmp[i-th bit] = ( x[idx] > x[idy] ? 1 : 0 );

Returns

Value of each lane is the result of a maximum difference computation between lanes of xbuff where the result of lane 0 goes to lane 0 of the output.

Parameters

xbuff	Input buffer of 32 elements with 16-bit precision
xstart	Starting position offset applied to all lanes of input from X buffer
xoffsets	4b offset for each lane, applied to the xbuffer. LSB apply to first lane
xoffsets_hi	4b offset for each lane, applied to the xbuffer. LSB apply to 16th lane
xsquare	Select order of the mini-permute square (default=0x3210). LSB apply to first element. Value per lane needs to be less than 4. max value for this field is (0x3333)
ystart	Starting position offset applied to all lanes of input from xbuffer for the second input
yoffsets	4b offset for each lane, applied to the xbuffer. LSB apply to first lane
yoffsets_hi	4b offset for each lane, applied to the xbuffer. LSB apply to 16th lane
ysquare	Select order of the mini-permute square (default=0x3210). LSB apply to first element. Value per lane needs be less than 4. max value for this field is (0x3333)
cmp	32bit value where each bit is the results of the comparison lane by lane, referred to the output lanes

Note

• This intrinsic uses the 'square' parameter, to have more information on how to use this please go here
• For more information on how the function f() selects data from the buffers go here.

maxdiffcmp32() [2/3]

v32int16 maxdiffcmp32	(	v32int16	xbuff,
		int	xstart,
		unsigned int	xoffsets,
		unsigned int	xoffsets_hi,
		unsigned int	xsquare,
		v32int16	ybuff,
		int	ystart,
		unsigned int	yoffsets,
		unsigned int	yoffsets_hi,
		unsigned int	ysquare,
		unsigned int &	cmp
	)

Performs a maximum difference computation between lanes of xbuff and ybuff.

for (int i = 0; i < 32; i++)
    idx = f( xstart, xoffsets[i],xsquare);
    idy = f( ystart, yoffsets[i],ysquare);
    o[i] = maxdiff(x[idx], y[idy])
 
xoffsets, xoffsets_hi, yoffsets, yoffsets_hi have 8 offset values each. 4 bits per offset.
For Example: for v16int32 output type, idx for output_lane_0 = f(xstart,xoffsets[0],xsquare)
For Example: for v16int32 output type, idx for output_lane_15 = f(xstart,xoffsets_hi[7],xsquare)
In case of v32int16, 1 offset is used for 2 adjacent lanes.
For more information on how the function f() selects data from the buffers refer to Lane selection note below.
    cmp[i-th bit] = ( x[idx] > y[idy] ? 1 : 0 );

Returns

Value of each lane is the result of a maximum difference computation between lanes of xbuff where the result of lane 0 goes to lane 0 of the output.

Parameters

xbuff	Input buffer of 32 elements with 16-bit precision
xstart	Starting position offset applied to all lanes of input from X buffer
xoffsets	4b offset for each lane, applied to the xbuffer. LSB apply to first lane
xoffsets_hi	4b offset for each lane, applied to the xbuffer. LSB apply to 16th lane
xsquare	Select order of the mini-permute square (default=0x3210). LSB apply to first element. Value per lane needs to be less than 4. max value for this field is (0x3333)
ybuff	Input buffer of 32 elements with 16-bit precision
ystart	Starting position offset applied to all lanes of input from ybuffer for the second input
yoffsets	4b offset for each lane, applied to the ybuffer. LSB apply to first lane
yoffsets_hi	4b offset for each lane, applied to the ybuffer. LSB apply to 16th lane
ysquare	Select order of the mini-permute square (default=0x3210). LSB apply to first element. Value per lane needs be less than 4. max value for this field is (0x3333)
cmp	32bit value where each bit is the results of the comparison lane by lane, referred to the output lanes

Note

• This intrinsic uses the 'square' parameter, to have more information on how to use this please go here
• For more information on how the function f() selects data from the buffers go here.

maxdiffcmp32() [3/3]

v32int16 maxdiffcmp32	(	v64int16	xbuff,
		int	xstart,
		unsigned int	xoffsets,
		unsigned int	xoffsets_hi,
		unsigned int	xsquare,
		int	ystart,
		unsigned int	yoffsets,
		unsigned int	yoffsets_hi,
		unsigned int	ysquare,
		unsigned int &	cmp
	)

Performs a maximum difference computation between lanes of xbuff.

for (int i = 0; i < 32; i++)
    idx = f( xstart, xoffsets[i],xsquare);
    idy = f( ystart, yoffsets[i],ysquare);
    o[i] = maxdiff(x[idx], x[idy])
 
xoffsets, xoffsets_hi, yoffsets, yoffsets_hi have 8 offset values each. 4 bits per offset.
For Example: for v16int32 output type, idx for output_lane_0 = f(xstart,xoffsets[0],xsquare)
For Example: for v16int32 output type, idx for output_lane_15 = f(xstart,xoffsets_hi[7],xsquare)
In case of v32int16, 1 offset is used for 2 adjacent lanes.
For more information on how the function f() selects data from the buffers refer to Lane selection note below.
    cmp[i-th bit] = ( x[idx] > x[idy] ? 1 : 0 );

Returns

Value of each lane is the result of a maximum difference computation between lanes of xbuff where the result of lane 0 goes to lane 0 of the output.

Parameters

xbuff	Input buffer of 64 elements with 16-bit precision
xstart	Starting position offset applied to all lanes of input from X buffer
xoffsets	4b offset for each lane, applied to the xbuffer. LSB apply to first lane
xoffsets_hi	4b offset for each lane, applied to the xbuffer. LSB apply to 16th lane
xsquare	Select order of the mini-permute square (default=0x3210). LSB apply to first element. Value per lane needs to be less than 4. max value for this field is (0x3333)
ystart	Starting position offset applied to all lanes of input from xbuffer for the second input
yoffsets	4b offset for each lane, applied to the xbuffer. LSB apply to first lane
yoffsets_hi	4b offset for each lane, applied to the xbuffer. LSB apply to 16th lane
ysquare	Select order of the mini-permute square (default=0x3210). LSB apply to first element. Value per lane needs be less than 4. max value for this field is (0x3333)
cmp	32bit value where each bit is the results of the comparison lane by lane, referred to the output lanes

Note

• This intrinsic uses the 'square' parameter, to have more information on how to use this please go here
• For more information on how the function f() selects data from the buffers go here.