BF1CVTL, BF2CVTL

Multi-vector convert from 8-bit floating-point to deinterleaved BFloat16

This instruction converts each 8-bit floating-point element of the source vector to BFloat16 while downscaling the value, and places the two-way deinterleaved results in the corresponding 16-bit elements of the destination vectors. BF1CVTL scales the values by 2-UInt(FPMR.LSCALE[5:0]). BF2CVTL scales the values by 2-UInt(FPMR.LSCALE2[5:0]). The 8-bit floating-point encoding format for BF1CVTL is selected by FPMR.F8S1. The 8-bit floating-point encoding format for BF2CVTL is selected by FPMR.F8S2.

This instruction is unpredicated.

Encoding: BF1CVTL

Variants: FEAT_SME2 && FEAT_FP8 (FEAT_SME2 && FEAT_FP8)

31	30	29	28	27	26	25	24	23	22	21	20	19	18	17	16	15	14	13	12	11	10	9	8	7	6	5	4	3	2	1	0
1	1	0	0	0	0	0	1	0	1	1	0	0	1	1	0	1	1	1	0	0	0										1
								opc														Zn					Zd				L

BF1CVTL { <Zd1>.H-<Zd2>.H }, <Zn>.B

Decoding algorithm

if !IsFeatureImplemented(FEAT_SME2) || !IsFeatureImplemented(FEAT_FP8) then
    EndOfDecode(Decode_UNDEF);
constant integer n = UInt(Zn);
constant integer d = UInt(Zd: '0');
constant boolean issrc2 = FALSE;

Encoding: BF2CVTL

Variants: FEAT_SME2 && FEAT_FP8 (FEAT_SME2 && FEAT_FP8)

31	30	29	28	27	26	25	24	23	22	21	20	19	18	17	16	15	14	13	12	11	10	9	8	7	6	5	4	3	2	1	0
1	1	0	0	0	0	0	1	1	1	1	0	0	1	1	0	1	1	1	0	0	0										1
								opc														Zn					Zd				L

BF2CVTL { <Zd1>.H-<Zd2>.H }, <Zn>.B

Decoding algorithm

if !IsFeatureImplemented(FEAT_SME2) || !IsFeatureImplemented(FEAT_FP8) then
    EndOfDecode(Decode_UNDEF);
constant integer n = UInt(Zn);
constant integer d = UInt(Zd: '0');
constant boolean issrc2 = TRUE;

Operation

CheckFPMREnabled();
CheckStreamingSVEEnabled();
constant integer VL = CurrentVL;
constant integer pairs = VL DIV 16;
constant bits(VL) operand = Z[n, VL];
bits(VL) result1;
bits(VL) result2;

for p = 0 to pairs-1
    constant bits(8) element1 = Elem[operand, 2*p + 0, 8];
    constant bits(8) element2 = Elem[operand, 2*p + 1, 8];
    Elem[result1, p, 16] = FP8ConvertBF(element1, issrc2, FPCR, FPMR);
    Elem[result2, p, 16] = FP8ConvertBF(element2, issrc2, FPCR, FPMR);

Z[d+0, VL] = result1;
Z[d+1, VL] = result2;

Explanations

<Zd1>: Is the name of the first scalable vector register of the destination multi-vector group, encoded as "Zd" times 2.
<Zd2>: Is the name of the second scalable vector register of the destination multi-vector group, encoded as "Zd" times 2 plus 1.
<Zn>: Is the name of the source scalable vector register, encoded in the "Zn" field.