Arm® A32/T32 Instruction Set Architecture: Armv8, for Armv8-A architecture profile Documentation

VUMMLA

The widening integer matrix multiply-accumulate instruction multiplies the 2x8 matrix of unsigned 8-bit integer values held in the first source vector by the 8x2 matrix of unsigned 8-bit integer values in the second source vector. The resulting 2x2 32-bit integer matrix product is destructively added to the 32-bit integer matrix accumulator held in the destination vector. This is equivalent to performing an 8-way dot product per destination element.

From Armv8.2, this is an optional instruction. ID_ISAR6.I8MM indicates whether this instruction is supported in the T32 and A32 instruction sets.

It has encodings from the following instruction sets: A32 ( A1 ) and T32 ( T1 ) .

A1
(Armv8.6)

if !HaveAArch32Int8MatMulExt() then UNDEFINED;
case B:U of
    when '00' op1_unsigned = FALSE; op2_unsigned = FALSE;
    when '01' op1_unsigned = TRUE;  op2_unsigned = TRUE;
    when '10' op1_unsigned = TRUE;  op2_unsigned = FALSE;
    when '11' UNDEFINED;
if Vd<0> == '1' || Vn<0> == '1' || Vm<0> == '1' then UNDEFINED;
integer d = UInt(D:Vd);
integer n = UInt(N:Vn);
integer m = UInt(M:Vm);

T1
(Armv8.6)

if InITBlock() then UNPREDICTABLE;
if !HaveAArch32Int8MatMulExt() then UNDEFINED;
case B:U of
    when '00' op1_unsigned = FALSE; op2_unsigned = FALSE;
    when '01' op1_unsigned = TRUE;  op2_unsigned = TRUE;
    when '10' op1_unsigned = TRUE;  op2_unsigned = FALSE;
    when '11' UNDEFINED;
if Vd<0> == '1' || Vn<0> == '1' || Vm<0> == '1' then UNDEFINED;
integer d = UInt(D:Vd);
integer n = UInt(N:Vn);
integer m = UInt(M:Vm);

Assembler Symbols

CheckAdvSIMDEnabled();
bits(128) operand1 = Q[n>>1];
bits(128) operand2 = Q[m>>1];
bits(128) addend   = Q[d>>1];

Q[d>>1] = MatMulAdd(addend, operand1, operand2, op1_unsigned, op2_unsigned);