VPMOVSSDB

VPMOVSSDB narrows signed doublewords to signed bytes as VPMOVSDB does, 4, 8 or 16 of them from an xmm, ymm or zmm register into the low bytes of an xmm register or straight to 4, 8 or 16 bytes of memory, but it clamps to -127 to 127, a range symmetric about zero: every value of -127 or less gives 81h, and no result is 80h, -128. It belongs to AVX10_V2_AUX of the ACE specification: opcode 41h with the F3 prefix in map 0F38, W0, where VPMOVSDB has 21h.

Why leave out -128#

A byte of -128 has no positive counterpart: negating it gives -128 again, and the range -128 to 127 reaches one step further below zero than above. The ACE specification gives keeping the sign symmetry as the reason for the narrower range.

Its operands#

As in VPMOVSDB, the source register is in ModRM.reg and the destination, register or memory, in ModRM.r/m. EVEX.vvvv must be 1111b, or the instruction raises #UD. A write mask has one bit per byte: the register form merges or zeroes the bytes it leaves out and clears the destination above the written bytes, while the memory form leaves them in memory as they were and takes no zeroing.

In our builds#

Clang 23’s assembler and GNU as 2.44 reject the mnemonic, and ACE’s intrinsic _mm512_cvtssepi32_epi8, one letter away from VPMOVSDB’s _mm512_cvtsepi32_epi8, is missing from Clang 23’s headers. Without it, one VPMAXSD against -127 before VPMOVSDB gives the same bytes. For a loop clamping int32_t values to -127 to 127 into an int8_t array, Clang 23 at -O2 with -march=x86-64-v4 clamps at both ends with VPMAXSD and VPMINSD and then truncates with VPMOVDB; GCC 14 also clamps at both ends and narrows through VPERMT2W and VPMOVWB. This site’s recorded cases include none of VPMOVSSDB.

Forms

Data flow

Where the elements of the sources go, element 0 on the right; results that depend on a control value are drawn for an example. How to read.

Operation

What the forms do, in pseudocode that our tests run against results recorded on real processors. How to read.

for i in range(VL // 32):
    if k1[i]:
        sa = zmm2.sd[i]
        xmm1.b[i] = max(-127, min(127, sa))
    elif zeroing:
        xmm1.b[i] = 0
    # else: xmm1.b[i] keeps its value
for i in range(VL // 32, 16):
    xmm1.b[i] = 0
# symmetric saturation: never -128
# xmm1 is zeroed from bit 128 up

for i in range(VL // 32):
    if k1[i]:
        sa = zmm2.sd[i]
        mem.b[i] = max(-127, min(127, sa))
    # else: mem.b[i] is not written
# symmetric saturation: never -128

Operands

Flags

None: XED records no flag that the forms test or change.

Exceptions

The exception class is XED's name for the set of exception conditions that a form follows. How to read.

Register forms

AVX512-E6NF

Memory forms

AVX512-E6

Attributes

What else XED records about the forms. How to read.

All forms

write masking

Memory forms

memory fault suppression

CPUID and processors

Each form belongs to an XED ISA set. A form can be used when the processor reports every CPUID bit of one of its ISA set's alternatives.

More AVX10_V2_AUX instructions

• VCVTBF42HF8
• VCVTBF62HF8
• VCVTBF82BF4S
• VCVTBF82BF6S
• VCVTBF82PS
• VCVTBIASPS2BF8
• VCVTBIASPS2BF8S
• VCVTBIASPS2HF8
• VCVTBIASPS2HF8S
• VCVTHF62HF8
• VCVTHF82BF4S
• VCVTHF82HF6S
• VCVTHF82PS
• VCVTPS2BF8
• VCVTPS2BF8S
• VCVTPS2HF8
• VCVTPS2HF8S
• VCVTROPS2HF8
• VCVTROPS2HF8S
• VUNPACKB