background image

CVTPD2PI—Convert Packed Double-Precision FP Values to Packed Dword Integers

INSTRUCTION SET REFERENCE, A-L

3-238 Vol. 2A

CVTPD2PI—Convert Packed Double-Precision FP Values to Packed Dword Integers

Instruction Operand Encoding

Description

Converts two packed double-precision floating-point values in the source operand (second operand) to two packed 
signed doubleword integers in the destination operand (first operand).
The source operand can be an XMM register or a 128-bit memory location. The destination operand is an MMX tech-
nology register. 
When a conversion is inexact, the value returned is rounded according to the rounding control bits in the MXCSR 
register. If a converted result is larger than the maximum signed doubleword integer, the floating-point invalid 
exception is raised, and if this exception is masked, the indefinite integer value (80000000H) is returned.
This instruction causes a transition from x87 FPU to MMX technology operation (that is, the x87 FPU top-of-stack 
pointer is set to 0 and the x87 FPU tag word is set to all 0s [valid]). If this instruction is executed while an x87 FPU 
floating-point exception is pending, the exception is handled before the CVTPD2PI instruction is executed.
In 64-bit mode, use of the REX.R prefix permits this instruction to access additional registers (XMM8-XMM15).

Operation

DEST[31:0] ← Convert_Double_Precision_Floating_Point_To_Integer32(SRC[63:0]);

DEST[63:32] ← Convert_Double_Precision_Floating_Point_To_Integer32(SRC[127:64]);

Intel C/C++ Compiler Intrinsic Equivalent

CVTPD1PI:

__m64 _mm_cvtpd_pi32(__m128d a)

SIMD Floating-Point Exceptions

Invalid, Precision.

Other Exceptions

See Table 22-4, “Exception Conditions for Legacy SIMD/MMX Instructions with FP Exception and 16-Byte Align-

ment,” in the Intel® 64 and IA-32 Architectures Software Developer’s Manual, Volume 3B.

Opcode/

Instruction

Op/ 

En

64-Bit 

Mode

Compat/

Leg Mode

Description

66 0F 2D /r
CVTPD2PI mmxmm/m128

RM

Valid

Valid

Convert two packed double-precision floating-

point values from xmm/m128 to two packed 

signed doubleword integers in mm.

Op/En

Operand 1

Operand 2

Operand 3

Operand 4

RM

ModRM:reg (w)

ModRM:r/m (r)

NA

NA