2 checking for sse/sse2/sse3 extension support, 4 initialization of the sse/sse2/sse3 extensions – Intel IA-32 User Manual
Page 500
12-2 Vol. 3A
SSE, SSE2 AND SSE3 SYSTEM PROGRAMMING
12.1.2
Checking for SSE/SSE2/SSE3 Extension Support
If the processor attempts to execute an unsupported SSE/SSE2/SSE3 instruction, the processor
will generate an invalid-opcode exception (#UD).
Before an operating system or executive attempts to use SSE/SSE2/SSE3 extensions, it should
check that support is present on the processor. To make this check, execute CPUID with an argu-
ment of 1 in the EAX register. Make sure:
•
CPUID.1:EDX.SSE[bit 25] = 1
•
CPUID.1:EDX.SSE2[bit 26] = 1
•
CPUID.1:ECX.SSE3[bit 0] = 1
12.1.3
Checking for Support for the FXSAVE and FXRSTOR
Instructions
A separate check must be made to insure that the processor supports FXSAVE and FXRSTOR.
To make this check, execute CPUID with an argument of 1 in the EAX register. Make sure:
•
CPUID.1:EDX.FXSR[bit 24] = 1
12.1.4
Initialization of the SSE/SSE2/SSE3 Extensions
The operating system or executive should carry out the following steps to set up
SSE/SSE2/SSE3 extensions for use by application programs:
1.
Set CR4.OSFXSR[bit 9] = 1. Setting this flag assumes that the operating system provides
facilities for saving and restoring SSE/SSE2/SSE3 states using FXSAVE and FXRSTOR
instructions. These instructions are commonly used to save the SSE/SSE2/SSE3 state
during task switches and when invoking the SIMD floating-point exception (#XF) handler
(see Section 12.4, “Saving the SSE/SSE2/SSE3 State on Task or Context Switches,” and
Section 12.1.6, “Providing an Handler for the SIMD Floating-Point Exception (#XF),”
respectively).
If the processor does not support the FXSAVE and FXRSTOR instructions, attempting to
set the OSFXSR flag will cause an exception (#GP) to be generated.
2.
Set CR4.OSXMMEXCPT[bit 10] = 1. Setting this flag assumes that the operating system
provides an SIMD floating-point exception (#XF) handler (see Section 12.1.6, “Providing
an Handler for the SIMD Floating-Point Exception (#XF)”).