Convert _mm_slli_epi{16,32,64} & _mm_srli_epi{16,32,64} to const generics #1020
Conversation
|
r? @Amanieu (rust-highfive has picked a reviewer for you, use r? to override) |
tmiasko
changed the title
tmiasko
changed the title
| } | ||
| constify_imm8!(imm8, call) | ||
| pub unsafe fn _mm_slli_epi16<const imm8: i32>(a: __m128i) -> __m128i { | ||
| transmute(pslliw(a.as_i16x8(), imm8)) |
There was a problem hiding this comment.
This should have a static_assert! to ensure the immediate is between 0 and 255.
There was a problem hiding this comment.
In fact use static_assert_imm8 that was added in #1021.
There was a problem hiding this comment.
Added range checks using static_assert_imm8.
What do you use as a reference point, to make decision whether it should be hard error or not? The fact that it is documented as imm8? clang, gcc, and icc, as far as I can see accept arbitrary values there (in fact run-time ones as well).
There was a problem hiding this comment.
We follow the intel intrinsics data which specifies that the argument is an 8-bit immediate. There is a separate intrinsic (_mm_sll_epi32) which allows run-time values.
<intrinsic tech="SSE2" vexEq="TRUE" name="_mm_slli_epi32">
<type>Integer</type>
<CPUID>SSE2</CPUID>
<category>Shift</category>
<return type="__m128i" varname="dst" etype="UI32"/>
<parameter type="__m128i" varname="a" etype="UI32"/>
<parameter type="int" varname="imm8" etype="IMM" immwidth="8"/>
<description>Shift packed 32-bit integers in "a" left by "imm8" while shifting in zeros, and store the results in "dst".</description>
<operation>
FOR j := 0 to 3
i := j*32
IF imm8[7:0] > 31
dst[i+31:i] := 0
ELSE
dst[i+31:i] := ZeroExtend32(a[i+31:i] << imm8[7:0])
FI
ENDFOR
</operation>
<instruction name="PSLLD" form="xmm, imm8" xed="PSLLD_XMMdq_IMMb"/>
<header>emmintrin.h</header>
</intrinsic>
The presence of an appropriate static_assert! should probably be enforced by stdarch-verify, I'll look into implementing that.
_mm_slli_epi64_mm_slli_epi32_mm_slli_epi16_mm_srli_epi64_mm_srli_epi32_mm_srli_epi16