RPCSX/llvm
1
0
Fork 0
mirror of https://github.com/RPCSX/llvm.git synced 2024-11-24 20:29:53 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

[PATCH] PowerPC: Expand load extend vector operations

Browse Source 
This patch expands the SEXTLOAD, ZEXTLOAD, and EXTLOAD operations for
vector types when altivec is enabled.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@167386 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Adhemerval Zanella 2012-11-05 17:15:56 +00:00
parent 08f9c8d776
commit cfe09ed28d
2 changed files with 165 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
lib/Target/PowerPC/PPCISelLowering.cpp
View File

@ -361,6 +361,16 @@ PPCTargetLowering::PPCTargetLowering(PPCTargetMachine &TM)
setOperationAction(ISD::CTLZ_ZERO_UNDEF, VT, Expand);
setOperationAction(ISD::CTTZ, VT, Expand);
setOperationAction(ISD::CTTZ_ZERO_UNDEF, VT, Expand);
setOperationAction(ISD::SIGN_EXTEND_INREG, VT, Expand);
for (unsigned j = (unsigned)MVT::FIRST_VECTOR_VALUETYPE;
j <= (unsigned)MVT::LAST_VECTOR_VALUETYPE; ++j) {
MVT::SimpleValueType InnerVT = (MVT::SimpleValueType)j;
setTruncStoreAction(VT, InnerVT, Expand);
}
setLoadExtAction(ISD::SEXTLOAD, VT, Expand);
setLoadExtAction(ISD::ZEXTLOAD, VT, Expand);
setLoadExtAction(ISD::EXTLOAD, VT, Expand);
}
for (unsigned i = (unsigned)MVT::FIRST_FP_VECTOR_VALUETYPE;

155
test/CodeGen/PowerPC/vec_extload.ll Normal file
View File

@ -0,0 +1,155 @@
; RUN: llc -mcpu=pwr6 -mattr=+altivec < %s | FileCheck %s
; Check vector extend load expansion with altivec enabled.
target datalayout = "E-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v128:128:128-n32:64"
target triple = "powerpc64-unknown-linux-gnu"
; Altivec does not provides an sext intruction, so it expands
; a set of vector stores (stvx), bytes load/sign expand/store
; (lbz/stb), and a final vector load (lvx) to load the result
; extended vector.
define <16 x i8> @v16si8_sext_in_reg(<16 x i8> %a) {
%b = trunc <16 x i8> %a to <16 x i4>
%c = sext <16 x i4> %b to <16 x i8>
ret <16 x i8> %c
}
; CHECK: v16si8_sext_in_reg:
; CHECK: stvx 2, {{[0-9]+}}, {{[0-9]+}}
; CHECK: lbz
; CHECK: stb
; CHECK: stvx 2, {{[0-9]+}}, {{[0-9]+}}
; CHECK: lbz
; CHECK: stb
; CHECK: stvx 2, {{[0-9]+}}, {{[0-9]+}}
; CHECK: lbz
; CHECK: stb
; CHECK: stvx 2, {{[0-9]+}}, {{[0-9]+}}
; CHECK: lbz
; CHECK: stb
; CHECK: stvx 2, {{[0-9]+}}, {{[0-9]+}}
; CHECK: lbz
; CHECK: stb
; CHECK: stvx 2, {{[0-9]+}}, {{[0-9]+}}
; CHECK: lbz
; CHECK: stb
; CHECK: stvx 2, {{[0-9]+}}, {{[0-9]+}}
; CHECK: lbz
; CHECK: stb
; CHECK: stvx 2, {{[0-9]+}}, {{[0-9]+}}
; CHECK: lbz
; CHECK: stb
; CHECK: stvx 2, {{[0-9]+}}, {{[0-9]+}}
; CHECK: lbz
; CHECK: stb
; CHECK: stvx 2, {{[0-9]+}}, {{[0-9]+}}
; CHECK: lbz
; CHECK: stb
; CHECK: stvx 2, {{[0-9]+}}, {{[0-9]+}}
; CHECK: lbz
; CHECK: stb
; CHECK: stvx 2, {{[0-9]+}}, {{[0-9]+}}
; CHECK: lbz
; CHECK: stb
; CHECK: stvx 2, {{[0-9]+}}, {{[0-9]+}}
; CHECK: lbz
; CHECK: stb
; CHECK: stvx 2, {{[0-9]+}}, {{[0-9]+}}
; CHECK: lbz
; CHECK: stb
; CHECK: stvx 2, {{[0-9]+}}, {{[0-9]+}}
; CHECK: lbz
; CHECK: stb
; CHECK: stvx 2, {{[0-9]+}}, {{[0-9]+}}
; CHECK: lbz
; CHECK: stb
; CHECK: lvx 2, {{[0-9]+}}, {{[0-9]+}}
; The zero extend uses a more clever logic: a vector splat
; and a logic and to set higher bits to 0.
define <16 x i8> @v16si8_zext_in_reg(<16 x i8> %a) {
%b = trunc <16 x i8> %a to <16 x i4>
%c = zext <16 x i4> %b to <16 x i8>
ret <16 x i8> %c
}
; CHECK: v16si8_zext_in_reg:
; CHECK: vspltisb [[VMASK:[0-9]+]], 15
; CHECK-NEXT: vand 2, 2, [[VMASK]]
; Same as v16si8_sext_in_reg, expands to load/store halfwords (lhz/sth).
define <8 x i16> @v8si16_sext_in_reg(<8 x i16> %a) {
%b = trunc <8 x i16> %a to <8 x i8>
%c = sext <8 x i8> %b to <8 x i16>
ret <8 x i16> %c
}
; CHECK: v8si16_sext_in_reg:
; CHECK: stvx 2, {{[0-9]+}}, {{[0-9]+}}
; CHECK: lhz
; CHECK: sth
; CHECK: stvx 2, {{[0-9]+}}, {{[0-9]+}}
; CHECK: lhz
; CHECK: sth
; CHECK: stvx 2, {{[0-9]+}}, {{[0-9]+}}
; CHECK: lhz
; CHECK: sth
; CHECK: stvx 2, {{[0-9]+}}, {{[0-9]+}}
; CHECK: lhz
; CHECK: sth
; CHECK: stvx 2, {{[0-9]+}}, {{[0-9]+}}
; CHECK: lhz
; CHECK: sth
; CHECK: stvx 2, {{[0-9]+}}, {{[0-9]+}}
; CHECK: lhz
; CHECK: sth
; CHECK: stvx 2, {{[0-9]+}}, {{[0-9]+}}
; CHECK: lhz
; CHECK: sth
; CHECK: stvx 2, {{[0-9]+}}, {{[0-9]+}}
; CHECK: lhz
; CHECK: sth
; CHECK: lvx 2, {{[0-9]+}}, {{[0-9]+}}
; Same as v8si16_sext_in_reg, but instead of creating the mask
; with a splat, loads it from memory.
define <8 x i16> @v8si16_zext_in_reg(<8 x i16> %a) {
%b = trunc <8 x i16> %a to <8 x i8>
%c = zext <8 x i8> %b to <8 x i16>
ret <8 x i16> %c
}
; CHECK: v8si16_zext_in_reg:
; CHECK: ld [[RMASKTOC:[0-9]+]], .LC{{[0-9]+}}@toc(2)
; CHECK-NEXT: lvx [[VMASK:[0-9]+]], {{[0-9]+}}, [[RMASKTOC]]
; CHECK-NEXT: vand 2, 2, [[VMASK]]
; Same as v16si8_sext_in_reg, expands to load halfword (lha) and
; store words (stw).
define <4 x i32> @v4si32_sext_in_reg(<4 x i32> %a) {
%b = trunc <4 x i32> %a to <4 x i16>
%c = sext <4 x i16> %b to <4 x i32>
ret <4 x i32> %c
}
; CHECK: v4si32_sext_in_reg:
; CHECK: stvx 2, {{[0-9]+}}, {{[0-9]+}}
; CHECK: lha
; CHECK: stw
; CHECK: stvx 2, {{[0-9]+}}, {{[0-9]+}}
; CHECK: lha
; CHECK: stw
; CHECK: stvx 2, {{[0-9]+}}, {{[0-9]+}}
; CHECK: lha
; CHECK: stw
; CHECK: stvx 2, {{[0-9]+}}, {{[0-9]+}}
; CHECK: lha
; CHECK: stw
; CHECK: lvx 2, {{[0-9]+}}, {{[0-9]+}}
; Same as v8si16_sext_in_reg.
define <4 x i32> @v4si32_zext_in_reg(<4 x i32> %a) {
%b = trunc <4 x i32> %a to <4 x i16>
%c = zext <4 x i16> %b to <4 x i32>
ret <4 x i32> %c
}
; CHECK: v4si32_zext_in_reg:
; CHECK: vspltisw [[VMASK:[0-9]+]], -16
; CHECK-NEXT: vsrw [[VMASK]], [[VMASK]], [[VMASK]]
; CHECK-NEXT: vand 2, 2, [[VMASK]]