{ X86::VSQRTPSYr_Int, X86::VSQRTPSYm_Int, TB_ALIGN_32 },
{ X86::VBROADCASTSSYrr, X86::VBROADCASTSSYrm, TB_NO_REVERSE },
{ X86::VBROADCASTSDYrr, X86::VBROADCASTSDYrm, TB_NO_REVERSE },
+
+ // BMI/BMI2 foldable instructions
+ { X86::RORX32ri, X86::RORX32mi, 0 },
+ { X86::RORX64ri, X86::RORX64mi, 0 },
};
for (unsigned i = 0, e = array_lengthof(OpTbl1); i != e; ++i) {
} // Defs = [EFLAGS]
+def ROT32L2R_imm8 : SDNodeXForm<imm, [{
+ // Convert a ROTL shamt to a ROTR shamt on 32-bit integer.
+ return getI8Imm(32 - N->getZExtValue());
+}]>;
+
+def ROT64L2R_imm8 : SDNodeXForm<imm, [{
+ // Convert a ROTL shamt to a ROTR shamt on 64-bit integer.
+ return getI8Imm(64 - N->getZExtValue());
+}]>;
+
multiclass bmi_rotate<string asm, RegisterClass RC, X86MemOperand x86memop> {
let neverHasSideEffects = 1 in {
def ri : Ii8<0xF0, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, i8imm:$src2),
defm SHRX64 : bmi_shift<"shrx{q}", GR64, i64mem>, T8XD, VEX_W;
defm SHLX32 : bmi_shift<"shlx{l}", GR32, i32mem>, T8, OpSize;
defm SHLX64 : bmi_shift<"shlx{q}", GR64, i64mem>, T8, OpSize, VEX_W;
+
+ // Prefer RORX which is non-destructive and doesn't update EFLAGS.
+ let AddedComplexity = 10 in {
+ def : Pat<(rotl GR32:$src, (i8 imm:$shamt)),
+ (RORX32ri GR32:$src, (ROT32L2R_imm8 imm:$shamt))>;
+ def : Pat<(rotl GR64:$src, (i8 imm:$shamt)),
+ (RORX64ri GR64:$src, (ROT64L2R_imm8 imm:$shamt))>;
+ }
+
+ def : Pat<(rotl (loadi32 addr:$src), (i8 imm:$shamt)),
+ (RORX32mi addr:$src, (ROT32L2R_imm8 imm:$shamt))>;
+ def : Pat<(rotl (loadi64 addr:$src), (i8 imm:$shamt)),
+ (RORX64mi addr:$src, (ROT64L2R_imm8 imm:$shamt))>;
}
-; RUN: llc -mtriple=i386-apple-darwin -o - < %s | FileCheck %s
+; RUN: llc -mtriple=i386-apple-darwin -mcpu=corei7 -o - < %s | FileCheck %s
define i32 @func(i8* %A) nounwind readnone {
entry:
-; RUN: llc < %s -march=x86 | FileCheck %s
+; RUN: llc < %s -march=x86 -mcpu=corei7 | FileCheck %s
+; RUN: llc < %s -march=x86 -mcpu=core-avx2 | FileCheck %s --check-prefix=BMI2
define i32 @foo(i32 %x, i32 %y, i32 %z) nounwind readnone {
entry:
entry:
; CHECK: xfoo:
; CHECK: roll $7
+; BMI2: xfoo:
+; BMI2: rorxl $25
%0 = lshr i32 %x, 25
%1 = shl i32 %x, 7
%2 = or i32 %0, %1
ret i32 %2
}
+define i32 @xfoop(i32* %p) nounwind readnone {
+entry:
+; BMI2: xfoop:
+; BMI2: rorxl $25, ({{.+}}), %{{.+}}
+ %x = load i32* %p
+ %a = lshr i32 %x, 25
+ %b = shl i32 %x, 7
+ %c = or i32 %a, %b
+ ret i32 %c
+}
+
define i32 @xbar(i32 %x, i32 %y, i32 %z) nounwind readnone {
entry:
; CHECK: xbar:
entry:
; CHECK: xun:
; CHECK: roll $25
+; BMI2: xun:
+; BMI2: rorxl $7
%0 = lshr i32 %x, 7
%1 = shl i32 %x, 25
%2 = or i32 %0, %1
ret i32 %2
}
+define i32 @xunp(i32* %p) nounwind readnone {
+entry:
+; BMI2: xunp:
+; BMI2: rorxl $7, ({{.+}}), %{{.+}}
+ %x = load i32* %p
+ %a = lshr i32 %x, 7
+ %b = shl i32 %x, 25
+ %c = or i32 %a, %b
+ ret i32 %c
+}
+
define i32 @xbu(i32 %x, i32 %y, i32 %z) nounwind readnone {
entry:
; CHECK: xbu:
-; RUN: llc < %s -march=x86-64 > %t
-; RUN: grep rol %t | count 3
+; RUN: llc < %s -march=x86-64 -mcpu=corei7 > %t
+; RUN: grep rol %t | count 5
; RUN: grep ror %t | count 1
; RUN: grep shld %t | count 2
; RUN: grep shrd %t | count 2
+; RUN: llc < %s -march=x86-64 -mcpu=core-avx2 | FileCheck %s --check-prefix=BMI2
define i64 @foo(i64 %x, i64 %y, i64 %z) nounwind readnone {
entry:
define i64 @xfoo(i64 %x, i64 %y, i64 %z) nounwind readnone {
entry:
+; BMI2: xfoo:
+; BMI2: rorxq $57
%0 = lshr i64 %x, 57
%1 = shl i64 %x, 7
%2 = or i64 %0, %1
ret i64 %2
}
+define i64 @xfoop(i64* %p) nounwind readnone {
+entry:
+; BMI2: xfoop:
+; BMI2: rorxq $57, ({{.+}}), %{{.+}}
+ %x = load i64* %p
+ %a = lshr i64 %x, 57
+ %b = shl i64 %x, 7
+ %c = or i64 %a, %b
+ ret i64 %c
+}
+
define i64 @xbar(i64 %x, i64 %y, i64 %z) nounwind readnone {
entry:
%0 = shl i64 %y, 7
define i64 @xun(i64 %x, i64 %y, i64 %z) nounwind readnone {
entry:
+; BMI2: xun:
+; BMI2: rorxq $7
%0 = lshr i64 %x, 7
%1 = shl i64 %x, 57
%2 = or i64 %0, %1
ret i64 %2
}
+define i64 @xunp(i64* %p) nounwind readnone {
+entry:
+; BMI2: xunp:
+; BMI2: rorxq $7, ({{.+}}), %{{.+}}
+ %x = load i64* %p
+ %a = lshr i64 %x, 7
+ %b = shl i64 %x, 57
+ %c = or i64 %a, %b
+ ret i64 %c
+}
+
define i64 @xbu(i64 %x, i64 %y, i64 %z) nounwind readnone {
entry:
%0 = lshr i64 %y, 7
-; RUN: llc < %s -march=x86-64 | grep rol | count 2
+; RUN: llc < %s -march=x86-64 -mcpu=corei7 | grep rol | count 2
define i64 @test1(i64 %x) nounwind {
entry: