1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-22 18:54:02 +01:00
llvm-mirror/test/CodeGen/Thumb2/shift_parts.ll
David Green e547a3cc78 [ARM] Optimise ASRL/LSRL to smaller shifts using demand bits.
The ASRL/LSRL long shifts are generated from 64bit shifts. Once we have
them, it might turn out that enough of the 64bit result was not required
that we can use a smaller shift to perform the same result. As the
smaller shift can in general be folded in more way, such as into add
instructions in one of the test cases here, we can use the demand bit
analysis to prefer the smaller shifts where we can.

Differential Revision: https://reviews.llvm.org/D75371
2020-03-13 10:09:03 +00:00

510 lines
13 KiB
LLVM

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc --verify-machineinstrs -mtriple=thumbv8.1m.main-none-eabi -mattr=+mve %s -o - | FileCheck %s -check-prefix=CHECK --check-prefix=CHECK-MVE
; RUN: llc --verify-machineinstrs -mtriple=thumbv8.1m.main-none-eabi %s -o - | FileCheck %s -check-prefix=CHECK --check-prefix=CHECK-NON-MVE
define i64 @shift_left_reg(i64 %x, i64 %y) {
; CHECK-MVE-LABEL: shift_left_reg:
; CHECK-MVE: @ %bb.0: @ %entry
; CHECK-MVE-NEXT: lsll r0, r1, r2
; CHECK-MVE-NEXT: bx lr
;
; CHECK-NON-MVE-LABEL: shift_left_reg:
; CHECK-NON-MVE: @ %bb.0: @ %entry
; CHECK-NON-MVE-NEXT: rsb.w r3, r2, #32
; CHECK-NON-MVE-NEXT: lsls r1, r2
; CHECK-NON-MVE-NEXT: lsr.w r3, r0, r3
; CHECK-NON-MVE-NEXT: orrs r1, r3
; CHECK-NON-MVE-NEXT: subs.w r3, r2, #32
; CHECK-NON-MVE-NEXT: it pl
; CHECK-NON-MVE-NEXT: lslpl.w r1, r0, r3
; CHECK-NON-MVE-NEXT: lsl.w r0, r0, r2
; CHECK-NON-MVE-NEXT: it pl
; CHECK-NON-MVE-NEXT: movpl r0, #0
; CHECK-NON-MVE-NEXT: bx lr
entry:
%shl = shl i64 %x, %y
ret i64 %shl
}
define i64 @shift_left_imm(i64 %x) {
; CHECK-MVE-LABEL: shift_left_imm:
; CHECK-MVE: @ %bb.0: @ %entry
; CHECK-MVE-NEXT: lsll r0, r1, #3
; CHECK-MVE-NEXT: bx lr
;
; CHECK-NON-MVE-LABEL: shift_left_imm:
; CHECK-NON-MVE: @ %bb.0: @ %entry
; CHECK-NON-MVE-NEXT: lsls r1, r1, #3
; CHECK-NON-MVE-NEXT: orr.w r1, r1, r0, lsr #29
; CHECK-NON-MVE-NEXT: lsls r0, r0, #3
; CHECK-NON-MVE-NEXT: bx lr
entry:
%shl = shl i64 %x, 3
ret i64 %shl
}
define i64 @shift_left_imm_big(i64 %x) {
; CHECK-LABEL: shift_left_imm_big:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: lsls r1, r0, #16
; CHECK-NEXT: movs r0, #0
; CHECK-NEXT: bx lr
entry:
%shl = shl i64 %x, 48
ret i64 %shl
}
define i64 @shift_left_imm_big2(i64 %x) {
; CHECK-LABEL: shift_left_imm_big2:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: mov r1, r0
; CHECK-NEXT: movs r0, #0
; CHECK-NEXT: bx lr
entry:
%shl = shl i64 %x, 32
ret i64 %shl
}
define i64 @shift_left_imm_big3(i64 %x) {
; CHECK-LABEL: shift_left_imm_big3:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: lsls r1, r0, #1
; CHECK-NEXT: movs r0, #0
; CHECK-NEXT: bx lr
entry:
%shl = shl i64 %x, 33
ret i64 %shl
}
define i64 @shift_right_reg(i64 %x, i64 %y) {
; CHECK-MVE-LABEL: shift_right_reg:
; CHECK-MVE: @ %bb.0: @ %entry
; CHECK-MVE-NEXT: rsbs r2, r2, #0
; CHECK-MVE-NEXT: lsll r0, r1, r2
; CHECK-MVE-NEXT: bx lr
;
; CHECK-NON-MVE-LABEL: shift_right_reg:
; CHECK-NON-MVE: @ %bb.0: @ %entry
; CHECK-NON-MVE-NEXT: rsb.w r3, r2, #32
; CHECK-NON-MVE-NEXT: lsrs r0, r2
; CHECK-NON-MVE-NEXT: lsl.w r3, r1, r3
; CHECK-NON-MVE-NEXT: orrs r0, r3
; CHECK-NON-MVE-NEXT: subs.w r3, r2, #32
; CHECK-NON-MVE-NEXT: it pl
; CHECK-NON-MVE-NEXT: lsrpl.w r0, r1, r3
; CHECK-NON-MVE-NEXT: lsr.w r1, r1, r2
; CHECK-NON-MVE-NEXT: it pl
; CHECK-NON-MVE-NEXT: movpl r1, #0
; CHECK-NON-MVE-NEXT: bx lr
entry:
%shr = lshr i64 %x, %y
ret i64 %shr
}
define i64 @shift_right_imm(i64 %x) {
; CHECK-MVE-LABEL: shift_right_imm:
; CHECK-MVE: @ %bb.0: @ %entry
; CHECK-MVE-NEXT: lsrl r0, r1, #3
; CHECK-MVE-NEXT: bx lr
;
; CHECK-NON-MVE-LABEL: shift_right_imm:
; CHECK-NON-MVE: @ %bb.0: @ %entry
; CHECK-NON-MVE-NEXT: lsrs r0, r0, #3
; CHECK-NON-MVE-NEXT: orr.w r0, r0, r1, lsl #29
; CHECK-NON-MVE-NEXT: lsrs r1, r1, #3
; CHECK-NON-MVE-NEXT: bx lr
entry:
%shr = lshr i64 %x, 3
ret i64 %shr
}
define i64 @shift_right_imm_big(i64 %x) {
; CHECK-LABEL: shift_right_imm_big:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: lsrs r0, r1, #16
; CHECK-NEXT: movs r1, #0
; CHECK-NEXT: bx lr
entry:
%shr = lshr i64 %x, 48
ret i64 %shr
}
define i64 @shift_right_imm_big2(i64 %x) {
; CHECK-LABEL: shift_right_imm_big2:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: mov r0, r1
; CHECK-NEXT: movs r1, #0
; CHECK-NEXT: bx lr
entry:
%shr = lshr i64 %x, 32
ret i64 %shr
}
define i64 @shift_right_imm_big3(i64 %x) {
; CHECK-LABEL: shift_right_imm_big3:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: lsrs r0, r1, #1
; CHECK-NEXT: movs r1, #0
; CHECK-NEXT: bx lr
entry:
%shr = lshr i64 %x, 33
ret i64 %shr
}
define i64 @shift_arithmetic_right_reg(i64 %x, i64 %y) {
; CHECK-MVE-LABEL: shift_arithmetic_right_reg:
; CHECK-MVE: @ %bb.0: @ %entry
; CHECK-MVE-NEXT: asrl r0, r1, r2
; CHECK-MVE-NEXT: bx lr
;
; CHECK-NON-MVE-LABEL: shift_arithmetic_right_reg:
; CHECK-NON-MVE: @ %bb.0: @ %entry
; CHECK-NON-MVE-NEXT: rsb.w r3, r2, #32
; CHECK-NON-MVE-NEXT: lsrs r0, r2
; CHECK-NON-MVE-NEXT: lsl.w r3, r1, r3
; CHECK-NON-MVE-NEXT: orrs r0, r3
; CHECK-NON-MVE-NEXT: subs.w r3, r2, #32
; CHECK-NON-MVE-NEXT: asr.w r2, r1, r2
; CHECK-NON-MVE-NEXT: it pl
; CHECK-NON-MVE-NEXT: asrpl.w r0, r1, r3
; CHECK-NON-MVE-NEXT: it pl
; CHECK-NON-MVE-NEXT: asrpl r2, r1, #31
; CHECK-NON-MVE-NEXT: mov r1, r2
; CHECK-NON-MVE-NEXT: bx lr
entry:
%shr = ashr i64 %x, %y
ret i64 %shr
}
define i64 @shift_arithmetic_right_imm(i64 %x) {
; CHECK-MVE-LABEL: shift_arithmetic_right_imm:
; CHECK-MVE: @ %bb.0: @ %entry
; CHECK-MVE-NEXT: asrl r0, r1, #3
; CHECK-MVE-NEXT: bx lr
;
; CHECK-NON-MVE-LABEL: shift_arithmetic_right_imm:
; CHECK-NON-MVE: @ %bb.0: @ %entry
; CHECK-NON-MVE-NEXT: lsrs r0, r0, #3
; CHECK-NON-MVE-NEXT: orr.w r0, r0, r1, lsl #29
; CHECK-NON-MVE-NEXT: asrs r1, r1, #3
; CHECK-NON-MVE-NEXT: bx lr
entry:
%shr = ashr i64 %x, 3
ret i64 %shr
}
%struct.bar = type { i16, i8, [5 x i8] }
define arm_aapcs_vfpcc void @fn1(%struct.bar* nocapture %a) {
; CHECK-MVE-LABEL: fn1:
; CHECK-MVE: @ %bb.0: @ %entry
; CHECK-MVE-NEXT: ldr r2, [r0, #4]
; CHECK-MVE-NEXT: movs r1, #0
; CHECK-MVE-NEXT: lsll r2, r1, #8
; CHECK-MVE-NEXT: strb r1, [r0, #7]
; CHECK-MVE-NEXT: str.w r2, [r0, #3]
; CHECK-MVE-NEXT: bx lr
;
; CHECK-NON-MVE-LABEL: fn1:
; CHECK-NON-MVE: @ %bb.0: @ %entry
; CHECK-NON-MVE-NEXT: ldr r1, [r0, #4]
; CHECK-NON-MVE-NEXT: lsrs r2, r1, #24
; CHECK-NON-MVE-NEXT: lsls r1, r1, #8
; CHECK-NON-MVE-NEXT: strb r2, [r0, #7]
; CHECK-NON-MVE-NEXT: str.w r1, [r0, #3]
; CHECK-NON-MVE-NEXT: bx lr
entry:
%carey = getelementptr inbounds %struct.bar, %struct.bar* %a, i32 0, i32 2
%0 = bitcast [5 x i8]* %carey to i40*
%bf.load = load i40, i40* %0, align 1
%bf.clear = and i40 %bf.load, -256
store i40 %bf.clear, i40* %0, align 1
ret void
}
%struct.a = type { i96 }
define void @lsll_128bit_shift(%struct.a* nocapture %x) local_unnamed_addr #0 {
; CHECK-LABEL: lsll_128bit_shift:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: movs r1, #0
; CHECK-NEXT: strd r1, r1, [r0]
; CHECK-NEXT: str r1, [r0, #8]
; CHECK-NEXT: bx lr
entry:
%0 = bitcast %struct.a* %x to i128*
%bf.load = load i128, i128* %0, align 8
%bf.clear4 = and i128 %bf.load, -79228162514264337593543950336
store i128 %bf.clear4, i128* %0, align 8
ret void
}
%struct.b = type { i184 }
define void @lsll_256bit_shift(%struct.b* nocapture %x) local_unnamed_addr #0 {
; CHECK-LABEL: lsll_256bit_shift:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: movs r1, #0
; CHECK-NEXT: str r1, [r0, #16]
; CHECK-NEXT: strd r1, r1, [r0, #8]
; CHECK-NEXT: strd r1, r1, [r0]
; CHECK-NEXT: ldrb r1, [r0, #23]
; CHECK-NEXT: lsls r1, r1, #24
; CHECK-NEXT: str r1, [r0, #20]
; CHECK-NEXT: bx lr
entry:
%0 = bitcast %struct.b* %x to i192*
%bf.load = load i192, i192* %0, align 8
%bf.clear4 = and i192 %bf.load, -24519928653854221733733552434404946937899825954937634816
store i192 %bf.clear4, i192* %0, align 8
ret void
}
define i32 @ashr_demand_bottom3(i64 %x) {
; CHECK-MVE-LABEL: ashr_demand_bottom3:
; CHECK-MVE: @ %bb.0: @ %entry
; CHECK-MVE-NEXT: lsrl r0, r1, #3
; CHECK-MVE-NEXT: bx lr
;
; CHECK-NON-MVE-LABEL: ashr_demand_bottom3:
; CHECK-NON-MVE: @ %bb.0: @ %entry
; CHECK-NON-MVE-NEXT: lsrs r0, r0, #3
; CHECK-NON-MVE-NEXT: orr.w r0, r0, r1, lsl #29
; CHECK-NON-MVE-NEXT: bx lr
entry:
%shr = ashr i64 %x, 3
%t = trunc i64 %shr to i32
ret i32 %t
}
define i32 @lshr_demand_bottom3(i64 %x) {
; CHECK-MVE-LABEL: lshr_demand_bottom3:
; CHECK-MVE: @ %bb.0: @ %entry
; CHECK-MVE-NEXT: lsrl r0, r1, #3
; CHECK-MVE-NEXT: bx lr
;
; CHECK-NON-MVE-LABEL: lshr_demand_bottom3:
; CHECK-NON-MVE: @ %bb.0: @ %entry
; CHECK-NON-MVE-NEXT: lsrs r0, r0, #3
; CHECK-NON-MVE-NEXT: orr.w r0, r0, r1, lsl #29
; CHECK-NON-MVE-NEXT: bx lr
entry:
%shr = lshr i64 %x, 3
%t = trunc i64 %shr to i32
ret i32 %t
}
define i32 @lsl_demand_bottom3(i64 %x) {
; CHECK-LABEL: lsl_demand_bottom3:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: lsls r0, r0, #3
; CHECK-NEXT: bx lr
entry:
%shr = shl i64 %x, 3
%t = trunc i64 %shr to i32
ret i32 %t
}
define i32 @ashr_demand_bottom31(i64 %x) {
; CHECK-MVE-LABEL: ashr_demand_bottom31:
; CHECK-MVE: @ %bb.0: @ %entry
; CHECK-MVE-NEXT: lsrl r0, r1, #31
; CHECK-MVE-NEXT: bx lr
;
; CHECK-NON-MVE-LABEL: ashr_demand_bottom31:
; CHECK-NON-MVE: @ %bb.0: @ %entry
; CHECK-NON-MVE-NEXT: lsrs r0, r0, #31
; CHECK-NON-MVE-NEXT: orr.w r0, r0, r1, lsl #1
; CHECK-NON-MVE-NEXT: bx lr
entry:
%shr = ashr i64 %x, 31
%t = trunc i64 %shr to i32
ret i32 %t
}
define i32 @lshr_demand_bottom31(i64 %x) {
; CHECK-MVE-LABEL: lshr_demand_bottom31:
; CHECK-MVE: @ %bb.0: @ %entry
; CHECK-MVE-NEXT: lsrl r0, r1, #31
; CHECK-MVE-NEXT: bx lr
;
; CHECK-NON-MVE-LABEL: lshr_demand_bottom31:
; CHECK-NON-MVE: @ %bb.0: @ %entry
; CHECK-NON-MVE-NEXT: lsrs r0, r0, #31
; CHECK-NON-MVE-NEXT: orr.w r0, r0, r1, lsl #1
; CHECK-NON-MVE-NEXT: bx lr
entry:
%shr = lshr i64 %x, 31
%t = trunc i64 %shr to i32
ret i32 %t
}
define i32 @lsl_demand_bottom31(i64 %x) {
; CHECK-LABEL: lsl_demand_bottom31:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: lsls r0, r0, #31
; CHECK-NEXT: bx lr
entry:
%shr = shl i64 %x, 31
%t = trunc i64 %shr to i32
ret i32 %t
}
define i32 @ashr_demand_bottom32(i64 %x) {
; CHECK-LABEL: ashr_demand_bottom32:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: mov r0, r1
; CHECK-NEXT: bx lr
entry:
%shr = ashr i64 %x, 32
%t = trunc i64 %shr to i32
ret i32 %t
}
define i32 @lshr_demand_bottom32(i64 %x) {
; CHECK-LABEL: lshr_demand_bottom32:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: mov r0, r1
; CHECK-NEXT: bx lr
entry:
%shr = lshr i64 %x, 32
%t = trunc i64 %shr to i32
ret i32 %t
}
define i32 @lsl_demand_bottom32(i64 %x) {
; CHECK-LABEL: lsl_demand_bottom32:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: movs r0, #0
; CHECK-NEXT: bx lr
entry:
%shr = shl i64 %x, 32
%t = trunc i64 %shr to i32
ret i32 %t
}
define i32 @ashr_demand_bottom44(i64 %x) {
; CHECK-LABEL: ashr_demand_bottom44:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: asrs r0, r1, #12
; CHECK-NEXT: bx lr
entry:
%shr = ashr i64 %x, 44
%t = trunc i64 %shr to i32
ret i32 %t
}
define i32 @lshr_demand_bottom44(i64 %x) {
; CHECK-LABEL: lshr_demand_bottom44:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: lsrs r0, r1, #12
; CHECK-NEXT: bx lr
entry:
%shr = lshr i64 %x, 44
%t = trunc i64 %shr to i32
ret i32 %t
}
define i32 @lsl_demand_bottom44(i64 %x) {
; CHECK-LABEL: lsl_demand_bottom44:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: movs r0, #0
; CHECK-NEXT: bx lr
entry:
%shr = shl i64 %x, 44
%t = trunc i64 %shr to i32
ret i32 %t
}
define i32 @ashr_demand_bottommask(i64 %x) {
; CHECK-LABEL: ashr_demand_bottommask:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: lsls r0, r1, #1
; CHECK-NEXT: bx lr
entry:
%shr = ashr i64 %x, 31
%t = trunc i64 %shr to i32
%a = and i32 %t, -2
ret i32 %a
}
define i32 @lshr_demand_bottommask(i64 %x) {
; CHECK-LABEL: lshr_demand_bottommask:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: lsls r0, r1, #1
; CHECK-NEXT: bx lr
entry:
%shr = lshr i64 %x, 31
%t = trunc i64 %shr to i32
%a = and i32 %t, -2
ret i32 %a
}
define i32 @lsl_demand_bottommask(i64 %x) {
; CHECK-LABEL: lsl_demand_bottommask:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: lsls r0, r0, #31
; CHECK-NEXT: bx lr
entry:
%shr = shl i64 %x, 31
%t = trunc i64 %shr to i32
%a = and i32 %t, -2
ret i32 %a
}
define i32 @ashr_demand_bottommask2(i64 %x) {
; CHECK-LABEL: ashr_demand_bottommask2:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: mvn r0, #2
; CHECK-NEXT: and.w r0, r0, r1, lsl #1
; CHECK-NEXT: bx lr
entry:
%shr = ashr i64 %x, 31
%t = trunc i64 %shr to i32
%a = and i32 %t, -4
ret i32 %a
}
define i32 @lshr_demand_bottommask2(i64 %x) {
; CHECK-LABEL: lshr_demand_bottommask2:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: mvn r0, #2
; CHECK-NEXT: and.w r0, r0, r1, lsl #1
; CHECK-NEXT: bx lr
entry:
%shr = lshr i64 %x, 31
%t = trunc i64 %shr to i32
%a = and i32 %t, -4
ret i32 %a
}
define i32 @lsl_demand_bottommask2(i64 %x) {
; CHECK-LABEL: lsl_demand_bottommask2:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: lsls r0, r0, #31
; CHECK-NEXT: bx lr
entry:
%shr = shl i64 %x, 31
%t = trunc i64 %shr to i32
%a = and i32 %t, -4
ret i32 %a
}
define i32 @lsl_demand_topmask(i64 %x) {
; CHECK-LABEL: lsl_demand_topmask:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: ubfx r0, r0, #1, #28
; CHECK-NEXT: bx lr
entry:
%sh = shl i64 %x, 31
%a = and i64 %sh, 1152921500311879680 ;0x0fffffff00000000
%l = ashr i64 %a, 32
%t = trunc i64 %l to i32
ret i32 %t
}