Skip to content

Commit

Permalink
[AIEX][Multi Slot] enable by default
Browse files Browse the repository at this point in the history
  • Loading branch information
@F-Stuckmann
F-Stuckmann committed Feb 25, 2025
1 parent 3fcf76d commit d0424ad
Show file tree
Hide file tree
Showing 3 changed files with 23 additions and 23 deletions.
2 changes: 1 addition & 1 deletion llvm/lib/Target/AIE/AIEInterBlockScheduling.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -62,7 +62,7 @@ static cl::opt<int> PostPipelinerMaxII(
cl::desc("[AIE] Maximum II to be tried in the post-ra pipeliner"));

static cl::opt<bool> EnableMultiSlotInstrMaterialization(
"aie-preassign-multi-slot-instr", cl::Hidden, cl::init(false),
"aie-preassign-multi-slot-instr", cl::Hidden, cl::init(true),
cl::desc("Statically materialize Multi-Slot Pseudo Instructions in "
"loops."));

Expand Down
20 changes: 10 additions & 10 deletions llvm/test/CodeGen/AIE/aie2/schedule/postpipeliner/bitwisexor.mir
View file
Original file line number Diff line number Diff line change
Expand Up @@ -26,27 +26,27 @@
; CHECK-NEXT: nop // Delay Slot 2
; CHECK-NEXT: nop // Delay Slot 1
; CHECK-NEXT: // %bb.1: // %for.body.preheader
; CHECK-NEXT: vlda wh1, [p0, #32]; vldb wh0, [p1, #32]
; CHECK-NEXT: vldb wl0, [p1], #64
; CHECK-NEXT: vldb wl1, [p0], #64
; CHECK-NEXT: vlda wh1, [p0, #32]; vldb wh0, [p1, #32]; nopx
; CHECK-NEXT: vlda wl1, [p0], #64; vldb wl0, [p1], #64
; CHECK-NEXT: nop
; CHECK-NEXT: nop
; CHECK-NEXT: nop
; CHECK-NEXT: vlda wh1, [p0, #32]; vldb wh0, [p1, #32]; add.nc lc, r0, #-3
; CHECK-NEXT: vldb wl0, [p1], #64; movxm ls, #.LBB0_2
; CHECK-NEXT: vldb wl1, [p0], #64; movxm le, #.L_LEnd0
; CHECK-NEXT: vlda wl1, [p0], #64; vldb wl0, [p1], #64; movxm ls, #.LBB0_2
; CHECK-NEXT: movxm le, #.L_LEnd0
; CHECK-NEXT: nopb ; nopa ; nops ; nopx ; vbneg_ltz.s8 x2, r25:r24, x0; nopv
; CHECK-NEXT: nopb ; nopa ; nops ; nopx ; vbneg_ltz.s8 x3, r25:r24, x1; nopv
; CHECK-NEXT: vldb wh0, [p1, #32]; vlda wh1, [p0, #32]; nops ; nopx ; vband x4, x0, x3; nopv
; CHECK-NEXT: vldb wl0, [p1], #64; nopa ; nops ; nopx ; vband x5, x1, x2; nopv
; CHECK-NEXT: vldb wl1, [p0], #64; nopa ; nops ; nopx ; vbor x6, x4, x5; nopv
; CHECK-NEXT: vldb wl0, [p1], #64; vlda wl1, [p0], #64; nops ; nopx ; vband x5, x1, x2; nopv
; CHECK-NEXT: nopb ; nopa ; nops ; nopx ; vbor x6, x4, x5; nopv
; CHECK-NEXT: nopb ; nopa ; nops ; nopx ; vbneg_ltz.s8 x2, r25:r24, x0; nopv
; CHECK-NEXT: nopb ; nopa ; vst wh6, [p2, #32]; nopx ; vbneg_ltz.s8 x3, r25:r24, x1; nopv
; CHECK-NEXT: .p2align 4
; CHECK-NEXT: .LBB0_2: // %for.body
; CHECK-NEXT: // =>This Inner Loop Header: Depth=1
; CHECK-NEXT: vldb wh0, [p1, #32]; vlda wh1, [p0, #32]; vst wl6, [p2], #64; nopx ; vband x4, x0, x3; nopv
; CHECK-NEXT: vldb wl0, [p1], #64; vband x5, x1, x2
; CHECK-NEXT: vldb wl1, [p0], #64; vbor x6, x4, x5
; CHECK-NEXT: vlda wh1, [p0, #32]; vldb wh0, [p1, #32]; nopx ; vband x4, x0, x3; vst wl6, [p2], #64
; CHECK-NEXT: vlda wl1, [p0], #64; vldb wl0, [p1], #64; vband x5, x1, x2
; CHECK-NEXT: vbor x6, x4, x5
; CHECK-NEXT: vbneg_ltz.s8 x2, r25:r24, x0
; CHECK-NEXT: .L_LEnd0:
; CHECK-NEXT: nopb ; nopa ; vst wh6, [p2, #32]; nopx ; vbneg_ltz.s8 x3, r25:r24, x1; nopv
Expand Down
24 changes: 12 additions & 12 deletions llvm/test/CodeGen/AIE/aie2/schedule/postpipeliner/gemm-nopresched.mir
View file
Original file line number Diff line number Diff line change
Expand Up @@ -25,38 +25,38 @@
; CHECK-NEXT: nop // Delay Slot 2
; CHECK-NEXT: nop // Delay Slot 1
; CHECK-NEXT: // %bb.1: // %for.body.preheader
; CHECK-NEXT: vlda wl11, [p1], m5; vldb wl8, [p0], #32; nopxm
; CHECK-NEXT: vlda wl11, [p1], m5; vldb wl8, [p0], #32; nopm
; CHECK-NEXT: vlda wh11, [p1], m6; vldb wh8, [p0], #32
; CHECK-NEXT: vlda wl5, [p1], m5; vldb wl1, [p0], #32
; CHECK-NEXT: vlda wh5, [p1], m6; vldb wh1, [p0], #32
; CHECK-NEXT: paddb [p0], m4
; CHECK-NEXT: vldb wl0, [p0], #32
; CHECK-NEXT: vldb wh0, [p0], #32; add.nc lc, r0, #-1
; CHECK-NEXT: vldb wl3, [p0], #32; movxm ls, #.LBB0_2
; CHECK-NEXT: vldb.3d wh3, [p0], d0; movxm le, #.L_LEnd0
; CHECK-NEXT: vldb wl0, [p1], m5; nopa ; nops ; nopxm ; nopv
; CHECK-NEXT: vldb wh0, [p1], m6; nopa ; nops ; nopx ; vshuffle x5, x5, x5, r6; nopv
; CHECK-NEXT: vldb wl7, [p1], m5; nopa ; nops ; nopxm ; nopv
; CHECK-NEXT: vldb.3d wh7, [p1], d1; nopa ; nops ; nopxm ; nopv
; CHECK-NEXT: vlda wl0, [p1], m5; vldb.3d wh3, [p0], d0; movxm le, #.L_LEnd0
; CHECK-NEXT: nopb ; vlda wh0, [p1], m6; nops ; nopxm ; nopv
; CHECK-NEXT: nopb ; vlda wl7, [p1], m5; nops ; nopx ; vshuffle x5, x5, x5, r6; nopv
; CHECK-NEXT: nopb ; vlda.3d wh7, [p1], d1; nops ; nopxm ; nopv
; CHECK-NEXT: nopb ; nopa ; nops ; nopxm ; nopv
; CHECK-NEXT: nopb ; nopa ; nops ; nopx ; vshuffle x6, x8, x0, r3; nopv
; CHECK-NEXT: nopb ; nopa ; nops ; nopx ; vshuffle x2, x8, x0, r16; nopv
; CHECK-NEXT: nopb ; nopa ; nops ; nopx ; vshuffle x10, x1, x3, r3; vmac.f bmh4, bmh4, x6, x5, r2
; CHECK-NEXT: .p2align 4
; CHECK-NEXT: .LBB0_2: // %for.body
; CHECK-NEXT: // =>This Inner Loop Header: Depth=1
; CHECK-NEXT: vldb wl8, [p0], #32; vlda wl11, [p1], m5; nops ; nopx ; vshuffle x9, x1, x3, r16; vmac.f bmh5, bmh5, x2, x5, r2
; CHECK-NEXT: vlda wh11, [p1], m6; vldb wh8, [p0], #32; nopx ; vshuffle x3, x11, x11, r6; vmac.f bmh6, bmh6, x10, x5, r2
; CHECK-NEXT: vldb wh8, [p0], #32; vlda wh11, [p1], m6; nops ; nopx ; vshuffle x3, x11, x11, r6; vmac.f bmh6, bmh6, x10, x5, r2
; CHECK-NEXT: vlda wl5, [p1], m5; vldb wl1, [p0], #32; vshuffle x0, x0, x0, r6; vmac.f bmh7, bmh7, x9, x5, r2
; CHECK-NEXT: vlda wh5, [p1], m6; vldb wh1, [p0], #32; vshuffle x7, x7, x7, r6; vmac.f bmh0, bmh0, x6, x3, r2
; CHECK-NEXT: paddb [p0], m4; vmac.f bmh1, bmh1, x2, x3, r2
; CHECK-NEXT: vldb wl0, [p0], #32; vmac.f bmh2, bmh2, x10, x3, r2
; CHECK-NEXT: vldb wh0, [p0], #32; vmac.f bmh3, bmh3, x9, x3, r2
; CHECK-NEXT: vldb wl3, [p0], #32; vmac.f bmh8, bmh8, x6, x0, r2
; CHECK-NEXT: vldb.3d wh3, [p0], d0; vmac.f bml0, bml0, x2, x0, r2
; CHECK-NEXT: vldb wl0, [p1], m5; vmac.f bml1, bml1, x10, x0, r2
; CHECK-NEXT: vldb wh0, [p1], m6; vshuffle x5, x5, x5, r6; vmac.f bml2, bml2, x9, x0, r2
; CHECK-NEXT: vldb wl7, [p1], m5; vmac.f bml3, bml3, x6, x7, r2
; CHECK-NEXT: vldb.3d wh7, [p1], d1; vmac.f bml4, bml4, x2, x7, r2
; CHECK-NEXT: vlda wl0, [p1], m5; vldb.3d wh3, [p0], d0; vmac.f bml0, bml0, x2, x0, r2
; CHECK-NEXT: vlda wh0, [p1], m6; vmac.f bml1, bml1, x10, x0, r2
; CHECK-NEXT: vlda wl7, [p1], m5; vshuffle x5, x5, x5, r6; vmac.f bml2, bml2, x9, x0, r2
; CHECK-NEXT: vlda.3d wh7, [p1], d1; vmac.f bml3, bml3, x6, x7, r2
; CHECK-NEXT: vmac.f bml4, bml4, x2, x7, r2
; CHECK-NEXT: vshuffle x6, x8, x0, r3; vmac.f bml6, bml6, x10, x7, r2
; CHECK-NEXT: vshuffle x2, x8, x0, r16; vmac.f bml5, bml5, x9, x7, r2
; CHECK-NEXT: .L_LEnd0:
Expand Down

0 comments on commit d0424ad

Please sign in to comment.