Skip to content

Commit 76ced7f

Browse files
PankajDwivedi-25PankajDwivedi-25
authored
[AMDGPU] Insert readfirstlane in the function returns in sgpr. (#135326)
insert `readfirstlane` in the function returns in sgpr.
1 parent 5133b43 commit 76ced7f

14 files changed

+733
-538
lines changed

‎llvm/lib/Target/AMDGPU/SIISelLowering.cpp‎

Lines changed: 6 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -3221,6 +3221,7 @@ SITargetLowering::LowerReturn(SDValue Chain, CallingConv::ID CallConv,
32213221
const SDLoc &DL, SelectionDAG &DAG) const {
32223222
MachineFunction &MF = DAG.getMachineFunction();
32233223
SIMachineFunctionInfo *Info = MF.getInfo<SIMachineFunctionInfo>();
3224+
const SIRegisterInfo *TRI = getSubtarget()->getRegisterInfo();
32243225

32253226
if (AMDGPU::isKernel(CallConv)) {
32263227
return AMDGPUTargetLowering::LowerReturn(Chain, CallConv, isVarArg, Outs,
@@ -3247,6 +3248,8 @@ SITargetLowering::LowerReturn(SDValue Chain, CallingConv::ID CallConv,
32473248
SmallVector<SDValue, 48> RetOps;
32483249
RetOps.push_back(Chain); // Operand #0 = Chain (updated below)
32493250

3251+
SDValue ReadFirstLane =
3252+
DAG.getTargetConstant(Intrinsic::amdgcn_readfirstlane, DL, MVT::i32);
32503253
// Copy the result values into the output registers.
32513254
for (unsigned I = 0, RealRVLocIdx = 0, E = RVLocs.size(); I != E;
32523255
++I, ++RealRVLocIdx) {
@@ -3274,7 +3277,9 @@ SITargetLowering::LowerReturn(SDValue Chain, CallingConv::ID CallConv,
32743277
default:
32753278
llvm_unreachable("Unknown loc info!");
32763279
}
3277-
3280+
if (TRI->isSGPRPhysReg(VA.getLocReg()))
3281+
Arg = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, DL, Arg.getValueType(),
3282+
ReadFirstLane, Arg);
32783283
Chain = DAG.getCopyToReg(Chain, DL, VA.getLocReg(), Arg, Glue);
32793284
Glue = Chain.getValue(1);
32803285
RetOps.push_back(DAG.getRegister(VA.getLocReg(), VA.getLocVT()));

‎llvm/test/CodeGen/AMDGPU/add64-low-32-bits-known-zero.ll‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -148,8 +148,8 @@ define <2 x i64> @v_add_v2i64_nonsplat_const_low_bits_known0_0(<2 x i64> %reg) {
148148
define amdgpu_ps <2 x i64> @s_add_v2i64_splat_const_low_bits_known0_0(<2 x i64> inreg %reg) {
149149
; GFX9-LABEL: s_add_v2i64_splat_const_low_bits_known0_0:
150150
; GFX9: ; %bb.0:
151-
; GFX9-NEXT: s_add_i32 s1, s1, 1
152151
; GFX9-NEXT: s_add_i32 s3, s3, 1
152+
; GFX9-NEXT: s_add_i32 s1, s1, 1
153153
; GFX9-NEXT: ; return to shader part epilog
154154
%add = add <2 x i64> %reg, <i64 4294967296, i64 4294967296> ; (1 << 32)
155155
ret <2 x i64> %add
@@ -158,8 +158,8 @@ define amdgpu_ps <2 x i64> @s_add_v2i64_splat_const_low_bits_known0_0(<2 x i64>
158158
define amdgpu_ps <2 x i64> @s_add_v2i64_nonsplat_const_low_bits_known0_0(<2 x i64> inreg %reg) {
159159
; GFX9-LABEL: s_add_v2i64_nonsplat_const_low_bits_known0_0:
160160
; GFX9: ; %bb.0:
161-
; GFX9-NEXT: s_add_i32 s1, s1, 1
162161
; GFX9-NEXT: s_add_i32 s3, s3, 2
162+
; GFX9-NEXT: s_add_i32 s1, s1, 1
163163
; GFX9-NEXT: ; return to shader part epilog
164164
%add = add <2 x i64> %reg, <i64 4294967296, i64 8589934592> ; (1 << 32), (1 << 33)
165165
ret <2 x i64> %add

‎llvm/test/CodeGen/AMDGPU/buffer-atomic-fadd.f64.ll‎

Lines changed: 32 additions & 16 deletions
Original file line numberDiff line numberDiff line change
@@ -110,9 +110,11 @@ define amdgpu_ps double @buffer_atomic_fadd_f64_offset_rtn(double %val, <4 x i32
110110
; GFX90A_GFX942-NEXT: [[COPY7:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE1]]
111111
; GFX90A_GFX942-NEXT: [[BUFFER_ATOMIC_ADD_F64_OFFSET_RTN:%[0-9]+]]:vreg_64_align2 = BUFFER_ATOMIC_ADD_F64_OFFSET_RTN [[COPY7]], killed [[REG_SEQUENCE]], [[COPY]], 0, 1, implicit $exec :: (volatile dereferenceable load store (s64), align 1, addrspace 8)
112112
; GFX90A_GFX942-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_ATOMIC_ADD_F64_OFFSET_RTN]].sub0
113+
; GFX90A_GFX942-NEXT: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 killed [[COPY8]], implicit $exec
113114
; GFX90A_GFX942-NEXT: [[COPY9:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_ATOMIC_ADD_F64_OFFSET_RTN]].sub1
114-
; GFX90A_GFX942-NEXT: $sgpr0 = COPY [[COPY8]]
115-
; GFX90A_GFX942-NEXT: $sgpr1 = COPY [[COPY9]]
115+
; GFX90A_GFX942-NEXT: [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 killed [[COPY9]], implicit $exec
116+
; GFX90A_GFX942-NEXT: $sgpr0 = COPY [[V_READFIRSTLANE_B32_]]
117+
; GFX90A_GFX942-NEXT: $sgpr1 = COPY [[V_READFIRSTLANE_B32_1]]
116118
; GFX90A_GFX942-NEXT: SI_RETURN_TO_EPILOG $sgpr0, $sgpr1
117119
%ret = call double @llvm.amdgcn.raw.buffer.atomic.fadd.f64(double %val, <4 x i32> %rsrc, i32 0, i32 %soffset, i32 0)
118120
ret double %ret
@@ -136,9 +138,11 @@ define amdgpu_ps double @buffer_atomic_fadd_f64_offen_rtn(double %val, <4 x i32>
136138
; GFX90A_GFX942-NEXT: [[COPY8:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE1]]
137139
; GFX90A_GFX942-NEXT: [[BUFFER_ATOMIC_ADD_F64_OFFEN_RTN:%[0-9]+]]:vreg_64_align2 = BUFFER_ATOMIC_ADD_F64_OFFEN_RTN [[COPY8]], [[COPY1]], killed [[REG_SEQUENCE]], [[COPY]], 0, 1, implicit $exec :: (volatile dereferenceable load store (s64), align 1, addrspace 8)
138140
; GFX90A_GFX942-NEXT: [[COPY9:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_ATOMIC_ADD_F64_OFFEN_RTN]].sub0
141+
; GFX90A_GFX942-NEXT: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 killed [[COPY9]], implicit $exec
139142
; GFX90A_GFX942-NEXT: [[COPY10:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_ATOMIC_ADD_F64_OFFEN_RTN]].sub1
140-
; GFX90A_GFX942-NEXT: $sgpr0 = COPY [[COPY9]]
141-
; GFX90A_GFX942-NEXT: $sgpr1 = COPY [[COPY10]]
143+
; GFX90A_GFX942-NEXT: [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 killed [[COPY10]], implicit $exec
144+
; GFX90A_GFX942-NEXT: $sgpr0 = COPY [[V_READFIRSTLANE_B32_]]
145+
; GFX90A_GFX942-NEXT: $sgpr1 = COPY [[V_READFIRSTLANE_B32_1]]
142146
; GFX90A_GFX942-NEXT: SI_RETURN_TO_EPILOG $sgpr0, $sgpr1
143147
%ret = call double @llvm.amdgcn.raw.buffer.atomic.fadd.f64(double %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
144148
ret double %ret
@@ -162,9 +166,11 @@ define amdgpu_ps double @buffer_atomic_fadd_f64_idxen_rtn(double %val, <4 x i32>
162166
; GFX90A_GFX942-NEXT: [[COPY8:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE1]]
163167
; GFX90A_GFX942-NEXT: [[BUFFER_ATOMIC_ADD_F64_IDXEN_RTN:%[0-9]+]]:vreg_64_align2 = BUFFER_ATOMIC_ADD_F64_IDXEN_RTN [[COPY8]], [[COPY1]], killed [[REG_SEQUENCE]], [[COPY]], 0, 1, implicit $exec :: (volatile dereferenceable load store (s64), align 1, addrspace 8)
164168
; GFX90A_GFX942-NEXT: [[COPY9:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_ATOMIC_ADD_F64_IDXEN_RTN]].sub0
169+
; GFX90A_GFX942-NEXT: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 killed [[COPY9]], implicit $exec
165170
; GFX90A_GFX942-NEXT: [[COPY10:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_ATOMIC_ADD_F64_IDXEN_RTN]].sub1
166-
; GFX90A_GFX942-NEXT: $sgpr0 = COPY [[COPY9]]
167-
; GFX90A_GFX942-NEXT: $sgpr1 = COPY [[COPY10]]
171+
; GFX90A_GFX942-NEXT: [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 killed [[COPY10]], implicit $exec
172+
; GFX90A_GFX942-NEXT: $sgpr0 = COPY [[V_READFIRSTLANE_B32_]]
173+
; GFX90A_GFX942-NEXT: $sgpr1 = COPY [[V_READFIRSTLANE_B32_1]]
168174
; GFX90A_GFX942-NEXT: SI_RETURN_TO_EPILOG $sgpr0, $sgpr1
169175
%ret = call double @llvm.amdgcn.struct.buffer.atomic.fadd.f64(double %val, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 %soffset, i32 0)
170176
ret double %ret
@@ -190,9 +196,11 @@ define amdgpu_ps double @buffer_atomic_fadd_f64_bothen_rtn(double %val, <4 x i32
190196
; GFX90A_GFX942-NEXT: [[COPY9:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE1]]
191197
; GFX90A_GFX942-NEXT: [[BUFFER_ATOMIC_ADD_F64_BOTHEN_RTN:%[0-9]+]]:vreg_64_align2 = BUFFER_ATOMIC_ADD_F64_BOTHEN_RTN [[COPY9]], killed [[REG_SEQUENCE2]], killed [[REG_SEQUENCE]], [[COPY]], 0, 1, implicit $exec :: (volatile dereferenceable load store (s64), align 1, addrspace 8)
192198
; GFX90A_GFX942-NEXT: [[COPY10:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_ATOMIC_ADD_F64_BOTHEN_RTN]].sub0
199+
; GFX90A_GFX942-NEXT: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 killed [[COPY10]], implicit $exec
193200
; GFX90A_GFX942-NEXT: [[COPY11:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_ATOMIC_ADD_F64_BOTHEN_RTN]].sub1
194-
; GFX90A_GFX942-NEXT: $sgpr0 = COPY [[COPY10]]
195-
; GFX90A_GFX942-NEXT: $sgpr1 = COPY [[COPY11]]
201+
; GFX90A_GFX942-NEXT: [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 killed [[COPY11]], implicit $exec
202+
; GFX90A_GFX942-NEXT: $sgpr0 = COPY [[V_READFIRSTLANE_B32_]]
203+
; GFX90A_GFX942-NEXT: $sgpr1 = COPY [[V_READFIRSTLANE_B32_1]]
196204
; GFX90A_GFX942-NEXT: SI_RETURN_TO_EPILOG $sgpr0, $sgpr1
197205
%ret = call double @llvm.amdgcn.struct.buffer.atomic.fadd.f64(double %val, <4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 0)
198206
ret double %ret
@@ -334,9 +342,11 @@ define amdgpu_ps double @buffer_ptr_atomic_fadd_f64_offset_rtn(double %val, ptr
334342
; GFX90A_GFX942-NEXT: [[COPY11:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE3]]
335343
; GFX90A_GFX942-NEXT: [[BUFFER_ATOMIC_ADD_F64_OFFSET_RTN:%[0-9]+]]:vreg_64_align2 = BUFFER_ATOMIC_ADD_F64_OFFSET_RTN [[COPY11]], killed [[REG_SEQUENCE2]], [[COPY]], 0, 1, implicit $exec :: (volatile dereferenceable load store (s64) on %ir.rsrc, align 1, addrspace 8)
336344
; GFX90A_GFX942-NEXT: [[COPY12:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_ATOMIC_ADD_F64_OFFSET_RTN]].sub0
345+
; GFX90A_GFX942-NEXT: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 killed [[COPY12]], implicit $exec
337346
; GFX90A_GFX942-NEXT: [[COPY13:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_ATOMIC_ADD_F64_OFFSET_RTN]].sub1
338-
; GFX90A_GFX942-NEXT: $sgpr0 = COPY [[COPY12]]
339-
; GFX90A_GFX942-NEXT: $sgpr1 = COPY [[COPY13]]
347+
; GFX90A_GFX942-NEXT: [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 killed [[COPY13]], implicit $exec
348+
; GFX90A_GFX942-NEXT: $sgpr0 = COPY [[V_READFIRSTLANE_B32_]]
349+
; GFX90A_GFX942-NEXT: $sgpr1 = COPY [[V_READFIRSTLANE_B32_1]]
340350
; GFX90A_GFX942-NEXT: SI_RETURN_TO_EPILOG $sgpr0, $sgpr1
341351
%ret = call double @llvm.amdgcn.raw.ptr.buffer.atomic.fadd.f64(double %val, ptr addrspace(8) %rsrc, i32 0, i32 %soffset, i32 0)
342352
ret double %ret
@@ -366,9 +376,11 @@ define amdgpu_ps double @buffer_ptr_atomic_fadd_f64_offen_rtn(double %val, ptr a
366376
; GFX90A_GFX942-NEXT: [[COPY12:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE3]]
367377
; GFX90A_GFX942-NEXT: [[BUFFER_ATOMIC_ADD_F64_OFFEN_RTN:%[0-9]+]]:vreg_64_align2 = BUFFER_ATOMIC_ADD_F64_OFFEN_RTN [[COPY12]], [[COPY1]], killed [[REG_SEQUENCE2]], [[COPY]], 0, 1, implicit $exec :: (volatile dereferenceable load store (s64) on %ir.rsrc, align 1, addrspace 8)
368378
; GFX90A_GFX942-NEXT: [[COPY13:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_ATOMIC_ADD_F64_OFFEN_RTN]].sub0
379+
; GFX90A_GFX942-NEXT: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 killed [[COPY13]], implicit $exec
369380
; GFX90A_GFX942-NEXT: [[COPY14:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_ATOMIC_ADD_F64_OFFEN_RTN]].sub1
370-
; GFX90A_GFX942-NEXT: $sgpr0 = COPY [[COPY13]]
371-
; GFX90A_GFX942-NEXT: $sgpr1 = COPY [[COPY14]]
381+
; GFX90A_GFX942-NEXT: [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 killed [[COPY14]], implicit $exec
382+
; GFX90A_GFX942-NEXT: $sgpr0 = COPY [[V_READFIRSTLANE_B32_]]
383+
; GFX90A_GFX942-NEXT: $sgpr1 = COPY [[V_READFIRSTLANE_B32_1]]
372384
; GFX90A_GFX942-NEXT: SI_RETURN_TO_EPILOG $sgpr0, $sgpr1
373385
%ret = call double @llvm.amdgcn.raw.ptr.buffer.atomic.fadd.f64(double %val, ptr addrspace(8) %rsrc, i32 %voffset, i32 %soffset, i32 0)
374386
ret double %ret
@@ -398,9 +410,11 @@ define amdgpu_ps double @buffer_ptr_atomic_fadd_f64_idxen_rtn(double %val, ptr a
398410
; GFX90A_GFX942-NEXT: [[COPY12:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE3]]
399411
; GFX90A_GFX942-NEXT: [[BUFFER_ATOMIC_ADD_F64_IDXEN_RTN:%[0-9]+]]:vreg_64_align2 = BUFFER_ATOMIC_ADD_F64_IDXEN_RTN [[COPY12]], [[COPY1]], killed [[REG_SEQUENCE2]], [[COPY]], 0, 1, implicit $exec :: (volatile dereferenceable load store (s64) on %ir.rsrc, align 1, addrspace 8)
400412
; GFX90A_GFX942-NEXT: [[COPY13:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_ATOMIC_ADD_F64_IDXEN_RTN]].sub0
413+
; GFX90A_GFX942-NEXT: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 killed [[COPY13]], implicit $exec
401414
; GFX90A_GFX942-NEXT: [[COPY14:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_ATOMIC_ADD_F64_IDXEN_RTN]].sub1
402-
; GFX90A_GFX942-NEXT: $sgpr0 = COPY [[COPY13]]
403-
; GFX90A_GFX942-NEXT: $sgpr1 = COPY [[COPY14]]
415+
; GFX90A_GFX942-NEXT: [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 killed [[COPY14]], implicit $exec
416+
; GFX90A_GFX942-NEXT: $sgpr0 = COPY [[V_READFIRSTLANE_B32_]]
417+
; GFX90A_GFX942-NEXT: $sgpr1 = COPY [[V_READFIRSTLANE_B32_1]]
404418
; GFX90A_GFX942-NEXT: SI_RETURN_TO_EPILOG $sgpr0, $sgpr1
405419
%ret = call double @llvm.amdgcn.struct.ptr.buffer.atomic.fadd.f64(double %val, ptr addrspace(8) %rsrc, i32 %vindex, i32 0, i32 %soffset, i32 0)
406420
ret double %ret
@@ -432,9 +446,11 @@ define amdgpu_ps double @buffer_ptr_atomic_fadd_f64_bothen_rtn(double %val, ptr
432446
; GFX90A_GFX942-NEXT: [[COPY13:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE3]]
433447
; GFX90A_GFX942-NEXT: [[BUFFER_ATOMIC_ADD_F64_BOTHEN_RTN:%[0-9]+]]:vreg_64_align2 = BUFFER_ATOMIC_ADD_F64_BOTHEN_RTN [[COPY13]], killed [[REG_SEQUENCE4]], killed [[REG_SEQUENCE2]], [[COPY]], 0, 1, implicit $exec :: (volatile dereferenceable load store (s64) on %ir.rsrc, align 1, addrspace 8)
434448
; GFX90A_GFX942-NEXT: [[COPY14:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_ATOMIC_ADD_F64_BOTHEN_RTN]].sub0
449+
; GFX90A_GFX942-NEXT: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 killed [[COPY14]], implicit $exec
435450
; GFX90A_GFX942-NEXT: [[COPY15:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_ATOMIC_ADD_F64_BOTHEN_RTN]].sub1
436-
; GFX90A_GFX942-NEXT: $sgpr0 = COPY [[COPY14]]
437-
; GFX90A_GFX942-NEXT: $sgpr1 = COPY [[COPY15]]
451+
; GFX90A_GFX942-NEXT: [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 killed [[COPY15]], implicit $exec
452+
; GFX90A_GFX942-NEXT: $sgpr0 = COPY [[V_READFIRSTLANE_B32_]]
453+
; GFX90A_GFX942-NEXT: $sgpr1 = COPY [[V_READFIRSTLANE_B32_1]]
438454
; GFX90A_GFX942-NEXT: SI_RETURN_TO_EPILOG $sgpr0, $sgpr1
439455
%ret = call double @llvm.amdgcn.struct.ptr.buffer.atomic.fadd.f64(double %val, ptr addrspace(8) %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 0)
440456
ret double %ret

‎llvm/test/CodeGen/AMDGPU/constrained-shift.ll‎

Lines changed: 15 additions & 15 deletions
Original file line numberDiff line numberDiff line change
@@ -168,26 +168,26 @@ define <4 x i32> @csh_v4i32(<4 x i32> %a, <4 x i32> %b) {
168168
define amdgpu_ps <4 x i32> @s_csh_v4i32(<4 x i32> inreg %a, <4 x i32> inreg %b) {
169169
; CHECK-LABEL: s_csh_v4i32:
170170
; CHECK: ; %bb.0:
171-
; CHECK-NEXT: s_lshl_b32 s8, s0, s4
172-
; CHECK-NEXT: s_lshl_b32 s9, s1, s5
173-
; CHECK-NEXT: s_lshl_b32 s10, s2, s6
174-
; CHECK-NEXT: s_lshl_b32 s11, s3, s7
175-
; CHECK-NEXT: s_lshr_b32 s12, s0, s4
176-
; CHECK-NEXT: s_lshr_b32 s13, s1, s5
177-
; CHECK-NEXT: s_lshr_b32 s14, s2, s6
178-
; CHECK-NEXT: s_lshr_b32 s15, s3, s7
179-
; CHECK-NEXT: s_ashr_i32 s3, s3, s7
180-
; CHECK-NEXT: s_ashr_i32 s2, s2, s6
181-
; CHECK-NEXT: s_ashr_i32 s1, s1, s5
171+
; CHECK-NEXT: s_lshl_b32 s8, s3, s7
172+
; CHECK-NEXT: s_lshl_b32 s9, s2, s6
173+
; CHECK-NEXT: s_lshl_b32 s10, s1, s5
174+
; CHECK-NEXT: s_lshl_b32 s11, s0, s4
175+
; CHECK-NEXT: s_lshr_b32 s12, s3, s7
176+
; CHECK-NEXT: s_lshr_b32 s13, s2, s6
177+
; CHECK-NEXT: s_lshr_b32 s14, s1, s5
178+
; CHECK-NEXT: s_lshr_b32 s15, s0, s4
182179
; CHECK-NEXT: s_ashr_i32 s0, s0, s4
180+
; CHECK-NEXT: s_ashr_i32 s1, s1, s5
181+
; CHECK-NEXT: s_ashr_i32 s2, s2, s6
182+
; CHECK-NEXT: s_ashr_i32 s3, s3, s7
183183
; CHECK-NEXT: s_add_i32 s4, s11, s15
184184
; CHECK-NEXT: s_add_i32 s5, s10, s14
185185
; CHECK-NEXT: s_add_i32 s6, s9, s13
186186
; CHECK-NEXT: s_add_i32 s7, s8, s12
187-
; CHECK-NEXT: s_add_i32 s0, s7, s0
188-
; CHECK-NEXT: s_add_i32 s1, s6, s1
189-
; CHECK-NEXT: s_add_i32 s2, s5, s2
190-
; CHECK-NEXT: s_add_i32 s3, s4, s3
187+
; CHECK-NEXT: s_add_i32 s3, s7, s3
188+
; CHECK-NEXT: s_add_i32 s2, s6, s2
189+
; CHECK-NEXT: s_add_i32 s1, s5, s1
190+
; CHECK-NEXT: s_add_i32 s0, s4, s0
191191
; CHECK-NEXT: ; return to shader part epilog
192192
;
193193
; GISEL-LABEL: s_csh_v4i32:

0 commit comments

Comments
 (0)