Revert "AMDGPU: Don't avoid clamp of bit shift in BFE pattern (#115372)"

llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp

@@ -22,6 +22,7 @@
 #include "SIISelLowering.h"
 #include "SIMachineFunctionInfo.h"
 #include "llvm/Analysis/UniformityAnalysis.h"
+#include "llvm/Analysis/ValueTracking.h"
 #include "llvm/CodeGen/FunctionLoweringInfo.h"
 #include "llvm/CodeGen/SelectionDAG.h"
 #include "llvm/CodeGen/SelectionDAGISel.h"

llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.h

@@ -17,7 +17,6 @@
 #include "GCNSubtarget.h"
 #include "SIMachineFunctionInfo.h"
 #include "SIModeRegisterDefaults.h"
-#include "llvm/Analysis/ValueTracking.h"
 #include "llvm/CodeGen/SelectionDAGISel.h"
 #include "llvm/Target/TargetMachine.h"
 

llvm/lib/Target/AMDGPU/SIInstructions.td

@@ -3556,23 +3556,6 @@ def : AMDGPUPat <
   (V_BFE_U32_e64 $src, (i32 0), $width)
 >;
 
-def uint5Bits : PatLeaf<(i32 VGPR_32:$width), [{
-  return CurDAG->computeKnownBits(SDValue(N, 0)).countMaxActiveBits() <= 5;
-}]>;
-
-// x << (bitwidth - y) >> (bitwidth - y)
-def : AMDGPUPat <
-  (DivergentBinFrag<srl> (shl_oneuse i32:$src, (sub 32, uint5Bits:$width)),
-                         (sub 32, uint5Bits:$width)),
-   (V_BFE_U32_e64 $src, (i32 0), $width)
->;
-
-def : AMDGPUPat <
-  (DivergentBinFrag<sra> (shl_oneuse i32:$src, (sub 32, uint5Bits:$width)),
-                         (sub 32, uint5Bits:$width)),
-   (V_BFE_I32_e64 $src, (i32 0), $width)
->;
-
 // SHA-256 Ma patterns
 
 // ((x & z) | (y & (x | z))) -> BFI (XOR x, y), z, y

llvm/test/CodeGen/AMDGPU/bfe-patterns.ll

@@ -17,8 +17,9 @@ define amdgpu_kernel void @v_ubfe_sub_i32(ptr addrspace(1) %out, ptr addrspace(1
 ; SI-NEXT:    buffer_load_dword v3, v[0:1], s[4:7], 0 addr64 glc
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    s_mov_b64 s[2:3], s[6:7]
-; SI-NEXT:    v_and_b32_e32 v3, 31, v3
-; SI-NEXT:    v_bfe_u32 v2, v2, 0, v3
+; SI-NEXT:    v_sub_i32_e32 v3, vcc, 32, v3
+; SI-NEXT:    v_lshlrev_b32_e32 v2, v3, v2
+; SI-NEXT:    v_lshrrev_b32_e32 v2, v3, v2
 ; SI-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
 ; SI-NEXT:    s_endpgm
 ;
@@ -37,8 +38,9 @@ define amdgpu_kernel void @v_ubfe_sub_i32(ptr addrspace(1) %out, ptr addrspace(1
 ; VI-NEXT:    v_mov_b32_e32 v1, s1
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
-; VI-NEXT:    v_and_b32_e32 v2, 31, v4
-; VI-NEXT:    v_bfe_u32 v2, v3, 0, v2
+; VI-NEXT:    v_sub_u32_e32 v2, vcc, 32, v4
+; VI-NEXT:    v_lshlrev_b32_e32 v3, v2, v3
+; VI-NEXT:    v_lshrrev_b32_e32 v2, v2, v3
 ; VI-NEXT:    flat_store_dword v[0:1], v2
 ; VI-NEXT:    s_endpgm
   %id.x = tail call i32 @llvm.amdgcn.workitem.id.x()
@@ -47,8 +49,7 @@ define amdgpu_kernel void @v_ubfe_sub_i32(ptr addrspace(1) %out, ptr addrspace(1
   %out.gep = getelementptr i32, ptr addrspace(1) %out, i32 %id.x
   %src = load volatile i32, ptr addrspace(1) %in0.gep
   %width = load volatile i32, ptr addrspace(1) %in0.gep
-  %width5 = and i32 %width, 31
-  %sub = sub i32 32, %width5
+  %sub = sub i32 32, %width
   %shl = shl i32 %src, %sub
   %bfe = lshr i32 %shl, %sub
   store i32 %bfe, ptr addrspace(1) %out.gep
@@ -71,7 +72,6 @@ define amdgpu_kernel void @v_ubfe_sub_multi_use_shl_i32(ptr addrspace(1) %out, p
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    s_mov_b64 s[2:3], s[6:7]
 ; SI-NEXT:    s_mov_b32 s6, -1
-; SI-NEXT:    v_and_b32_e32 v3, 31, v3
 ; SI-NEXT:    v_sub_i32_e32 v3, vcc, 32, v3
 ; SI-NEXT:    v_lshlrev_b32_e32 v2, v3, v2
 ; SI-NEXT:    v_lshrrev_b32_e32 v3, v3, v2
@@ -95,8 +95,7 @@ define amdgpu_kernel void @v_ubfe_sub_multi_use_shl_i32(ptr addrspace(1) %out, p
 ; VI-NEXT:    v_mov_b32_e32 v1, s1
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
-; VI-NEXT:    v_and_b32_e32 v2, 31, v4
-; VI-NEXT:    v_sub_u32_e32 v2, vcc, 32, v2
+; VI-NEXT:    v_sub_u32_e32 v2, vcc, 32, v4
 ; VI-NEXT:    v_lshlrev_b32_e32 v3, v2, v3
 ; VI-NEXT:    v_lshrrev_b32_e32 v2, v2, v3
 ; VI-NEXT:    flat_store_dword v[0:1], v2
@@ -109,8 +108,7 @@ define amdgpu_kernel void @v_ubfe_sub_multi_use_shl_i32(ptr addrspace(1) %out, p
   %out.gep = getelementptr i32, ptr addrspace(1) %out, i32 %id.x
   %src = load volatile i32, ptr addrspace(1) %in0.gep
   %width = load volatile i32, ptr addrspace(1) %in0.gep
-  %width5 = and i32 %width, 31
-  %sub = sub i32 32, %width5
+  %sub = sub i32 32, %width
   %shl = shl i32 %src, %sub
   %bfe = lshr i32 %shl, %sub
   store i32 %bfe, ptr addrspace(1) %out.gep
@@ -221,8 +219,9 @@ define amdgpu_kernel void @v_sbfe_sub_i32(ptr addrspace(1) %out, ptr addrspace(1
 ; SI-NEXT:    buffer_load_dword v3, v[0:1], s[4:7], 0 addr64 glc
 ; SI-NEXT:    s_waitcnt vmcnt(0)
 ; SI-NEXT:    s_mov_b64 s[2:3], s[6:7]
-; SI-NEXT:    v_and_b32_e32 v3, 31, v3
-; SI-NEXT:    v_bfe_i32 v2, v2, 0, v3
+; SI-NEXT:    v_sub_i32_e32 v3, vcc, 32, v3
+; SI-NEXT:    v_lshlrev_b32_e32 v2, v3, v2
+; SI-NEXT:    v_ashrrev_i32_e32 v2, v3, v2
 ; SI-NEXT:    buffer_store_dword v2, v[0:1], s[0:3], 0 addr64
 ; SI-NEXT:    s_endpgm
 ;
@@ -241,8 +240,9 @@ define amdgpu_kernel void @v_sbfe_sub_i32(ptr addrspace(1) %out, ptr addrspace(1
 ; VI-NEXT:    v_mov_b32_e32 v1, s1
 ; VI-NEXT:    v_add_u32_e32 v0, vcc, s0, v2
 ; VI-NEXT:    v_addc_u32_e32 v1, vcc, 0, v1, vcc
-; VI-NEXT:    v_and_b32_e32 v2, 31, v4
-; VI-NEXT:    v_bfe_i32 v2, v3, 0, v2
+; VI-NEXT:    v_sub_u32_e32 v2, vcc, 32, v4
+; VI-NEXT:    v_lshlrev_b32_e32 v3, v2, v3
+; VI-NEXT:    v_ashrrev_i32_e32 v2, v2, v3
 ; VI-NEXT:    flat_store_dword v[0:1], v2
 ; VI-NEXT:    s_endpgm
   %id.x = tail call i32 @llvm.amdgcn.workitem.id.x()
@@ -251,8 +251,7 @@ define amdgpu_kernel void @v_sbfe_sub_i32(ptr addrspace(1) %out, ptr addrspace(1
   %out.gep = getelementptr i32, ptr addrspace(1) %out, i32 %id.x
   %src = load volatile i32, ptr addrspace(1) %in0.gep
   %width = load volatile i32, ptr addrspace(1) %in0.gep
-  %width5 = and i32 %width, 31
-  %sub = sub i32 32, %width5
+  %sub = sub i32 32, %width
   %shl = shl i32 %src, %sub
   %bfe = ashr i32 %shl, %sub
   store i32 %bfe, ptr addrspace(1) %out.gep

llvm/test/CodeGen/AMDGPU/extract-lowbits.ll

@@ -150,39 +150,47 @@ define i32 @bzhi32_c4_commutative(i32 %val, i32 %numlowbits) nounwind {
 ; ---------------------------------------------------------------------------- ;
 
 define i32 @bzhi32_d0(i32 %val, i32 %numlowbits) nounwind {
-; GCN-LABEL: bzhi32_d0:
-; GCN:       ; %bb.0:
-; GCN-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT:    v_and_b32_e32 v1, 31, v1
-; GCN-NEXT:    v_bfe_u32 v0, v0, 0, v1
-; GCN-NEXT:    s_setpc_b64 s[30:31]
-  %numlow5bits = and i32 %numlowbits, 31
-  %numhighbits = sub i32 32, %numlow5bits
+; SI-LABEL: bzhi32_d0:
+; SI:       ; %bb.0:
+; SI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SI-NEXT:    v_sub_i32_e32 v1, vcc, 32, v1
+; SI-NEXT:    v_lshlrev_b32_e32 v0, v1, v0
+; SI-NEXT:    v_lshrrev_b32_e32 v0, v1, v0
+; SI-NEXT:    s_setpc_b64 s[30:31]
+;
+; VI-LABEL: bzhi32_d0:
+; VI:       ; %bb.0:
+; VI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; VI-NEXT:    v_sub_u32_e32 v1, vcc, 32, v1
+; VI-NEXT:    v_lshlrev_b32_e32 v0, v1, v0
+; VI-NEXT:    v_lshrrev_b32_e32 v0, v1, v0
+; VI-NEXT:    s_setpc_b64 s[30:31]
+  %numhighbits = sub i32 32, %numlowbits
   %highbitscleared = shl i32 %val, %numhighbits
   %masked = lshr i32 %highbitscleared, %numhighbits
   ret i32 %masked
 }
 
-define i32 @bzhi32_d0_even(i32 %val, i32 %numlowbits) nounwind {
-; SI-LABEL: bzhi32_d0_even:
+define i32 @bzhi32_d0_5bits(i32 %val, i32 %numlowbits) nounwind {
+; SI-LABEL: bzhi32_d0_5bits:
 ; SI:       ; %bb.0:
 ; SI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; SI-NEXT:    v_lshlrev_b32_e32 v1, 1, v1
+; SI-NEXT:    v_and_b32_e32 v1, 31, v1
 ; SI-NEXT:    v_sub_i32_e32 v1, vcc, 32, v1
 ; SI-NEXT:    v_lshlrev_b32_e32 v0, v1, v0
 ; SI-NEXT:    v_lshrrev_b32_e32 v0, v1, v0
 ; SI-NEXT:    s_setpc_b64 s[30:31]
 ;
-; VI-LABEL: bzhi32_d0_even:
+; VI-LABEL: bzhi32_d0_5bits:
 ; VI:       ; %bb.0:
 ; VI-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; VI-NEXT:    v_lshlrev_b32_e32 v1, 1, v1
+; VI-NEXT:    v_and_b32_e32 v1, 31, v1
 ; VI-NEXT:    v_sub_u32_e32 v1, vcc, 32, v1
 ; VI-NEXT:    v_lshlrev_b32_e32 v0, v1, v0
 ; VI-NEXT:    v_lshrrev_b32_e32 v0, v1, v0
 ; VI-NEXT:    s_setpc_b64 s[30:31]
-  %times2 = shl i32 %numlowbits, 1
-  %numhighbits = sub i32 32, %times2
+  %numlow5bits = and i32 %numlowbits, 31
+  %numhighbits = sub i32 32, %numlow5bits
   %highbitscleared = shl i32 %val, %numhighbits
   %masked = lshr i32 %highbitscleared, %numhighbits
   ret i32 %masked