[GISel][RISCV] Compute CTPOP of small odd-sized integer correctly

llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp

@@ -7678,6 +7678,10 @@ LegalizerHelper::lowerBitCount(MachineInstr &MI) {
     unsigned Size = Ty.getSizeInBits();
     MachineIRBuilder &B = MIRBuilder;
 
+    // Bail out on irregular type lengths.
+    if (Size >= 128 || Size % 8 != 0)
+      return UnableToLegalize;
+
     // Count set bits in blocks of 2 bits. Default approach would be
     // B2Count = { val & 0x55555555 } + { (val >> 1) & 0x55555555 }
     // We use following formula instead:

llvm/lib/Target/RISCV/GISel/RISCVLegalizerInfo.cpp

@@ -238,7 +238,10 @@ RISCVLegalizerInfo::RISCVLegalizerInfo(const RISCVSubtarget &ST)
         .clampScalar(0, sXLen, sXLen)
         .scalarSameSizeAs(1, 0);
   } else {
-    CTPOPActions.maxScalar(0, sXLen).scalarSameSizeAs(1, 0).lower();
+    CTPOPActions.widenScalarToNextPow2(0, /*Min*/ 8)
+        .clampScalar(0, s8, sXLen)
+        .scalarSameSizeAs(1, 0)
+        .lower();
   }
 
   getActionDefinitionsBuilder(G_CONSTANT)

llvm/test/CodeGen/RISCV/GlobalISel/bitmanip.ll

@@ -205,3 +205,143 @@ define i24 @bitreverse_i24(i24 %x) {
   %rev = call i24 @llvm.bitreverse.i24(i24 %x)
   ret i24 %rev
 }
+
+define i2 @test_ctpop_i2(i2 %a) {
+; RV32-LABEL: test_ctpop_i2:
+; RV32:       # %bb.0:
+; RV32-NEXT:    addi sp, sp, -16
+; RV32-NEXT:    .cfi_def_cfa_offset 16
+; RV32-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32-NEXT:    .cfi_offset ra, -4
+; RV32-NEXT:    andi a0, a0, 3
+; RV32-NEXT:    srli a1, a0, 1
+; RV32-NEXT:    sub a0, a0, a1
+; RV32-NEXT:    zext.b a1, a0
+; RV32-NEXT:    srli a1, a1, 2
+; RV32-NEXT:    andi a1, a1, 51
+; RV32-NEXT:    andi a0, a0, 51
+; RV32-NEXT:    add a0, a1, a0
+; RV32-NEXT:    srli a1, a0, 4
+; RV32-NEXT:    add a0, a1, a0
+; RV32-NEXT:    andi a0, a0, 15
+; RV32-NEXT:    li a1, 1
+; RV32-NEXT:    call __mulsi3
+; RV32-NEXT:    zext.b a0, a0
+; RV32-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32-NEXT:    .cfi_restore ra
+; RV32-NEXT:    addi sp, sp, 16
+; RV32-NEXT:    .cfi_def_cfa_offset 0
+; RV32-NEXT:    ret
+;
+; RV64-LABEL: test_ctpop_i2:
+; RV64:       # %bb.0:
+; RV64-NEXT:    addi sp, sp, -16
+; RV64-NEXT:    .cfi_def_cfa_offset 16
+; RV64-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64-NEXT:    .cfi_offset ra, -8
+; RV64-NEXT:    andi a0, a0, 3
+; RV64-NEXT:    srli a1, a0, 1
+; RV64-NEXT:    sub a0, a0, a1
+; RV64-NEXT:    zext.b a1, a0
+; RV64-NEXT:    srli a1, a1, 2
+; RV64-NEXT:    andi a1, a1, 51
+; RV64-NEXT:    andi a0, a0, 51
+; RV64-NEXT:    add a0, a1, a0
+; RV64-NEXT:    srli a1, a0, 4
+; RV64-NEXT:    add a0, a1, a0
+; RV64-NEXT:    andi a0, a0, 15
+; RV64-NEXT:    li a1, 1
+; RV64-NEXT:    call __muldi3
+; RV64-NEXT:    zext.b a0, a0
+; RV64-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64-NEXT:    .cfi_restore ra
+; RV64-NEXT:    addi sp, sp, 16
+; RV64-NEXT:    .cfi_def_cfa_offset 0
+; RV64-NEXT:    ret
+  %1 = call i2 @llvm.ctpop.i2(i2 %a)
+  ret i2 %1
+}
+
+define i11 @test_ctpop_i11(i11 %a) {
+; RV32-LABEL: test_ctpop_i11:
+; RV32:       # %bb.0:
+; RV32-NEXT:    addi sp, sp, -16
+; RV32-NEXT:    .cfi_def_cfa_offset 16
+; RV32-NEXT:    sw ra, 12(sp) # 4-byte Folded Spill
+; RV32-NEXT:    sw s0, 8(sp) # 4-byte Folded Spill
+; RV32-NEXT:    .cfi_offset ra, -4
+; RV32-NEXT:    .cfi_offset s0, -8
+; RV32-NEXT:    andi a0, a0, 2047
+; RV32-NEXT:    lui a1, 5
+; RV32-NEXT:    lui a2, 16
+; RV32-NEXT:    srli a3, a0, 1
+; RV32-NEXT:    addi a1, a1, 1365
+; RV32-NEXT:    and a1, a3, a1
+; RV32-NEXT:    lui a3, 3
+; RV32-NEXT:    addi s0, a2, -1
+; RV32-NEXT:    addi a2, a3, 819
+; RV32-NEXT:    sub a0, a0, a1
+; RV32-NEXT:    and a1, a0, s0
+; RV32-NEXT:    and a0, a0, a2
+; RV32-NEXT:    srli a1, a1, 2
+; RV32-NEXT:    and a1, a1, a2
+; RV32-NEXT:    lui a2, 1
+; RV32-NEXT:    add a0, a1, a0
+; RV32-NEXT:    srli a1, a0, 4
+; RV32-NEXT:    add a0, a1, a0
+; RV32-NEXT:    addi a1, a2, -241
+; RV32-NEXT:    and a0, a0, a1
+; RV32-NEXT:    li a1, 257
+; RV32-NEXT:    call __mulsi3
+; RV32-NEXT:    and a0, a0, s0
+; RV32-NEXT:    srli a0, a0, 8
+; RV32-NEXT:    lw ra, 12(sp) # 4-byte Folded Reload
+; RV32-NEXT:    lw s0, 8(sp) # 4-byte Folded Reload
+; RV32-NEXT:    .cfi_restore ra
+; RV32-NEXT:    .cfi_restore s0
+; RV32-NEXT:    addi sp, sp, 16
+; RV32-NEXT:    .cfi_def_cfa_offset 0
+; RV32-NEXT:    ret
+;
+; RV64-LABEL: test_ctpop_i11:
+; RV64:       # %bb.0:
+; RV64-NEXT:    addi sp, sp, -16
+; RV64-NEXT:    .cfi_def_cfa_offset 16
+; RV64-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; RV64-NEXT:    sd s0, 0(sp) # 8-byte Folded Spill
+; RV64-NEXT:    .cfi_offset ra, -8
+; RV64-NEXT:    .cfi_offset s0, -16
+; RV64-NEXT:    andi a0, a0, 2047
+; RV64-NEXT:    lui a1, 5
+; RV64-NEXT:    lui a2, 16
+; RV64-NEXT:    srli a3, a0, 1
+; RV64-NEXT:    addi a1, a1, 1365
+; RV64-NEXT:    and a1, a3, a1
+; RV64-NEXT:    lui a3, 3
+; RV64-NEXT:    addi s0, a2, -1
+; RV64-NEXT:    addi a2, a3, 819
+; RV64-NEXT:    sub a0, a0, a1
+; RV64-NEXT:    and a1, a0, s0
+; RV64-NEXT:    and a0, a0, a2
+; RV64-NEXT:    srli a1, a1, 2
+; RV64-NEXT:    and a1, a1, a2
+; RV64-NEXT:    lui a2, 1
+; RV64-NEXT:    add a0, a1, a0
+; RV64-NEXT:    srli a1, a0, 4
+; RV64-NEXT:    add a0, a1, a0
+; RV64-NEXT:    addi a1, a2, -241
+; RV64-NEXT:    and a0, a0, a1
+; RV64-NEXT:    li a1, 257
+; RV64-NEXT:    call __muldi3
+; RV64-NEXT:    and a0, a0, s0
+; RV64-NEXT:    srli a0, a0, 8
+; RV64-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; RV64-NEXT:    ld s0, 0(sp) # 8-byte Folded Reload
+; RV64-NEXT:    .cfi_restore ra
+; RV64-NEXT:    .cfi_restore s0
+; RV64-NEXT:    addi sp, sp, 16
+; RV64-NEXT:    .cfi_def_cfa_offset 0
+; RV64-NEXT:    ret
+  %1 = call i11 @llvm.ctpop.i11(i11 %a)
+  ret i11 %1
+}

llvm/test/CodeGen/RISCV/GlobalISel/legalizer/legalize-ctpop-rv64.mir

@@ -216,3 +216,115 @@ body:             |
     PseudoRET implicit $x10
 
 ...
+---
+name:            ctpop_i2
+body:             |
+  bb.1:
+    liveins: $x10
+
+    ; RV64I-LABEL: name: ctpop_i2
+    ; RV64I: liveins: $x10
+    ; RV64I-NEXT: {{  $}}
+    ; RV64I-NEXT: [[COPY:%[0-9]+]]:_(s64) = COPY $x10
+    ; RV64I-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 1
+    ; RV64I-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 3
+    ; RV64I-NEXT: [[AND:%[0-9]+]]:_(s64) = G_AND [[COPY]], [[C1]]
+    ; RV64I-NEXT: [[LSHR:%[0-9]+]]:_(s64) = G_LSHR [[AND]], [[C]](s64)
+    ; RV64I-NEXT: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 85
+    ; RV64I-NEXT: [[AND1:%[0-9]+]]:_(s64) = G_AND [[LSHR]], [[C2]]
+    ; RV64I-NEXT: [[COPY1:%[0-9]+]]:_(s64) = COPY [[AND]](s64)
+    ; RV64I-NEXT: [[SUB:%[0-9]+]]:_(s64) = G_SUB [[COPY1]], [[AND1]]
+    ; RV64I-NEXT: [[C3:%[0-9]+]]:_(s64) = G_CONSTANT i64 2
+    ; RV64I-NEXT: [[C4:%[0-9]+]]:_(s64) = G_CONSTANT i64 255
+    ; RV64I-NEXT: [[AND2:%[0-9]+]]:_(s64) = G_AND [[SUB]], [[C4]]
+    ; RV64I-NEXT: [[LSHR1:%[0-9]+]]:_(s64) = G_LSHR [[AND2]], [[C3]](s64)
+    ; RV64I-NEXT: [[C5:%[0-9]+]]:_(s64) = G_CONSTANT i64 51
+    ; RV64I-NEXT: [[AND3:%[0-9]+]]:_(s64) = G_AND [[LSHR1]], [[C5]]
+    ; RV64I-NEXT: [[AND4:%[0-9]+]]:_(s64) = G_AND [[SUB]], [[C5]]
+    ; RV64I-NEXT: [[ADD:%[0-9]+]]:_(s64) = G_ADD [[AND3]], [[AND4]]
+    ; RV64I-NEXT: [[C6:%[0-9]+]]:_(s64) = G_CONSTANT i64 4
+    ; RV64I-NEXT: [[LSHR2:%[0-9]+]]:_(s64) = G_LSHR [[ADD]], [[C6]](s64)
+    ; RV64I-NEXT: [[ADD1:%[0-9]+]]:_(s64) = G_ADD [[LSHR2]], [[ADD]]
+    ; RV64I-NEXT: [[C7:%[0-9]+]]:_(s64) = G_CONSTANT i64 15
+    ; RV64I-NEXT: [[AND5:%[0-9]+]]:_(s64) = G_AND [[ADD1]], [[C7]]
+    ; RV64I-NEXT: [[C8:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
+    ; RV64I-NEXT: [[MUL:%[0-9]+]]:_(s64) = G_MUL [[AND5]], [[C]]
+    ; RV64I-NEXT: [[LSHR3:%[0-9]+]]:_(s64) = G_LSHR [[MUL]], [[C8]](s64)
+    ; RV64I-NEXT: $x10 = COPY [[LSHR3]](s64)
+    ; RV64I-NEXT: PseudoRET implicit $x10
+    ;
+    ; RV64ZBB-LABEL: name: ctpop_i2
+    ; RV64ZBB: liveins: $x10
+    ; RV64ZBB-NEXT: {{  $}}
+    ; RV64ZBB-NEXT: [[COPY:%[0-9]+]]:_(s64) = COPY $x10
+    ; RV64ZBB-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 3
+    ; RV64ZBB-NEXT: [[AND:%[0-9]+]]:_(s64) = G_AND [[COPY]], [[C]]
+    ; RV64ZBB-NEXT: [[CTPOP:%[0-9]+]]:_(s64) = G_CTPOP [[AND]](s64)
+    ; RV64ZBB-NEXT: [[COPY1:%[0-9]+]]:_(s64) = COPY [[CTPOP]](s64)
+    ; RV64ZBB-NEXT: $x10 = COPY [[COPY1]](s64)
+    ; RV64ZBB-NEXT: PseudoRET implicit $x10
+    %1:_(s64) = COPY $x10
+    %0:_(s2) = G_TRUNC %1(s64)
+    %2:_(s2) = G_CTPOP %0(s2)
+    %3:_(s64) = G_ANYEXT %2(s2)
+    $x10 = COPY %3(s64)
+    PseudoRET implicit $x10
+
+...
+---
+name:            ctpop_i11
+body:             |
+  bb.1:
+    liveins: $x10
+
+    ; RV64I-LABEL: name: ctpop_i11
+    ; RV64I: liveins: $x10
+    ; RV64I-NEXT: {{  $}}
+    ; RV64I-NEXT: [[COPY:%[0-9]+]]:_(s64) = COPY $x10
+    ; RV64I-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 1
+    ; RV64I-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 2047
+    ; RV64I-NEXT: [[AND:%[0-9]+]]:_(s64) = G_AND [[COPY]], [[C1]]
+    ; RV64I-NEXT: [[LSHR:%[0-9]+]]:_(s64) = G_LSHR [[AND]], [[C]](s64)
+    ; RV64I-NEXT: [[C2:%[0-9]+]]:_(s64) = G_CONSTANT i64 21845
+    ; RV64I-NEXT: [[AND1:%[0-9]+]]:_(s64) = G_AND [[LSHR]], [[C2]]
+    ; RV64I-NEXT: [[COPY1:%[0-9]+]]:_(s64) = COPY [[AND]](s64)
+    ; RV64I-NEXT: [[SUB:%[0-9]+]]:_(s64) = G_SUB [[COPY1]], [[AND1]]
+    ; RV64I-NEXT: [[C3:%[0-9]+]]:_(s64) = G_CONSTANT i64 2
+    ; RV64I-NEXT: [[C4:%[0-9]+]]:_(s64) = G_CONSTANT i64 65535
+    ; RV64I-NEXT: [[AND2:%[0-9]+]]:_(s64) = G_AND [[SUB]], [[C4]]
+    ; RV64I-NEXT: [[LSHR1:%[0-9]+]]:_(s64) = G_LSHR [[AND2]], [[C3]](s64)
+    ; RV64I-NEXT: [[C5:%[0-9]+]]:_(s64) = G_CONSTANT i64 13107
+    ; RV64I-NEXT: [[AND3:%[0-9]+]]:_(s64) = G_AND [[LSHR1]], [[C5]]
+    ; RV64I-NEXT: [[AND4:%[0-9]+]]:_(s64) = G_AND [[SUB]], [[C5]]
+    ; RV64I-NEXT: [[ADD:%[0-9]+]]:_(s64) = G_ADD [[AND3]], [[AND4]]
+    ; RV64I-NEXT: [[C6:%[0-9]+]]:_(s64) = G_CONSTANT i64 4
+    ; RV64I-NEXT: [[LSHR2:%[0-9]+]]:_(s64) = G_LSHR [[ADD]], [[C6]](s64)
+    ; RV64I-NEXT: [[ADD1:%[0-9]+]]:_(s64) = G_ADD [[LSHR2]], [[ADD]]
+    ; RV64I-NEXT: [[C7:%[0-9]+]]:_(s64) = G_CONSTANT i64 3855
+    ; RV64I-NEXT: [[AND5:%[0-9]+]]:_(s64) = G_AND [[ADD1]], [[C7]]
+    ; RV64I-NEXT: [[C8:%[0-9]+]]:_(s64) = G_CONSTANT i64 257
+    ; RV64I-NEXT: [[C9:%[0-9]+]]:_(s64) = G_CONSTANT i64 8
+    ; RV64I-NEXT: [[MUL:%[0-9]+]]:_(s64) = G_MUL [[AND5]], [[C8]]
+    ; RV64I-NEXT: [[AND6:%[0-9]+]]:_(s64) = G_AND [[MUL]], [[C4]]
+    ; RV64I-NEXT: [[LSHR3:%[0-9]+]]:_(s64) = G_LSHR [[AND6]], [[C9]](s64)
+    ; RV64I-NEXT: $x10 = COPY [[LSHR3]](s64)
+    ; RV64I-NEXT: PseudoRET implicit $x10
+    ;
+    ; RV64ZBB-LABEL: name: ctpop_i11
+    ; RV64ZBB: liveins: $x10
+    ; RV64ZBB-NEXT: {{  $}}
+    ; RV64ZBB-NEXT: [[COPY:%[0-9]+]]:_(s64) = COPY $x10
+    ; RV64ZBB-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 2047
+    ; RV64ZBB-NEXT: [[AND:%[0-9]+]]:_(s64) = G_AND [[COPY]], [[C]]
+    ; RV64ZBB-NEXT: [[CTPOP:%[0-9]+]]:_(s64) = G_CTPOP [[AND]](s64)
+    ; RV64ZBB-NEXT: [[COPY1:%[0-9]+]]:_(s64) = COPY [[CTPOP]](s64)
+    ; RV64ZBB-NEXT: $x10 = COPY [[COPY1]](s64)
+    ; RV64ZBB-NEXT: PseudoRET implicit $x10
+    %1:_(s64) = COPY $x10
+    %0:_(s11) = G_TRUNC %1(s64)
+    %2:_(s11) = G_CTPOP %0(s11)
+    %3:_(s64) = G_ANYEXT %2(s11)
+    $x10 = COPY %3(s64)
+    PseudoRET implicit $x10
+
+...