[X86] Add more patterns for BZHI isel

[craig.topper]

Summary

This patch adds more patterns that a reasonable person might write that can be compiled to BZHI.

This adds support for

(~0U >> (32 - b)) & a;

and

a << (32 - b) >> (32 - b);

This was inspired by the code in APInt::clearUnusedBits.

This can pass an index of 32 to the bzhi instruction which a quick test of Haswell hardware shows will not mask any bits. Though the description text in the Intel manual says the "index is saturated to OperandSize-1". The pseudocode in the same manual indicates no bits will be zeroed for this case.

I think this is still missing cases where the subtract portion is an 8-bit operation.

[RKSimon]

According to the AMD APM v3:

If the value of index is greater than or equal to the operand size, index is set to (op_size-1). In this case, the CF flag is set.

[craig.topper]

So AMD's description is similar to the incorrect Intel description. Do you have an AMD machine that you can verify this with?

I tested with

printf("%x", _bzhi_u32(0xffffffff, 32));

On Haswell that returned 0xffffffff.

[RKSimon]

In D32616#741875, @craig.topper wrote:

So AMD's description is similar to the incorrect Intel description. Do you have an AMD machine that you can verify this with?

I tested with

printf("%x", _bzhi_u32(0xffffffff, 32));

On Haswell that returned 0xffffffff.

Excavator returned 0xffffffff as well. The Ryzen is in pieces so I can't test it right now.

[RKSimon]

Ryzen returns 0xffffffff as well.

LGTM.

llvm/trunk/lib/Target/X86/X86InstrInfo.td

let Predicates = [HasBMI2] in {
 
   def : Pat<(and GR64:$src, (add (shl 1, GR8:$lz), -1)),
             (BZHI64rr GR64:$src,
               (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GR8:$lz, sub_8bit))>;
 
   def : Pat<(and (loadi64 addr:$src), (add (shl 1, GR8:$lz), -1)),
             (BZHI64rm addr:$src,
               (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GR8:$lz, sub_8bit))>;
+
+  // x & (-1 >> (32 - y))
+  def : Pat<(and GR32:$src, (srl -1, (i8 (trunc (sub 32, GR32:$lz))))),
+            (BZHI32rr GR32:$src, GR32:$lz)>;
+  def : Pat<(and (loadi32 addr:$src), (srl -1, (i8 (trunc (sub 32, GR32:$lz))))),
+            (BZHI32rm addr:$src, GR32:$lz)>;
+
+  // x & (-1 >> (64 - y))
+  def : Pat<(and GR64:$src, (srl -1, (i8 (trunc (sub 64, GR32:$lz))))),
+            (BZHI64rr GR64:$src,
+              (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GR32:$lz, sub_32bit))>;
+  def : Pat<(and (loadi64 addr:$src), (srl -1, (i8 (trunc (sub 64, GR32:$lz))))),
+            (BZHI64rm addr:$src,
+              (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GR32:$lz, sub_32bit))>;
+
+  // x << (32 - y) >> (32 - y)
+  def : Pat<(srl (shl GR32:$src, (i8 (trunc (sub 32, GR32:$lz)))),
+                 (i8 (trunc (sub 32, GR32:$lz)))),
+            (BZHI32rr GR32:$src, GR32:$lz)>;
+  def : Pat<(srl (shl (loadi32 addr:$src), (i8 (trunc (sub 32, GR32:$lz)))),
+                 (i8 (trunc (sub 32, GR32:$lz)))),
+            (BZHI32rm addr:$src, GR32:$lz)>;
+
+  // x << (64 - y) >> (64 - y)
+  def : Pat<(srl (shl GR64:$src, (i8 (trunc (sub 64, GR32:$lz)))),
+                 (i8 (trunc (sub 64, GR32:$lz)))),
+            (BZHI64rr GR64:$src,
+              (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GR32:$lz, sub_32bit))>;
+  def : Pat<(srl (shl (loadi64 addr:$src), (i8 (trunc (sub 64, GR32:$lz)))),
+                 (i8 (trunc (sub 64, GR32:$lz)))),
+            (BZHI64rm addr:$src,
+              (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GR32:$lz, sub_32bit))>;
 } // HasBMI2
 
 let Predicates = [HasBMI] in {
   def : Pat<(X86bextr GR32:$src1, GR32:$src2),
             (BEXTR32rr GR32:$src1, GR32:$src2)>;
   def : Pat<(X86bextr (loadi32 addr:$src1), GR32:$src2),
             (BEXTR32rm addr:$src1, GR32:$src2)>;
   def : Pat<(X86bextr GR64:$src1, GR64:$src2),

llvm/trunk/test/CodeGen/X86/bmi.ll

 entry:
   %conv = zext i8 %index to i32
   %shl = shl i32 1, %conv
   %sub = add nsw i32 %shl, -1
   %and = and i32 %x, %sub
   ret i32 %and
 }
 
+define i32 @bzhi32d(i32 %a, i32 %b) {
+; CHECK-LABEL: bzhi32d:
+; CHECK:       # BB#0: # %entry
+; CHECK-NEXT:    bzhil %esi, %edi, %eax
+; CHECK-NEXT:    retq
+entry:
+  %sub = sub i32 32, %b
+  %shr = lshr i32 -1, %sub
+  %and = and i32 %shr, %a
+  ret i32 %and
+}
+
+define i32 @bzhi32e(i32 %a, i32 %b) {
+; CHECK-LABEL: bzhi32e:
+; CHECK:       # BB#0: # %entry
+; CHECK-NEXT:    bzhil %esi, %edi, %eax
+; CHECK-NEXT:    retq
+entry:
+  %sub = sub i32 32, %b
+  %shl = shl i32 %a, %sub
+  %shr = lshr i32 %shl, %sub
+  ret i32 %shr
+}
+
 define i64 @bzhi64b(i64 %x, i8 zeroext %index) {
 ; CHECK-LABEL: bzhi64b:
 ; CHECK:       # BB#0: # %entry
 ; CHECK-NEXT:    # kill: %ESI<def> %ESI<kill> %RSI<def>
 ; CHECK-NEXT:    bzhiq %rsi, %rdi, %rax
 ; CHECK-NEXT:    retq
 entry:
   %conv = zext i8 %index to i64
   %shl = shl i64 1, %conv
   %sub = add nsw i64 %shl, -1
   %and = and i64 %x, %sub
   ret i64 %and
 }
 
+define i64 @bzhi64c(i64 %a, i64 %b) {
+; CHECK-LABEL: bzhi64c:
+; CHECK:       # BB#0: # %entry
+; CHECK-NEXT:    bzhiq %rsi, %rdi, %rax
+; CHECK-NEXT:    retq
+entry:
+  %sub = sub i64 64, %b
+  %shr = lshr i64 -1, %sub
+  %and = and i64 %shr, %a
+  ret i64 %and
+}
+
+define i64 @bzhi64d(i64 %a, i32 %b) {
+; CHECK-LABEL: bzhi64d:
+; CHECK:       # BB#0: # %entry
+; CHECK-NEXT:    # kill: %ESI<def> %ESI<kill> %RSI<def>
+; CHECK-NEXT:    bzhiq %rsi, %rdi, %rax
+; CHECK-NEXT:    retq
+entry:
+  %sub = sub i32 64, %b
+  %sh_prom = zext i32 %sub to i64
+  %shr = lshr i64 -1, %sh_prom
+  %and = and i64 %shr, %a
+  ret i64 %and
+}
+
+define i64 @bzhi64e(i64 %a, i64 %b) {
+; CHECK-LABEL: bzhi64e:
+; CHECK:       # BB#0: # %entry
+; CHECK-NEXT:    bzhiq %rsi, %rdi, %rax
+; CHECK-NEXT:    retq
+entry:
+  %sub = sub i64 64, %b
+  %shl = shl i64 %a, %sub
+  %shr = lshr i64 %shl, %sub
+  ret i64 %shr
+}
+
+define i64 @bzhi64f(i64 %a, i32 %b) {
+; CHECK-LABEL: bzhi64f:
+; CHECK:       # BB#0: # %entry
+; CHECK-NEXT:    # kill: %ESI<def> %ESI<kill> %RSI<def>
+; CHECK-NEXT:    bzhiq %rsi, %rdi, %rax
+; CHECK-NEXT:    retq
+entry:
+  %sub = sub i32 64, %b
+  %sh_prom = zext i32 %sub to i64
+  %shl = shl i64 %a, %sh_prom
+  %shr = lshr i64 %shl, %sh_prom
+  ret i64 %shr
+}
+
 define i64 @bzhi64_constant_mask(i64 %x) {
 ; CHECK-LABEL: bzhi64_constant_mask:
 ; CHECK:       # BB#0: # %entry
 ; CHECK-NEXT:    movb $62, %al
 ; CHECK-NEXT:    bzhiq %rax, %rdi, %rax
 ; CHECK-NEXT:    retq
 entry:
   %and = and i64 %x, 4611686018427387903