motiejus/zig
1
Fork 0
Code Releases Activity

ir: Support div/mod/rem on vector types

Browse Source 
Closes #4050
This commit is contained in:
LemonBoy
2020-03-14 13:23:41 +01:00
committed by Andrew Kelley
parent e2dc63644a
commit 54ffcf95a8
3 changed files with 327 additions and 174 deletions
Download Patch File Download Diff File Expand all files Collapse all files

140
src/codegen.cpp
View File

@@ -2591,12 +2591,7 @@ static LLVMValueRef gen_overflow_shr_op(CodeGen *g, ZigType *type_entry,
}
static LLVMValueRef gen_float_op(CodeGen *g, LLVMValueRef val, ZigType *type_entry, BuiltinFnId op) {
if ((op == BuiltinFnIdCeil ||
op == BuiltinFnIdFloor) &&
type_entry->id == ZigTypeIdInt)
return val;
assert(type_entry->id == ZigTypeIdFloat);
assert(type_entry->id == ZigTypeIdFloat || type_entry->id == ZigTypeIdVector);
LLVMValueRef floor_fn = get_float_fn(g, type_entry, ZigLLVMFnIdFloatOp, op);
return LLVMBuildCall(g->builder, floor_fn, &val, 1, "");
}
@@ -2612,6 +2607,21 @@ static LLVMValueRef bigint_to_llvm_const(LLVMTypeRef type_ref, BigInt *bigint) {
if (bigint->digit_count == 0) {
return LLVMConstNull(type_ref);
}
if (LLVMGetTypeKind(type_ref) == LLVMVectorTypeKind) {
const unsigned vector_len = LLVMGetVectorSize(type_ref);
LLVMTypeRef elem_type = LLVMGetElementType(type_ref);
LLVMValueRef *values = heap::c_allocator.allocate_nonzero<LLVMValueRef>(vector_len);
// Create a vector with all the elements having the same value
for (unsigned i = 0; i < vector_len; i++) {
values[i] = bigint_to_llvm_const(elem_type, bigint);
}
LLVMValueRef result = LLVMConstVector(values, vector_len);
heap::c_allocator.deallocate(values, vector_len);
return result;
}
LLVMValueRef unsigned_val;
if (bigint->digit_count == 1) {
unsigned_val = LLVMConstInt(type_ref, bigint_ptr(bigint)[0], false);
@@ -2625,22 +2635,40 @@ static LLVMValueRef bigint_to_llvm_const(LLVMTypeRef type_ref, BigInt *bigint) {
}
}
// Collapses a <N x i1> vector into a single i1 whose value is 1 iff all the
// vector elements are 1
static LLVMValueRef scalarize_cmp_result(CodeGen *g, LLVMValueRef val) {
assert(LLVMGetTypeKind(LLVMTypeOf(val)) == LLVMVectorTypeKind);
LLVMTypeRef scalar_type = LLVMIntType(LLVMGetVectorSize(LLVMTypeOf(val)));
LLVMValueRef all_ones = LLVMConstAllOnes(scalar_type);
LLVMValueRef casted = LLVMBuildBitCast(g->builder, val, scalar_type, "");
return LLVMBuildICmp(g->builder, LLVMIntEQ, casted, all_ones, "");
}
static LLVMValueRef gen_div(CodeGen *g, bool want_runtime_safety, bool want_fast_math,
LLVMValueRef val1, LLVMValueRef val2,
ZigType *type_entry, DivKind div_kind)
LLVMValueRef val1, LLVMValueRef val2, ZigType *operand_type, DivKind div_kind)
{
ZigType *scalar_type = (operand_type->id == ZigTypeIdVector) ?
operand_type->data.vector.elem_type : operand_type;
ZigLLVMSetFastMath(g->builder, want_fast_math);
LLVMValueRef zero = LLVMConstNull(get_llvm_type(g, type_entry));
if (want_runtime_safety && (want_fast_math || type_entry->id != ZigTypeIdFloat)) {
LLVMValueRef zero = LLVMConstNull(get_llvm_type(g, operand_type));
if (want_runtime_safety && (want_fast_math || scalar_type->id != ZigTypeIdFloat)) {
// Safety check: divisor != 0
LLVMValueRef is_zero_bit;
if (type_entry->id == ZigTypeIdInt) {
if (scalar_type->id == ZigTypeIdInt) {
is_zero_bit = LLVMBuildICmp(g->builder, LLVMIntEQ, val2, zero, "");
} else if (type_entry->id == ZigTypeIdFloat) {
} else if (scalar_type->id == ZigTypeIdFloat) {
is_zero_bit = LLVMBuildFCmp(g->builder, LLVMRealOEQ, val2, zero, "");
} else {
zig_unreachable();
}
if (operand_type->id == ZigTypeIdVector) {
is_zero_bit = scalarize_cmp_result(g, is_zero_bit);
}
LLVMBasicBlockRef div_zero_fail_block = LLVMAppendBasicBlock(g->cur_fn_val, "DivZeroFail");
LLVMBasicBlockRef div_zero_ok_block = LLVMAppendBasicBlock(g->cur_fn_val, "DivZeroOk");
LLVMBuildCondBr(g->builder, is_zero_bit, div_zero_fail_block, div_zero_ok_block);
@@ -2650,16 +2678,21 @@ static LLVMValueRef gen_div(CodeGen *g, bool want_runtime_safety, bool want_fast
LLVMPositionBuilderAtEnd(g->builder, div_zero_ok_block);
if (type_entry->id == ZigTypeIdInt && type_entry->data.integral.is_signed) {
LLVMValueRef neg_1_value = LLVMConstInt(get_llvm_type(g, type_entry), -1, true);
// Safety check: check for overflow (dividend = minInt and divisor = -1)
if (scalar_type->id == ZigTypeIdInt && scalar_type->data.integral.is_signed) {
LLVMValueRef neg_1_value = LLVMConstAllOnes(get_llvm_type(g, operand_type));
BigInt int_min_bi = {0};
eval_min_max_value_int(g, type_entry, &int_min_bi, false);
LLVMValueRef int_min_value = bigint_to_llvm_const(get_llvm_type(g, type_entry), &int_min_bi);
eval_min_max_value_int(g, scalar_type, &int_min_bi, false);
LLVMValueRef int_min_value = bigint_to_llvm_const(get_llvm_type(g, operand_type), &int_min_bi);
LLVMBasicBlockRef overflow_fail_block = LLVMAppendBasicBlock(g->cur_fn_val, "DivOverflowFail");
LLVMBasicBlockRef overflow_ok_block = LLVMAppendBasicBlock(g->cur_fn_val, "DivOverflowOk");
LLVMValueRef num_is_int_min = LLVMBuildICmp(g->builder, LLVMIntEQ, val1, int_min_value, "");
LLVMValueRef den_is_neg_1 = LLVMBuildICmp(g->builder, LLVMIntEQ, val2, neg_1_value, "");
LLVMValueRef overflow_fail_bit = LLVMBuildAnd(g->builder, num_is_int_min, den_is_neg_1, "");
if (operand_type->id == ZigTypeIdVector) {
overflow_fail_bit = scalarize_cmp_result(g, overflow_fail_bit);
}
LLVMBuildCondBr(g->builder, overflow_fail_bit, overflow_fail_block, overflow_ok_block);
LLVMPositionBuilderAtEnd(g->builder, overflow_fail_block);
@@ -2669,18 +2702,22 @@ static LLVMValueRef gen_div(CodeGen *g, bool want_runtime_safety, bool want_fast
}
}
if (type_entry->id == ZigTypeIdFloat) {
if (scalar_type->id == ZigTypeIdFloat) {
LLVMValueRef result = LLVMBuildFDiv(g->builder, val1, val2, "");
switch (div_kind) {
case DivKindFloat:
return result;
case DivKindExact:
if (want_runtime_safety) {
LLVMValueRef floored = gen_float_op(g, result, type_entry, BuiltinFnIdFloor);
// Safety check: a / b == floor(a / b)
LLVMValueRef floored = gen_float_op(g, result, operand_type, BuiltinFnIdFloor);
LLVMBasicBlockRef ok_block = LLVMAppendBasicBlock(g->cur_fn_val, "DivExactOk");
LLVMBasicBlockRef fail_block = LLVMAppendBasicBlock(g->cur_fn_val, "DivExactFail");
LLVMValueRef ok_bit = LLVMBuildFCmp(g->builder, LLVMRealOEQ, floored, result, "");
if (operand_type->id == ZigTypeIdVector) {
ok_bit = scalarize_cmp_result(g, ok_bit);
}
LLVMBuildCondBr(g->builder, ok_bit, ok_block, fail_block);
LLVMPositionBuilderAtEnd(g->builder, fail_block);
@@ -2695,54 +2732,61 @@ static LLVMValueRef gen_div(CodeGen *g, bool want_runtime_safety, bool want_fast
LLVMBasicBlockRef gez_block = LLVMAppendBasicBlock(g->cur_fn_val, "DivTruncGEZero");
LLVMBasicBlockRef end_block = LLVMAppendBasicBlock(g->cur_fn_val, "DivTruncEnd");
LLVMValueRef ltz = LLVMBuildFCmp(g->builder, LLVMRealOLT, val1, zero, "");
if (operand_type->id == ZigTypeIdVector) {
ltz = scalarize_cmp_result(g, ltz);
}
LLVMBuildCondBr(g->builder, ltz, ltz_block, gez_block);
LLVMPositionBuilderAtEnd(g->builder, ltz_block);
LLVMValueRef ceiled = gen_float_op(g, result, type_entry, BuiltinFnIdCeil);
LLVMValueRef ceiled = gen_float_op(g, result, operand_type, BuiltinFnIdCeil);
LLVMBasicBlockRef ceiled_end_block = LLVMGetInsertBlock(g->builder);
LLVMBuildBr(g->builder, end_block);
LLVMPositionBuilderAtEnd(g->builder, gez_block);
LLVMValueRef floored = gen_float_op(g, result, type_entry, BuiltinFnIdFloor);
LLVMValueRef floored = gen_float_op(g, result, operand_type, BuiltinFnIdFloor);
LLVMBasicBlockRef floored_end_block = LLVMGetInsertBlock(g->builder);
LLVMBuildBr(g->builder, end_block);
LLVMPositionBuilderAtEnd(g->builder, end_block);
LLVMValueRef phi = LLVMBuildPhi(g->builder, get_llvm_type(g, type_entry), "");
LLVMValueRef phi = LLVMBuildPhi(g->builder, get_llvm_type(g, operand_type), "");
LLVMValueRef incoming_values[] = { ceiled, floored };
LLVMBasicBlockRef incoming_blocks[] = { ceiled_end_block, floored_end_block };
LLVMAddIncoming(phi, incoming_values, incoming_blocks, 2);
return phi;
}
case DivKindFloor:
return gen_float_op(g, result, type_entry, BuiltinFnIdFloor);
return gen_float_op(g, result, operand_type, BuiltinFnIdFloor);
}
zig_unreachable();
}
assert(type_entry->id == ZigTypeIdInt);
assert(scalar_type->id == ZigTypeIdInt);
switch (div_kind) {
case DivKindFloat:
zig_unreachable();
case DivKindTrunc:
if (type_entry->data.integral.is_signed) {
if (scalar_type->data.integral.is_signed) {
return LLVMBuildSDiv(g->builder, val1, val2, "");
} else {
return LLVMBuildUDiv(g->builder, val1, val2, "");
}
case DivKindExact:
if (want_runtime_safety) {
// Safety check: a % b == 0
LLVMValueRef remainder_val;
if (type_entry->data.integral.is_signed) {
if (scalar_type->data.integral.is_signed) {
remainder_val = LLVMBuildSRem(g->builder, val1, val2, "");
} else {
remainder_val = LLVMBuildURem(g->builder, val1, val2, "");
}
LLVMValueRef ok_bit = LLVMBuildICmp(g->builder, LLVMIntEQ, remainder_val, zero, "");
LLVMBasicBlockRef ok_block = LLVMAppendBasicBlock(g->cur_fn_val, "DivExactOk");
LLVMBasicBlockRef fail_block = LLVMAppendBasicBlock(g->cur_fn_val, "DivExactFail");
LLVMValueRef ok_bit = LLVMBuildICmp(g->builder, LLVMIntEQ, remainder_val, zero, "");
if (operand_type->id == ZigTypeIdVector) {
ok_bit = scalarize_cmp_result(g, ok_bit);
}
LLVMBuildCondBr(g->builder, ok_bit, ok_block, fail_block);
LLVMPositionBuilderAtEnd(g->builder, fail_block);
@@ -2750,14 +2794,14 @@ static LLVMValueRef gen_div(CodeGen *g, bool want_runtime_safety, bool want_fast
LLVMPositionBuilderAtEnd(g->builder, ok_block);
}
if (type_entry->data.integral.is_signed) {
if (scalar_type->data.integral.is_signed) {
return LLVMBuildExactSDiv(g->builder, val1, val2, "");
} else {
return LLVMBuildExactUDiv(g->builder, val1, val2, "");
}
case DivKindFloor:
{
if (!type_entry->data.integral.is_signed) {
if (!scalar_type->data.integral.is_signed) {
return LLVMBuildUDiv(g->builder, val1, val2, "");
}
// const d = @divTrunc(a, b);
@@ -2784,22 +2828,30 @@ enum RemKind {
};
static LLVMValueRef gen_rem(CodeGen *g, bool want_runtime_safety, bool want_fast_math,
LLVMValueRef val1, LLVMValueRef val2,
ZigType *type_entry, RemKind rem_kind)
LLVMValueRef val1, LLVMValueRef val2, ZigType *operand_type, RemKind rem_kind)
{
ZigType *scalar_type = (operand_type->id == ZigTypeIdVector) ?
operand_type->data.vector.elem_type : operand_type;
ZigLLVMSetFastMath(g->builder, want_fast_math);
LLVMValueRef zero = LLVMConstNull(get_llvm_type(g, type_entry));
LLVMValueRef zero = LLVMConstNull(get_llvm_type(g, operand_type));
if (want_runtime_safety) {
// Safety check: divisor != 0
LLVMValueRef is_zero_bit;
if (type_entry->id == ZigTypeIdInt) {
LLVMIntPredicate pred = type_entry->data.integral.is_signed ? LLVMIntSLE : LLVMIntEQ;
if (scalar_type->id == ZigTypeIdInt) {
LLVMIntPredicate pred = scalar_type->data.integral.is_signed ? LLVMIntSLE : LLVMIntEQ;
is_zero_bit = LLVMBuildICmp(g->builder, pred, val2, zero, "");
} else if (type_entry->id == ZigTypeIdFloat) {
} else if (scalar_type->id == ZigTypeIdFloat) {
is_zero_bit = LLVMBuildFCmp(g->builder, LLVMRealOEQ, val2, zero, "");
} else {
zig_unreachable();
}
if (operand_type->id == ZigTypeIdVector) {
is_zero_bit = scalarize_cmp_result(g, is_zero_bit);
}
LLVMBasicBlockRef rem_zero_ok_block = LLVMAppendBasicBlock(g->cur_fn_val, "RemZeroOk");
LLVMBasicBlockRef rem_zero_fail_block = LLVMAppendBasicBlock(g->cur_fn_val, "RemZeroFail");
LLVMBuildCondBr(g->builder, is_zero_bit, rem_zero_fail_block, rem_zero_ok_block);
@@ -2810,7 +2862,7 @@ static LLVMValueRef gen_rem(CodeGen *g, bool want_runtime_safety, bool want_fast
LLVMPositionBuilderAtEnd(g->builder, rem_zero_ok_block);
}
if (type_entry->id == ZigTypeIdFloat) {
if (scalar_type->id == ZigTypeIdFloat) {
if (rem_kind == RemKindRem) {
return LLVMBuildFRem(g->builder, val1, val2, "");
} else {
@@ -2821,8 +2873,8 @@ static LLVMValueRef gen_rem(CodeGen *g, bool want_runtime_safety, bool want_fast
return LLVMBuildSelect(g->builder, ltz, c, a, "");
}
} else {
assert(type_entry->id == ZigTypeIdInt);
if (type_entry->data.integral.is_signed) {
assert(scalar_type->id == ZigTypeIdInt);
if (scalar_type->data.integral.is_signed) {
if (rem_kind == RemKindRem) {
return LLVMBuildSRem(g->builder, val1, val2, "");
} else {
@@ -3010,22 +3062,22 @@ static LLVMValueRef ir_render_bin_op(CodeGen *g, IrExecutableGen *executable,
}
case IrBinOpDivUnspecified:
return gen_div(g, want_runtime_safety, ir_want_fast_math(g, &bin_op_instruction->base),
op1_value, op2_value, scalar_type, DivKindFloat);
op1_value, op2_value, operand_type, DivKindFloat);
case IrBinOpDivExact:
return gen_div(g, want_runtime_safety, ir_want_fast_math(g, &bin_op_instruction->base),
op1_value, op2_value, scalar_type, DivKindExact);
op1_value, op2_value, operand_type, DivKindExact);
case IrBinOpDivTrunc:
return gen_div(g, want_runtime_safety, ir_want_fast_math(g, &bin_op_instruction->base),
op1_value, op2_value, scalar_type, DivKindTrunc);
op1_value, op2_value, operand_type, DivKindTrunc);
case IrBinOpDivFloor:
return gen_div(g, want_runtime_safety, ir_want_fast_math(g, &bin_op_instruction->base),
op1_value, op2_value, scalar_type, DivKindFloor);
op1_value, op2_value, operand_type, DivKindFloor);
case IrBinOpRemRem:
return gen_rem(g, want_runtime_safety, ir_want_fast_math(g, &bin_op_instruction->base),
op1_value, op2_value, scalar_type, RemKindRem);
op1_value, op2_value, operand_type, RemKindRem);
case IrBinOpRemMod:
return gen_rem(g, want_runtime_safety, ir_want_fast_math(g, &bin_op_instruction->base),
op1_value, op2_value, scalar_type, RemKindMod);
op1_value, op2_value, operand_type, RemKindMod);
}
zig_unreachable();
}