From: Nicolai Hähnle <nicolai.haeh...@amd.com>
---
src/amd/common/ac_nir_to_llvm.c | 235 ++++++++++++++++++++--------------------
1 file changed, 118 insertions(+), 117 deletions(-)
diff --git a/src/amd/common/ac_nir_to_llvm.c b/src/amd/common/ac_nir_to_llvm.c
index a590888..0b1eea0 100644
--- a/src/amd/common/ac_nir_to_llvm.c
+++ b/src/amd/common/ac_nir_to_llvm.c
@@ -1097,201 +1097,201 @@ static LLVMValueRef get_alu_src(struct
nir_to_llvm_context *ctx,
LLVMValueRef swizzle = LLVMConstVector(masks,
num_components);
value = LLVMBuildShuffleVector(ctx->builder, value,
value,
swizzle, "");
}
}
assert(!src.negate);
assert(!src.abs);
return value;
}
-static LLVMValueRef emit_int_cmp(struct nir_to_llvm_context *ctx,
+static LLVMValueRef emit_int_cmp(struct ac_llvm_context *ctx,
LLVMIntPredicate pred, LLVMValueRef src0,
LLVMValueRef src1)
{
LLVMValueRef result = LLVMBuildICmp(ctx->builder, pred, src0, src1, "");
return LLVMBuildSelect(ctx->builder, result,
LLVMConstInt(ctx->i32, 0xFFFFFFFF, false),
LLVMConstInt(ctx->i32, 0, false), "");
}
-static LLVMValueRef emit_float_cmp(struct nir_to_llvm_context *ctx,
+static LLVMValueRef emit_float_cmp(struct ac_llvm_context *ctx,
LLVMRealPredicate pred, LLVMValueRef src0,
LLVMValueRef src1)
{
LLVMValueRef result;
- src0 = to_float(&ctx->ac, src0);
- src1 = to_float(&ctx->ac, src1);
+ src0 = to_float(ctx, src0);
+ src1 = to_float(ctx, src1);
result = LLVMBuildFCmp(ctx->builder, pred, src0, src1, "");
return LLVMBuildSelect(ctx->builder, result,
LLVMConstInt(ctx->i32, 0xFFFFFFFF, false),
LLVMConstInt(ctx->i32, 0, false), "");
}
-static LLVMValueRef emit_intrin_1f_param(struct nir_to_llvm_context *ctx,
+static LLVMValueRef emit_intrin_1f_param(struct ac_llvm_context *ctx,
const char *intrin,
LLVMTypeRef result_type,
LLVMValueRef src0)
{
char name[64];
LLVMValueRef params[] = {
- to_float(&ctx->ac, src0),
+ to_float(ctx, src0),
};
- sprintf(name, "%s.f%d", intrin, get_elem_bits(&ctx->ac, result_type));
- return ac_build_intrinsic(&ctx->ac, name, result_type, params, 1,
AC_FUNC_ATTR_READNONE);
+ sprintf(name, "%s.f%d", intrin, get_elem_bits(ctx, result_type));
+ return ac_build_intrinsic(ctx, name, result_type, params, 1,
AC_FUNC_ATTR_READNONE);
}
-static LLVMValueRef emit_intrin_2f_param(struct nir_to_llvm_context *ctx,
+static LLVMValueRef emit_intrin_2f_param(struct ac_llvm_context *ctx,
const char *intrin,
LLVMTypeRef result_type,
LLVMValueRef src0, LLVMValueRef src1)
{
char name[64];
LLVMValueRef params[] = {
- to_float(&ctx->ac, src0),
- to_float(&ctx->ac, src1),
+ to_float(ctx, src0),
+ to_float(ctx, src1),
};
- sprintf(name, "%s.f%d", intrin, get_elem_bits(&ctx->ac, result_type));
- return ac_build_intrinsic(&ctx->ac, name, result_type, params, 2,
AC_FUNC_ATTR_READNONE);
+ sprintf(name, "%s.f%d", intrin, get_elem_bits(ctx, result_type));
+ return ac_build_intrinsic(ctx, name, result_type, params, 2,
AC_FUNC_ATTR_READNONE);
}
-static LLVMValueRef emit_intrin_3f_param(struct nir_to_llvm_context *ctx,
+static LLVMValueRef emit_intrin_3f_param(struct ac_llvm_context *ctx,
const char *intrin,
LLVMTypeRef result_type,
LLVMValueRef src0, LLVMValueRef src1,
LLVMValueRef src2)
{
char name[64];
LLVMValueRef params[] = {
- to_float(&ctx->ac, src0),
- to_float(&ctx->ac, src1),
- to_float(&ctx->ac, src2),
+ to_float(ctx, src0),
+ to_float(ctx, src1),
+ to_float(ctx, src2),
};
- sprintf(name, "%s.f%d", intrin, get_elem_bits(&ctx->ac, result_type));
- return ac_build_intrinsic(&ctx->ac, name, result_type, params, 3,
AC_FUNC_ATTR_READNONE);
+ sprintf(name, "%s.f%d", intrin, get_elem_bits(ctx, result_type));
+ return ac_build_intrinsic(ctx, name, result_type, params, 3,
AC_FUNC_ATTR_READNONE);
}
-static LLVMValueRef emit_bcsel(struct nir_to_llvm_context *ctx,
+static LLVMValueRef emit_bcsel(struct ac_llvm_context *ctx,
LLVMValueRef src0, LLVMValueRef src1,
LLVMValueRef src2)
{
LLVMValueRef v = LLVMBuildICmp(ctx->builder, LLVMIntNE, src0,
- ctx->i32zero, "");
+ ctx->i32_0, "");
return LLVMBuildSelect(ctx->builder, v, src1, src2, "");
}
-static LLVMValueRef emit_find_lsb(struct nir_to_llvm_context *ctx,
+static LLVMValueRef emit_find_lsb(struct ac_llvm_context *ctx,
LLVMValueRef src0)
{
LLVMValueRef params[2] = {
src0,
/* The value of 1 means that ffs(x=0) = undef, so LLVM won't
* add special code to check for x=0. The reason is that
* the LLVM behavior for x=0 is different from what we
* need here.
*
* The hardware already implements the correct behavior.
*/
LLVMConstInt(ctx->i1, 1, false),
};
- return ac_build_intrinsic(&ctx->ac, "llvm.cttz.i32", ctx->i32, params,
2, AC_FUNC_ATTR_READNONE);
+ return ac_build_intrinsic(ctx, "llvm.cttz.i32", ctx->i32, params, 2,
AC_FUNC_ATTR_READNONE);
}
-static LLVMValueRef emit_ifind_msb(struct nir_to_llvm_context *ctx,
+static LLVMValueRef emit_ifind_msb(struct ac_llvm_context *ctx,
LLVMValueRef src0)
{
- return ac_build_imsb(&ctx->ac, src0, ctx->i32);
+ return ac_build_imsb(ctx, src0, ctx->i32);
}
-static LLVMValueRef emit_ufind_msb(struct nir_to_llvm_context *ctx,
+static LLVMValueRef emit_ufind_msb(struct ac_llvm_context *ctx,
LLVMValueRef src0)
{
- return ac_build_umsb(&ctx->ac, src0, ctx->i32);
+ return ac_build_umsb(ctx, src0, ctx->i32);
}
-static LLVMValueRef emit_minmax_int(struct nir_to_llvm_context *ctx,
+static LLVMValueRef emit_minmax_int(struct ac_llvm_context *ctx,
LLVMIntPredicate pred,
LLVMValueRef src0, LLVMValueRef src1)
{
return LLVMBuildSelect(ctx->builder,
LLVMBuildICmp(ctx->builder, pred, src0, src1,
""),
src0,
src1, "");
}
-static LLVMValueRef emit_iabs(struct nir_to_llvm_context *ctx,
+static LLVMValueRef emit_iabs(struct ac_llvm_context *ctx,
LLVMValueRef src0)
{
return emit_minmax_int(ctx, LLVMIntSGT, src0,
LLVMBuildNeg(ctx->builder, src0, ""));
}
-static LLVMValueRef emit_fsign(struct nir_to_llvm_context *ctx,
+static LLVMValueRef emit_fsign(struct ac_llvm_context *ctx,
LLVMValueRef src0)
{
LLVMValueRef cmp, val;
- cmp = LLVMBuildFCmp(ctx->builder, LLVMRealOGT, src0, ctx->f32zero, "");
- val = LLVMBuildSelect(ctx->builder, cmp, ctx->f32one, src0, "");
- cmp = LLVMBuildFCmp(ctx->builder, LLVMRealOGE, val, ctx->f32zero, "");
+ cmp = LLVMBuildFCmp(ctx->builder, LLVMRealOGT, src0, ctx->f32_0, "");
+ val = LLVMBuildSelect(ctx->builder, cmp, ctx->f32_1, src0, "");
+ cmp = LLVMBuildFCmp(ctx->builder, LLVMRealOGE, val, ctx->f32_0, "");
val = LLVMBuildSelect(ctx->builder, cmp, val, LLVMConstReal(ctx->f32,
-1.0), "");
return val;
}
-static LLVMValueRef emit_isign(struct nir_to_llvm_context *ctx,
+static LLVMValueRef emit_isign(struct ac_llvm_context *ctx,
LLVMValueRef src0)
{
LLVMValueRef cmp, val;
- cmp = LLVMBuildICmp(ctx->builder, LLVMIntSGT, src0, ctx->i32zero, "");
- val = LLVMBuildSelect(ctx->builder, cmp, ctx->i32one, src0, "");
- cmp = LLVMBuildICmp(ctx->builder, LLVMIntSGE, val, ctx->i32zero, "");
+ cmp = LLVMBuildICmp(ctx->builder, LLVMIntSGT, src0, ctx->i32_0, "");
+ val = LLVMBuildSelect(ctx->builder, cmp, ctx->i32_1, src0, "");
+ cmp = LLVMBuildICmp(ctx->builder, LLVMIntSGE, val, ctx->i32_0, "");
val = LLVMBuildSelect(ctx->builder, cmp, val, LLVMConstInt(ctx->i32,
-1, true), "");
return val;
}
-static LLVMValueRef emit_ffract(struct nir_to_llvm_context *ctx,
+static LLVMValueRef emit_ffract(struct ac_llvm_context *ctx,
LLVMValueRef src0)
{
const char *intr = "llvm.floor.f32";
- LLVMValueRef fsrc0 = to_float(&ctx->ac, src0);
+ LLVMValueRef fsrc0 = to_float(ctx, src0);
LLVMValueRef params[] = {
fsrc0,
};
- LLVMValueRef floor = ac_build_intrinsic(&ctx->ac, intr,
+ LLVMValueRef floor = ac_build_intrinsic(ctx, intr,
ctx->f32, params, 1,
AC_FUNC_ATTR_READNONE);
return LLVMBuildFSub(ctx->builder, fsrc0, floor, "");
}
-static LLVMValueRef emit_uint_carry(struct nir_to_llvm_context *ctx,
+static LLVMValueRef emit_uint_carry(struct ac_llvm_context *ctx,
const char *intrin,
LLVMValueRef src0, LLVMValueRef src1)
{
LLVMTypeRef ret_type;
LLVMTypeRef types[] = { ctx->i32, ctx->i1 };
LLVMValueRef res;
LLVMValueRef params[] = { src0, src1 };
ret_type = LLVMStructTypeInContext(ctx->context, types,
2, true);
- res = ac_build_intrinsic(&ctx->ac, intrin, ret_type,
+ res = ac_build_intrinsic(ctx, intrin, ret_type,
params, 2, AC_FUNC_ATTR_READNONE);
res = LLVMBuildExtractValue(ctx->builder, res, 1, "");
res = LLVMBuildZExt(ctx->builder, res, ctx->i32, "");
return res;
}
-static LLVMValueRef emit_b2f(struct nir_to_llvm_context *ctx,
+static LLVMValueRef emit_b2f(struct ac_llvm_context *ctx,
LLVMValueRef src0)
{
return LLVMBuildAnd(ctx->builder, src0, LLVMBuildBitCast(ctx->builder,
LLVMConstReal(ctx->f32, 1.0), ctx->i32, ""), "");
}
static LLVMValueRef emit_f2f16(struct nir_to_llvm_context *ctx,
LLVMValueRef src0)
{
LLVMValueRef result;
LLVMValueRef cond;
@@ -1310,128 +1310,129 @@ static LLVMValueRef emit_f2f16(struct
nir_to_llvm_context *ctx,
/* need to convert back up to f32 */
result = LLVMBuildFPExt(ctx->builder, result, ctx->f32, "");
if (ctx->options->chip_class >= VI)
result = LLVMBuildSelect(ctx->builder, cond, ctx->f32zero,
result, "");
return result;
}
-static LLVMValueRef emit_umul_high(struct nir_to_llvm_context *ctx,
+static LLVMValueRef emit_umul_high(struct ac_llvm_context *ctx,
LLVMValueRef src0, LLVMValueRef src1)
{
LLVMValueRef dst64, result;
src0 = LLVMBuildZExt(ctx->builder, src0, ctx->i64, "");
src1 = LLVMBuildZExt(ctx->builder, src1, ctx->i64, "");
dst64 = LLVMBuildMul(ctx->builder, src0, src1, "");
dst64 = LLVMBuildLShr(ctx->builder, dst64, LLVMConstInt(ctx->i64, 32,
false), "");
result = LLVMBuildTrunc(ctx->builder, dst64, ctx->i32, "");
return result;
}
-static LLVMValueRef emit_imul_high(struct nir_to_llvm_context *ctx,
+static LLVMValueRef emit_imul_high(struct ac_llvm_context *ctx,
LLVMValueRef src0, LLVMValueRef src1)
{
LLVMValueRef dst64, result;
src0 = LLVMBuildSExt(ctx->builder, src0, ctx->i64, "");
src1 = LLVMBuildSExt(ctx->builder, src1, ctx->i64, "");
dst64 = LLVMBuildMul(ctx->builder, src0, src1, "");
dst64 = LLVMBuildAShr(ctx->builder, dst64, LLVMConstInt(ctx->i64, 32,
false), "");
result = LLVMBuildTrunc(ctx->builder, dst64, ctx->i32, "");
return result;
}
-static LLVMValueRef emit_bitfield_extract(struct nir_to_llvm_context *ctx,
+static LLVMValueRef emit_bitfield_extract(struct ac_llvm_context *ctx,
bool is_signed,
const LLVMValueRef srcs[3])
{
LLVMValueRef result;
LLVMValueRef icond = LLVMBuildICmp(ctx->builder, LLVMIntEQ, srcs[2],
LLVMConstInt(ctx->i32, 32, false), "");
- result = ac_build_bfe(&ctx->ac, srcs[0], srcs[1], srcs[2], is_signed);
+ result = ac_build_bfe(ctx, srcs[0], srcs[1], srcs[2], is_signed);
result = LLVMBuildSelect(ctx->builder, icond, srcs[0], result, "");
return result;
}
-static LLVMValueRef emit_bitfield_insert(struct nir_to_llvm_context *ctx,
+static LLVMValueRef emit_bitfield_insert(struct ac_llvm_context *ctx,
LLVMValueRef src0, LLVMValueRef src1,
LLVMValueRef src2, LLVMValueRef src3)
{
LLVMValueRef bfi_args[3], result;
bfi_args[0] = LLVMBuildShl(ctx->builder,
LLVMBuildSub(ctx->builder,
LLVMBuildShl(ctx->builder,
- ctx->i32one,
+ ctx->i32_1,
src3, ""),
- ctx->i32one, ""),
+ ctx->i32_1, ""),
src2, "");
bfi_args[1] = LLVMBuildShl(ctx->builder, src1, src2, "");
bfi_args[2] = src0;
LLVMValueRef icond = LLVMBuildICmp(ctx->builder, LLVMIntEQ, src3,
LLVMConstInt(ctx->i32, 32, false), "");
/* Calculate:
* (arg0 & arg1) | (~arg0 & arg2) = arg2 ^ (arg0 & (arg1 ^ arg2)
* Use the right-hand side, which the LLVM backend can convert to V_BFI.
*/
result = LLVMBuildXor(ctx->builder, bfi_args[2],
LLVMBuildAnd(ctx->builder, bfi_args[0],
LLVMBuildXor(ctx->builder,
bfi_args[1], bfi_args[2], ""), ""), "");
result = LLVMBuildSelect(ctx->builder, icond, src1, result, "");
return result;
}
-static LLVMValueRef emit_pack_half_2x16(struct nir_to_llvm_context *ctx,
+static LLVMValueRef emit_pack_half_2x16(struct ac_llvm_context *ctx,
LLVMValueRef src0)
{
LLVMValueRef const16 = LLVMConstInt(ctx->i32, 16, false);
int i;
LLVMValueRef comp[2];
- src0 = to_float(&ctx->ac, src0);
- comp[0] = LLVMBuildExtractElement(ctx->builder, src0, ctx->i32zero, "");
- comp[1] = LLVMBuildExtractElement(ctx->builder, src0, ctx->i32one, "");
+ src0 = to_float(ctx, src0);
+ comp[0] = LLVMBuildExtractElement(ctx->builder, src0, ctx->i32_0, "");
+ comp[1] = LLVMBuildExtractElement(ctx->builder, src0, ctx->i32_1, "");
for (i = 0; i < 2; i++) {
comp[i] = LLVMBuildFPTrunc(ctx->builder, comp[i], ctx->f16, "");
comp[i] = LLVMBuildBitCast(ctx->builder, comp[i], ctx->i16, "");
comp[i] = LLVMBuildZExt(ctx->builder, comp[i], ctx->i32, "");
}
comp[1] = LLVMBuildShl(ctx->builder, comp[1], const16, "");
comp[0] = LLVMBuildOr(ctx->builder, comp[0], comp[1], "");
return comp[0];
}
-static LLVMValueRef emit_unpack_half_2x16(struct nir_to_llvm_context *ctx,
+static LLVMValueRef emit_unpack_half_2x16(struct ac_llvm_context *ctx,
LLVMValueRef src0)
{
LLVMValueRef const16 = LLVMConstInt(ctx->i32, 16, false);
LLVMValueRef temps[2], result, val;
int i;
for (i = 0; i < 2; i++) {
val = i == 1 ? LLVMBuildLShr(ctx->builder, src0, const16, "") :
src0;
val = LLVMBuildTrunc(ctx->builder, val, ctx->i16, "");
val = LLVMBuildBitCast(ctx->builder, val, ctx->f16, "");
temps[i] = LLVMBuildFPExt(ctx->builder, val, ctx->f32, "");
}
- result = LLVMBuildInsertElement(ctx->builder, LLVMGetUndef(ctx->v2f32),
temps[0],
- ctx->i32zero, "");
+ LLVMTypeRef v2f32 = LLVMVectorType(ctx->f32, 2);
+ result = LLVMBuildInsertElement(ctx->builder, LLVMGetUndef(v2f32),
temps[0],
+ ctx->i32_0, "");
result = LLVMBuildInsertElement(ctx->builder, result, temps[1],
- ctx->i32one, "");
+ ctx->i32_1, "");
return result;
}
static LLVMValueRef emit_ddxy(struct nir_to_llvm_context *ctx,
nir_op op,
LLVMValueRef src0)
{
unsigned mask;
int idx;
LLVMValueRef result;
@@ -1548,21 +1549,21 @@ static void visit_alu(struct nir_to_llvm_context *ctx,
const nir_alu_instr *inst
case nir_op_imod:
result = LLVMBuildSRem(ctx->builder, src[0], src[1], "");
break;
case nir_op_umod:
result = LLVMBuildURem(ctx->builder, src[0], src[1], "");
break;
case nir_op_fmod:
src[0] = to_float(&ctx->ac, src[0]);
src[1] = to_float(&ctx->ac, src[1]);
result = ac_build_fdiv(&ctx->ac, src[0], src[1]);
- result = emit_intrin_1f_param(ctx, "llvm.floor",
+ result = emit_intrin_1f_param(&ctx->ac, "llvm.floor",
to_float_type(&ctx->ac,
def_type), result);
result = LLVMBuildFMul(ctx->builder, src[1] , result, "");
result = LLVMBuildFSub(ctx->builder, src[0], result, "");
break;
case nir_op_frem:
src[0] = to_float(&ctx->ac, src[0]);
src[1] = to_float(&ctx->ac, src[1]);
result = LLVMBuildFRem(ctx->builder, src[0], src[1], "");
break;
case nir_op_irem:
@@ -1600,151 +1601,151 @@ static void visit_alu(struct nir_to_llvm_context
*ctx, const nir_alu_instr *inst
case nir_op_ishl:
result = LLVMBuildShl(ctx->builder, src[0], src[1], "");
break;
case nir_op_ishr:
result = LLVMBuildAShr(ctx->builder, src[0], src[1], "");
break;
case nir_op_ushr:
result = LLVMBuildLShr(ctx->builder, src[0], src[1], "");
break;
case nir_op_ilt:
- result = emit_int_cmp(ctx, LLVMIntSLT, src[0], src[1]);
+ result = emit_int_cmp(&ctx->ac, LLVMIntSLT, src[0], src[1]);
break;
case nir_op_ine:
- result = emit_int_cmp(ctx, LLVMIntNE, src[0], src[1]);
+ result = emit_int_cmp(&ctx->ac, LLVMIntNE, src[0], src[1]);
break;
case nir_op_ieq:
- result = emit_int_cmp(ctx, LLVMIntEQ, src[0], src[1]);
+ result = emit_int_cmp(&ctx->ac, LLVMIntEQ, src[0], src[1]);
break;
case nir_op_ige:
- result = emit_int_cmp(ctx, LLVMIntSGE, src[0], src[1]);
+ result = emit_int_cmp(&ctx->ac, LLVMIntSGE, src[0], src[1]);
break;
case nir_op_ult:
- result = emit_int_cmp(ctx, LLVMIntULT, src[0], src[1]);
+ result = emit_int_cmp(&ctx->ac, LLVMIntULT, src[0], src[1]);
break;
case nir_op_uge:
- result = emit_int_cmp(ctx, LLVMIntUGE, src[0], src[1]);
+ result = emit_int_cmp(&ctx->ac, LLVMIntUGE, src[0], src[1]);
break;
case nir_op_feq:
- result = emit_float_cmp(ctx, LLVMRealUEQ, src[0], src[1]);
+ result = emit_float_cmp(&ctx->ac, LLVMRealUEQ, src[0], src[1]);
break;
case nir_op_fne:
- result = emit_float_cmp(ctx, LLVMRealUNE, src[0], src[1]);
+ result = emit_float_cmp(&ctx->ac, LLVMRealUNE, src[0], src[1]);
break;
case nir_op_flt:
- result = emit_float_cmp(ctx, LLVMRealULT, src[0], src[1]);
+ result = emit_float_cmp(&ctx->ac, LLVMRealULT, src[0], src[1]);
break;
case nir_op_fge:
- result = emit_float_cmp(ctx, LLVMRealUGE, src[0], src[1]);
+ result = emit_float_cmp(&ctx->ac, LLVMRealUGE, src[0], src[1]);
break;
case nir_op_fabs:
- result = emit_intrin_1f_param(ctx, "llvm.fabs",
+ result = emit_intrin_1f_param(&ctx->ac, "llvm.fabs",
to_float_type(&ctx->ac,
def_type), src[0]);
break;
case nir_op_iabs:
- result = emit_iabs(ctx, src[0]);
+ result = emit_iabs(&ctx->ac, src[0]);
break;
case nir_op_imax:
- result = emit_minmax_int(ctx, LLVMIntSGT, src[0], src[1]);
+ result = emit_minmax_int(&ctx->ac, LLVMIntSGT, src[0], src[1]);
break;
case nir_op_imin:
- result = emit_minmax_int(ctx, LLVMIntSLT, src[0], src[1]);
+ result = emit_minmax_int(&ctx->ac, LLVMIntSLT, src[0], src[1]);
break;
case nir_op_umax:
- result = emit_minmax_int(ctx, LLVMIntUGT, src[0], src[1]);
+ result = emit_minmax_int(&ctx->ac, LLVMIntUGT, src[0], src[1]);
break;
case nir_op_umin:
- result = emit_minmax_int(ctx, LLVMIntULT, src[0], src[1]);
+ result = emit_minmax_int(&ctx->ac, LLVMIntULT, src[0], src[1]);
break;
case nir_op_isign:
- result = emit_isign(ctx, src[0]);
+ result = emit_isign(&ctx->ac, src[0]);
break;
case nir_op_fsign:
src[0] = to_float(&ctx->ac, src[0]);
- result = emit_fsign(ctx, src[0]);
+ result = emit_fsign(&ctx->ac, src[0]);
break;
case nir_op_ffloor:
- result = emit_intrin_1f_param(ctx, "llvm.floor",
+ result = emit_intrin_1f_param(&ctx->ac, "llvm.floor",
to_float_type(&ctx->ac,
def_type), src[0]);
break;
case nir_op_ftrunc:
- result = emit_intrin_1f_param(ctx, "llvm.trunc",
+ result = emit_intrin_1f_param(&ctx->ac, "llvm.trunc",
to_float_type(&ctx->ac,
def_type), src[0]);
break;
case nir_op_fceil:
- result = emit_intrin_1f_param(ctx, "llvm.ceil",
+ result = emit_intrin_1f_param(&ctx->ac, "llvm.ceil",
to_float_type(&ctx->ac,
def_type), src[0]);
break;
case nir_op_fround_even:
- result = emit_intrin_1f_param(ctx, "llvm.rint",
+ result = emit_intrin_1f_param(&ctx->ac, "llvm.rint",
to_float_type(&ctx->ac,
def_type),src[0]);
break;
case nir_op_ffract:
- result = emit_ffract(ctx, src[0]);
+ result = emit_ffract(&ctx->ac, src[0]);
break;
case nir_op_fsin:
- result = emit_intrin_1f_param(ctx, "llvm.sin",
+ result = emit_intrin_1f_param(&ctx->ac, "llvm.sin",
to_float_type(&ctx->ac,
def_type), src[0]);
break;
case nir_op_fcos:
- result = emit_intrin_1f_param(ctx, "llvm.cos",
+ result = emit_intrin_1f_param(&ctx->ac, "llvm.cos",
to_float_type(&ctx->ac,
def_type), src[0]);
break;
case nir_op_fsqrt:
- result = emit_intrin_1f_param(ctx, "llvm.sqrt",
+ result = emit_intrin_1f_param(&ctx->ac, "llvm.sqrt",
to_float_type(&ctx->ac,
def_type), src[0]);
break;
case nir_op_fexp2:
- result = emit_intrin_1f_param(ctx, "llvm.exp2",
+ result = emit_intrin_1f_param(&ctx->ac, "llvm.exp2",
to_float_type(&ctx->ac,
def_type), src[0]);
break;
case nir_op_flog2:
- result = emit_intrin_1f_param(ctx, "llvm.log2",
+ result = emit_intrin_1f_param(&ctx->ac, "llvm.log2",
to_float_type(&ctx->ac,
def_type), src[0]);
break;
case nir_op_frsq:
- result = emit_intrin_1f_param(ctx, "llvm.sqrt",
+ result = emit_intrin_1f_param(&ctx->ac, "llvm.sqrt",
to_float_type(&ctx->ac,
def_type), src[0]);
result = ac_build_fdiv(&ctx->ac, ctx->f32one, result);
break;
case nir_op_fpow:
- result = emit_intrin_2f_param(ctx, "llvm.pow",
+ result = emit_intrin_2f_param(&ctx->ac, "llvm.pow",
to_float_type(&ctx->ac,
def_type), src[0], src[1]);
break;
case nir_op_fmax:
- result = emit_intrin_2f_param(ctx, "llvm.maxnum",
+ result = emit_intrin_2f_param(&ctx->ac, "llvm.maxnum",
to_float_type(&ctx->ac,
def_type), src[0], src[1]);
if (instr->dest.dest.ssa.bit_size == 32)
- result = emit_intrin_1f_param(ctx, "llvm.canonicalize",
+ result = emit_intrin_1f_param(&ctx->ac,
"llvm.canonicalize",
to_float_type(&ctx->ac,
def_type),
result);
break;
case nir_op_fmin:
- result = emit_intrin_2f_param(ctx, "llvm.minnum",
+ result = emit_intrin_2f_param(&ctx->ac, "llvm.minnum",
to_float_type(&ctx->ac,
def_type), src[0], src[1]);
if (instr->dest.dest.ssa.bit_size == 32)
- result = emit_intrin_1f_param(ctx, "llvm.canonicalize",
+ result = emit_intrin_1f_param(&ctx->ac,
"llvm.canonicalize",
to_float_type(&ctx->ac,
def_type),
result);
break;
case nir_op_ffma:
- result = emit_intrin_3f_param(ctx, "llvm.fma",
+ result = emit_intrin_3f_param(&ctx->ac, "llvm.fma",
to_float_type(&ctx->ac,
def_type), src[0], src[1], src[2]);
break;
case nir_op_ibitfield_extract:
- result = emit_bitfield_extract(ctx, true, src);
+ result = emit_bitfield_extract(&ctx->ac, true, src);
break;
case nir_op_ubitfield_extract:
- result = emit_bitfield_extract(ctx, false, src);
+ result = emit_bitfield_extract(&ctx->ac, false, src);
break;
case nir_op_bitfield_insert:
- result = emit_bitfield_insert(ctx, src[0], src[1], src[2],
src[3]);
+ result = emit_bitfield_insert(&ctx->ac, src[0], src[1], src[2],
src[3]);
break;
case nir_op_bitfield_reverse:
result = ac_build_intrinsic(&ctx->ac, "llvm.bitreverse.i32",
ctx->i32, src, 1, AC_FUNC_ATTR_READNONE);
break;
case nir_op_bit_count:
result = ac_build_intrinsic(&ctx->ac, "llvm.ctpop.i32",
ctx->i32, src, 1, AC_FUNC_ATTR_READNONE);
break;
case nir_op_vec2:
case nir_op_vec3:
case nir_op_vec4:
@@ -1784,54 +1785,54 @@ static void visit_alu(struct nir_to_llvm_context *ctx,
const nir_alu_instr *inst
result = LLVMBuildTrunc(ctx->builder, src[0], def_type,
"");
break;
case nir_op_i2i32:
case nir_op_i2i64:
if (get_elem_bits(&ctx->ac, LLVMTypeOf(src[0])) <
get_elem_bits(&ctx->ac, def_type))
result = LLVMBuildSExt(ctx->builder, src[0], def_type,
"");
else
result = LLVMBuildTrunc(ctx->builder, src[0], def_type,
"");
break;
case nir_op_bcsel:
- result = emit_bcsel(ctx, src[0], src[1], src[2]);
+ result = emit_bcsel(&ctx->ac, src[0], src[1], src[2]);
break;
case nir_op_find_lsb:
- result = emit_find_lsb(ctx, src[0]);
+ result = emit_find_lsb(&ctx->ac, src[0]);
break;
case nir_op_ufind_msb:
- result = emit_ufind_msb(ctx, src[0]);
+ result = emit_ufind_msb(&ctx->ac, src[0]);
break;
case nir_op_ifind_msb:
- result = emit_ifind_msb(ctx, src[0]);
+ result = emit_ifind_msb(&ctx->ac, src[0]);
break;
case nir_op_uadd_carry:
- result = emit_uint_carry(ctx, "llvm.uadd.with.overflow.i32",
src[0], src[1]);
+ result = emit_uint_carry(&ctx->ac,
"llvm.uadd.with.overflow.i32", src[0], src[1]);
break;
case nir_op_usub_borrow:
- result = emit_uint_carry(ctx, "llvm.usub.with.overflow.i32",
src[0], src[1]);
+ result = emit_uint_carry(&ctx->ac,
"llvm.usub.with.overflow.i32", src[0], src[1]);
break;
case nir_op_b2f:
- result = emit_b2f(ctx, src[0]);
+ result = emit_b2f(&ctx->ac, src[0]);
break;
case nir_op_fquantize2f16:
result = emit_f2f16(ctx, src[0]);
break;
case nir_op_umul_high:
- result = emit_umul_high(ctx, src[0], src[1]);
+ result = emit_umul_high(&ctx->ac, src[0], src[1]);
break;
case nir_op_imul_high:
- result = emit_imul_high(ctx, src[0], src[1]);
+ result = emit_imul_high(&ctx->ac, src[0], src[1]);
break;
case nir_op_pack_half_2x16:
- result = emit_pack_half_2x16(ctx, src[0]);
+ result = emit_pack_half_2x16(&ctx->ac, src[0]);
break;
case nir_op_unpack_half_2x16:
- result = emit_unpack_half_2x16(ctx, src[0]);
+ result = emit_unpack_half_2x16(&ctx->ac, src[0]);
break;
case nir_op_fddx:
case nir_op_fddy:
case nir_op_fddx_fine:
case nir_op_fddy_fine:
case nir_op_fddx_coarse:
case nir_op_fddy_coarse:
result = emit_ddxy(ctx, instr->op, src[0]);
break;
default:
@@ -3624,22 +3625,22 @@ static LLVMValueRef load_sample_position(struct
nir_to_llvm_context *ctx,
sample_id = LLVMBuildAdd(ctx->builder, sample_id,
ctx->sample_pos_offset, "");
result = ac_build_indexed_load(&ctx->ac, ptr, sample_id, false);
return result;
}
static LLVMValueRef load_sample_pos(struct nir_to_llvm_context *ctx)
{
LLVMValueRef values[2];
- values[0] = emit_ffract(ctx, ctx->frag_pos[0]);
- values[1] = emit_ffract(ctx, ctx->frag_pos[1]);
+ values[0] = emit_ffract(&ctx->ac, ctx->frag_pos[0]);
+ values[1] = emit_ffract(&ctx->ac, ctx->frag_pos[1]);
return ac_build_gather_values(&ctx->ac, values, 2);
}
static LLVMValueRef visit_interp(struct nir_to_llvm_context *ctx,
const nir_intrinsic_instr *instr)
{
LLVMValueRef result[2];
LLVMValueRef interp_param, attr_number;
unsigned location;
unsigned chan;
@@ -4407,21 +4408,21 @@ static void visit_tex(struct nir_to_llvm_context *ctx,
nir_tex_instr *instr)
txf_address[2] = ctx->i32zero;
txf_address[3] = ctx->i32zero;
set_tex_fetch_args(ctx, &txf_args, instr, nir_texop_txf,
fmask_ptr, NULL,
txf_address, txf_count, 0xf);
result = build_tex_intrinsic(ctx, instr, false, &txf_args);
result = LLVMBuildExtractElement(ctx->builder, result,
ctx->i32zero, "");
- result = emit_int_cmp(ctx, LLVMIntEQ, result, ctx->i32zero);
+ result = emit_int_cmp(&ctx->ac, LLVMIntEQ, result,
ctx->i32zero);
goto write_result;
}
if (instr->sampler_dim == GLSL_SAMPLER_DIM_MS &&
instr->op != nir_texop_txs) {
unsigned sample_chan = instr->is_array ? 3 : 2;
address[sample_chan] = adjust_sample_index_using_fmask(ctx,
address[0],
address[1],
instr->is_array ? address[2] : NULL,
@@ -4946,23 +4947,23 @@ setup_locals(struct nir_to_llvm_context *ctx,
for (i = 0; i < ctx->num_locals; i++) {
for (j = 0; j < 4; j++) {
ctx->locals[i * 4 + j] =
si_build_alloca_undef(ctx, ctx->f32, "temp");
}
}
}
static LLVMValueRef
-emit_float_saturate(struct nir_to_llvm_context *ctx, LLVMValueRef v, float lo,
float hi)
+emit_float_saturate(struct ac_llvm_context *ctx, LLVMValueRef v, float lo,
float hi)
{
- v = to_float(&ctx->ac, v);
+ v = to_float(ctx, v);
v = emit_intrin_2f_param(ctx, "llvm.maxnum.f32", ctx->f32, v,
LLVMConstReal(ctx->f32, lo));
return emit_intrin_2f_param(ctx, "llvm.minnum.f32", ctx->f32, v,
LLVMConstReal(ctx->f32, hi));
}
static LLVMValueRef emit_pack_int16(struct nir_to_llvm_context *ctx,
LLVMValueRef src0, LLVMValueRef src1)
{
LLVMValueRef const16 = LLVMConstInt(ctx->i32, 16, false);
LLVMValueRef comp[2];
@@ -5056,21 +5057,21 @@ si_llvm_init_export_args(struct nir_to_llvm_context
*ctx,
ctx->i32, "");
}
args->compr = 1;
args->out[0] = emit_pack_int16(ctx, val[0], val[1]);
args->out[1] = emit_pack_int16(ctx, val[2], val[3]);
break;
case V_028714_SPI_SHADER_SNORM16_ABGR:
for (unsigned chan = 0; chan < 4; chan++) {
- val[chan] = emit_float_saturate(ctx,
values[chan], -1, 1);
+ val[chan] = emit_float_saturate(&ctx->ac,
values[chan], -1, 1);
val[chan] = LLVMBuildFMul(ctx->builder,
val[chan],
LLVMConstReal(ctx->f32,
32767), "");
/* If positive, add 0.5, else add -0.5. */
val[chan] = LLVMBuildFAdd(ctx->builder,
val[chan],
LLVMBuildSelect(ctx->builder,
LLVMBuildFCmp(ctx->builder, LLVMRealOGE,
val[chan],
ctx->f32zero, ""),
LLVMConstReal(ctx->f32,
0.5),
LLVMConstReal(ctx->f32,
-0.5), ""), "");
@@ -5080,38 +5081,38 @@ si_llvm_init_export_args(struct nir_to_llvm_context
*ctx,
args->compr = 1;
args->out[0] = emit_pack_int16(ctx, val[0], val[1]);
args->out[1] = emit_pack_int16(ctx, val[2], val[3]);
break;
case V_028714_SPI_SHADER_UINT16_ABGR: {
LLVMValueRef max = LLVMConstInt(ctx->i32, is_int8 ? 255
: 65535, 0);
for (unsigned chan = 0; chan < 4; chan++) {
val[chan] = to_integer(&ctx->ac, values[chan]);
- val[chan] = emit_minmax_int(ctx, LLVMIntULT,
val[chan], max);
+ val[chan] = emit_minmax_int(&ctx->ac,
LLVMIntULT, val[chan], max);
}
args->compr = 1;
args->out[0] = emit_pack_int16(ctx, val[0], val[1]);
args->out[1] = emit_pack_int16(ctx, val[2], val[3]);
break;
}
case V_028714_SPI_SHADER_SINT16_ABGR: {
LLVMValueRef max = LLVMConstInt(ctx->i32, is_int8 ? 127
: 32767, 0);
LLVMValueRef min = LLVMConstInt(ctx->i32, is_int8 ?
-128 : -32768, 0);
/* Clamp. */
for (unsigned chan = 0; chan < 4; chan++) {
val[chan] = to_integer(&ctx->ac, values[chan]);
- val[chan] = emit_minmax_int(ctx, LLVMIntSLT,
val[chan], max);
- val[chan] = emit_minmax_int(ctx, LLVMIntSGT,
val[chan], min);
+ val[chan] = emit_minmax_int(&ctx->ac,
LLVMIntSLT, val[chan], max);
+ val[chan] = emit_minmax_int(&ctx->ac,
LLVMIntSGT, val[chan], min);
}
args->compr = 1;
args->out[0] = emit_pack_int16(ctx, val[0], val[1]);
args->out[1] = emit_pack_int16(ctx, val[2], val[3]);
break;
}
default:
case V_028714_SPI_SHADER_32_ABGR:
--
2.9.3
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
