[clang][RISCV] Support RVV bfloat16 C intrinsics #89354
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
llvmbot
added
clang
|
@llvm/pr-subscribers-clang @llvm/pr-subscribers-backend-risc-v Author: Brandon Wu (4vtomat) ChangesIt follows the interface defined here: Patch is 4.84 MiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/89354.diff 252 Files Affected:
diff --git a/clang/include/clang/Basic/riscv_vector.td b/clang/include/clang/Basic/riscv_vector.td
index 87a18e8474ef50..4120b2e663f666 100644
--- a/clang/include/clang/Basic/riscv_vector.td
+++ b/clang/include/clang/Basic/riscv_vector.td
@@ -14,14 +14,14 @@
include "riscv_vector_common.td"
-defvar TypeList = ["c","s","i","l","x","f","d"];
+defvar TypeList = ["c","s","i","l","x","f","d","y"];
defvar EEWList = [["8", "(Log2EEW:3)"],
["16", "(Log2EEW:4)"],
["32", "(Log2EEW:5)"],
["64", "(Log2EEW:6)"]];
class IsFloat<string type> {
- bit val = !or(!eq(type, "x"), !eq(type, "f"), !eq(type, "d"));
+ bit val = !or(!eq(type, "x"), !eq(type, "f"), !eq(type, "d"), !eq(type, "y"));
}
let SupportOverloading = false,
@@ -118,7 +118,8 @@ multiclass RVVIndexedLoad<string op> {
defvar eew_type = eew_list[1];
let Name = op # eew # "_v", IRName = op, MaskedIRName = op # "_mask",
RequiredFeatures = !if(!eq(type, "x"), ["Zvfhmin"],
- []<string>) in {
+ !if(!eq(type, "y"), ["Zvfbfmin"],
+ []<string>)) in {
def: RVVOutOp1Builtin<"v", "vPCe" # eew_type # "Uv", type>;
if !not(IsFloat<type>.val) then {
def: RVVOutOp1Builtin<"Uv", "UvPCUe" # eew_type # "Uv", type>;
@@ -129,7 +130,8 @@ multiclass RVVIndexedLoad<string op> {
defvar eew64_type = "(Log2EEW:6)";
let Name = op # eew64 # "_v", IRName = op, MaskedIRName = op # "_mask",
RequiredFeatures = !if(!eq(type, "x"), ["Zvfhmin", "RV64"],
- ["RV64"]) in {
+ !if(!eq(type, "y"), ["Zvfbfmin", "RV64"],
+ ["RV64"])) in {
def: RVVOutOp1Builtin<"v", "vPCe" # eew64_type # "Uv", type>;
if !not(IsFloat<type>.val) then {
def: RVVOutOp1Builtin<"Uv", "UvPCUe" # eew64_type # "Uv", type>;
@@ -223,7 +225,8 @@ multiclass RVVIndexedStore<string op> {
defvar eew_type = eew_list[1];
let Name = op # eew # "_v", IRName = op, MaskedIRName = op # "_mask",
RequiredFeatures = !if(!eq(type, "x"), ["Zvfhmin"],
- []<string>) in {
+ !if(!eq(type, "y"), ["Zvfbfmin"],
+ []<string>)) in {
def : RVVBuiltin<"v", "0Pe" # eew_type # "Uvv", type>;
if !not(IsFloat<type>.val) then {
def : RVVBuiltin<"Uv", "0PUe" # eew_type # "UvUv", type>;
@@ -234,7 +237,8 @@ multiclass RVVIndexedStore<string op> {
defvar eew64_type = "(Log2EEW:6)";
let Name = op # eew64 # "_v", IRName = op, MaskedIRName = op # "_mask",
RequiredFeatures = !if(!eq(type, "x"), ["Zvfhmin", "RV64"],
- ["RV64"]) in {
+ !if(!eq(type, "y"), ["Zvfbfmin", "RV64"],
+ ["RV64"])) in {
def : RVVBuiltin<"v", "0Pe" # eew64_type # "Uvv", type>;
if !not(IsFloat<type>.val) then {
def : RVVBuiltin<"Uv", "0PUe" # eew64_type # "UvUv", type>;
@@ -357,7 +361,11 @@ multiclass RVVNonTupleVCreateBuiltin<int dst_lmul, list<int> src_lmul_list> {
defvar src_s = FixedVString<src_lmul, num, "v">.S;
def vcreate # src_v # dst_v : RVVBuiltin<src_v # dst_v,
dst_v # src_s,
- "csilxfd", dst_v>;
+ "csilxfdy", dst_v>;
+ let RequiredFeatures = ["Zvfbfmin"] in
+ def vcreate_bf16 # src_v # dst_v : RVVBuiltin<src_v # dst_v,
+ dst_v # src_s,
+ "y", dst_v>;
defvar src_uv = FixedVString<src_lmul, num, "Uv">.V;
defvar src_us = FixedVString<src_lmul, num, "Uv">.S;
@@ -683,6 +691,8 @@ defm vle8: RVVVLEBuiltin<["c"]>;
defm vle16: RVVVLEBuiltin<["s"]>;
let Name = "vle16_v", RequiredFeatures = ["Zvfhmin"] in
defm vle16_h: RVVVLEBuiltin<["x"]>;
+let Name = "vle16_v", RequiredFeatures = ["Zvfbfmin"] in
+ defm vle16_bf16 : RVVVLEBuiltin<["y"]>;
defm vle32: RVVVLEBuiltin<["i","f"]>;
defm vle64: RVVVLEBuiltin<["l","d"]>;
@@ -691,6 +701,8 @@ defm vse8 : RVVVSEBuiltin<["c"]>;
defm vse16: RVVVSEBuiltin<["s"]>;
let Name = "vse16_v", RequiredFeatures = ["Zvfhmin"] in
defm vse16_h: RVVVSEBuiltin<["x"]>;
+let Name = "vse16_v", RequiredFeatures = ["Zvfbfmin"] in
+ defm vse16_bf16: RVVVSEBuiltin<["y"]>;
defm vse32: RVVVSEBuiltin<["i","f"]>;
defm vse64: RVVVSEBuiltin<["l","d"]>;
@@ -699,6 +711,8 @@ defm vlse8: RVVVLSEBuiltin<["c"]>;
defm vlse16: RVVVLSEBuiltin<["s"]>;
let Name = "vlse16_v", RequiredFeatures = ["Zvfhmin"] in
defm vlse16_h: RVVVLSEBuiltin<["x"]>;
+let Name = "vlse16_v", RequiredFeatures = ["Zvfbfmin"] in
+ defm vlse16_bf16: RVVVLSEBuiltin<["y"]>;
defm vlse32: RVVVLSEBuiltin<["i","f"]>;
defm vlse64: RVVVLSEBuiltin<["l","d"]>;
@@ -706,6 +720,8 @@ defm vsse8 : RVVVSSEBuiltin<["c"]>;
defm vsse16: RVVVSSEBuiltin<["s"]>;
let Name = "vsse16_v", RequiredFeatures = ["Zvfhmin"] in
defm vsse16_h: RVVVSSEBuiltin<["x"]>;
+let Name = "vsse16_v", RequiredFeatures = ["Zvfbfmin"] in
+ defm vsse16_bf: RVVVSSEBuiltin<["y"]>;
defm vsse32: RVVVSSEBuiltin<["i","f"]>;
defm vsse64: RVVVSSEBuiltin<["l","d"]>;
@@ -721,6 +737,8 @@ defm vle8ff: RVVVLEFFBuiltin<["c"]>;
defm vle16ff: RVVVLEFFBuiltin<["s"]>;
let Name = "vle16ff_v", RequiredFeatures = ["Zvfhmin"] in
defm vle16ff: RVVVLEFFBuiltin<["x"]>;
+let Name = "vle16ff_v", RequiredFeatures = ["Zvfbfmin"] in
+ defm vle16ff: RVVVLEFFBuiltin<["y"]>;
defm vle32ff: RVVVLEFFBuiltin<["i", "f"]>;
defm vle64ff: RVVVLEFFBuiltin<["l", "d"]>;
@@ -732,14 +750,16 @@ multiclass RVVUnitStridedSegLoadTuple<string op> {
!eq(type, "l") : "64",
!eq(type, "x") : "16",
!eq(type, "f") : "32",
- !eq(type, "d") : "64");
+ !eq(type, "d") : "64",
+ !eq(type, "y") : "16");
foreach nf = NFList in {
let Name = op # nf # "e" # eew # "_v",
IRName = op # nf,
MaskedIRName = op # nf # "_mask",
NF = nf,
RequiredFeatures = !if(!eq(type, "x"), ["Zvfhmin"],
- []<string>),
+ !if(!eq(type, "y"), ["Zvfbfmin"],
+ []<string>)),
ManualCodegen = [{
{
llvm::Type *ElementVectorType = cast<StructType>(ResultType)->elements()[0];
@@ -793,7 +813,8 @@ multiclass RVVUnitStridedSegStoreTuple<string op> {
!eq(type, "l") : "64",
!eq(type, "x") : "16",
!eq(type, "f") : "32",
- !eq(type, "d") : "64");
+ !eq(type, "d") : "64",
+ !eq(type, "y") : "16");
foreach nf = NFList in {
let Name = op # nf # "e" # eew # "_v",
IRName = op # nf,
@@ -801,7 +822,8 @@ multiclass RVVUnitStridedSegStoreTuple<string op> {
NF = nf,
HasMaskedOffOperand = false,
RequiredFeatures = !if(!eq(type, "x"), ["Zvfhmin"],
- []<string>),
+ !if(!eq(type, "y"), ["Zvfbfmin"],
+ []<string>)),
ManualCodegen = [{
{
// Masked
@@ -846,14 +868,16 @@ multiclass RVVUnitStridedSegLoadFFTuple<string op> {
!eq(type, "l") : "64",
!eq(type, "x") : "16",
!eq(type, "f") : "32",
- !eq(type, "d") : "64");
+ !eq(type, "d") : "64",
+ !eq(type, "y") : "16");
foreach nf = NFList in {
let Name = op # nf # "e" # eew # "ff_v",
IRName = op # nf # "ff",
MaskedIRName = op # nf # "ff_mask",
NF = nf,
RequiredFeatures = !if(!eq(type, "x"), ["Zvfhmin"],
- []<string>),
+ !if(!eq(type, "y"), ["Zvfbfmin"],
+ []<string>)),
ManualCodegen = [{
{
llvm::Type *ElementVectorType = cast<StructType>(ResultType)->elements()[0];
@@ -921,14 +945,16 @@ multiclass RVVStridedSegLoadTuple<string op> {
!eq(type, "l") : "64",
!eq(type, "x") : "16",
!eq(type, "f") : "32",
- !eq(type, "d") : "64");
+ !eq(type, "d") : "64",
+ !eq(type, "y") : "16");
foreach nf = NFList in {
let Name = op # nf # "e" # eew # "_v",
IRName = op # nf,
MaskedIRName = op # nf # "_mask",
NF = nf,
RequiredFeatures = !if(!eq(type, "x"), ["Zvfhmin"],
- []<string>),
+ !if(!eq(type, "y"), ["Zvfbfmin"],
+ []<string>)),
ManualCodegen = [{
{
llvm::Type *ElementVectorType = cast<StructType>(ResultType)->elements()[0];
@@ -983,7 +1009,8 @@ multiclass RVVStridedSegStoreTuple<string op> {
!eq(type, "l") : "64",
!eq(type, "x") : "16",
!eq(type, "f") : "32",
- !eq(type, "d") : "64");
+ !eq(type, "d") : "64",
+ !eq(type, "y") : "16");
foreach nf = NFList in {
let Name = op # nf # "e" # eew # "_v",
IRName = op # nf,
@@ -992,7 +1019,8 @@ multiclass RVVStridedSegStoreTuple<string op> {
HasMaskedOffOperand = false,
MaskedPolicyScheme = NonePolicy,
RequiredFeatures = !if(!eq(type, "x"), ["Zvfhmin"],
- []<string>),
+ !if(!eq(type, "y"), ["Zvfbfmin"],
+ []<string>)),
ManualCodegen = [{
{
// Masked
@@ -1041,7 +1069,8 @@ multiclass RVVIndexedSegLoadTuple<string op> {
MaskedIRName = op # nf # "_mask",
NF = nf,
RequiredFeatures = !if(!eq(type, "x"), ["Zvfhmin"],
- []<string>),
+ !if(!eq(type, "y"), ["Zvfbfmin"],
+ []<string>)),
ManualCodegen = [{
{
llvm::Type *ElementVectorType = cast<StructType>(ResultType)->elements()[0];
@@ -1104,7 +1133,8 @@ multiclass RVVIndexedSegStoreTuple<string op> {
HasMaskedOffOperand = false,
MaskedPolicyScheme = NonePolicy,
RequiredFeatures = !if(!eq(type, "x"), ["Zvfhmin"],
- []<string>),
+ !if(!eq(type, "y"), ["Zvfbfmin"],
+ []<string>)),
ManualCodegen = [{
{
// Masked
@@ -2308,6 +2338,12 @@ let HasMasked = false, HasVL = false, IRName = "" in {
def vreinterpret_h_i : RVVBuiltin<"vFv", "Fvv", "s", "Fv">;
def vreinterpret_h_u : RVVBuiltin<"UvFv", "FvUv", "s", "Fv">;
}
+ let RequiredFeatures = ["Zvfbfmin"] in {
+ def vreinterpret_i_bf16 : RVVBuiltin<"vIv", "Ivv", "y", "Iv">;
+ def vreinterpret_u_bf16 : RVVBuiltin<"vUv", "Uvv", "y", "Uv">;
+ def vreinterpret_bf16_i : RVVBuiltin<"Ivv", "vIv", "y", "v">;
+ def vreinterpret_bf16_u : RVVBuiltin<"Uvv", "vUv", "y", "v">;
+ }
// Reinterpret between different SEW under the same LMUL
foreach dst_sew = ["(FixedSEW:8)", "(FixedSEW:16)", "(FixedSEW:32)",
@@ -2372,12 +2408,16 @@ let HasMasked = false, HasVL = false, IRName = "" in {
return llvm::PoisonValue::get(ResultType);
}] in {
def vundefined : RVVBuiltin<"v", "v", "csilxfd">;
+ let RequiredFeatures = ["Zvfbfmin"] in
+ def vundefined_bf16 : RVVBuiltin<"v", "v", "y">;
def vundefined_u : RVVBuiltin<"Uv", "Uv", "csil">;
foreach nf = NFList in {
let NF = nf in {
defvar T = "(Tuple:" # nf # ")";
def : RVVBuiltin<T # "v", T # "v", "csilxfd">;
+ let RequiredFeatures = ["Zvfbfmin"] in
+ def : RVVBuiltin<T # "v", T # "v", "y">;
def : RVVBuiltin<T # "Uv", T # "Uv", "csil">;
}
}
@@ -2396,6 +2436,9 @@ let HasMasked = false, HasVL = false, IRName = "" in {
"(SFixedLog2LMUL:0)", "(SFixedLog2LMUL:1)", "(SFixedLog2LMUL:2)"] in {
def vlmul_trunc # dst_lmul : RVVBuiltin<"v" # dst_lmul # "v",
dst_lmul # "vv", "csilxfd", dst_lmul # "v">;
+ let RequiredFeatures = ["Zvfbfmin"] in
+ def vlmul_trunc_bf16 # dst_lmul : RVVBuiltin<"v" # dst_lmul # "v",
+ dst_lmul # "vv", "y", dst_lmul # "v">;
def vlmul_trunc_u # dst_lmul : RVVBuiltin<"Uv" # dst_lmul # "Uv",
dst_lmul # "UvUv", "csil", dst_lmul # "Uv">;
}
@@ -2414,6 +2457,9 @@ let HasMasked = false, HasVL = false, IRName = "" in {
"(LFixedLog2LMUL:1)", "(LFixedLog2LMUL:2)", "(LFixedLog2LMUL:3)"] in {
def vlmul_ext # dst_lmul : RVVBuiltin<"v" # dst_lmul # "v",
dst_lmul # "vv", "csilxfd", dst_lmul # "v">;
+ let RequiredFeatures = ["Zvfbfmin"] in
+ def vlmul_ext_bf16 # dst_lmul : RVVBuiltin<"v" # dst_lmul # "v",
+ dst_lmul # "vv", "y", dst_lmul # "v">;
def vlmul_ext_u # dst_lmul : RVVBuiltin<"Uv" # dst_lmul # "Uv",
dst_lmul # "UvUv", "csil", dst_lmul # "Uv">;
}
@@ -2441,12 +2487,12 @@ let HasMasked = false, HasVL = false, IRName = "" in {
}
}] in {
foreach dst_lmul = ["(SFixedLog2LMUL:0)", "(SFixedLog2LMUL:1)", "(SFixedLog2LMUL:2)"] in {
- def : RVVBuiltin<"v" # dst_lmul # "v", dst_lmul # "vvKz", "csilxfd", dst_lmul # "v">;
+ def : RVVBuiltin<"v" # dst_lmul # "v", dst_lmul # "vvKz", "csilxfdy", dst_lmul # "v">;
def : RVVBuiltin<"Uv" # dst_lmul # "Uv", dst_lmul # "UvUvKz", "csil", dst_lmul # "Uv">;
}
foreach nf = NFList in {
defvar T = "(Tuple:" # nf # ")";
- def : RVVBuiltin<T # "vv", "v" # T # "vKz", "csilxfd", "v">;
+ def : RVVBuiltin<T # "vv", "v" # T # "vKz", "csilxfdy", "v">;
def : RVVBuiltin<T # "UvUv", "Uv" # T # "UvKz", "csil", "Uv">;
}
}
@@ -2474,11 +2520,15 @@ let HasMasked = false, HasVL = false, IRName = "" in {
}] in {
foreach dst_lmul = ["(LFixedLog2LMUL:1)", "(LFixedLog2LMUL:2)", "(LFixedLog2LMUL:3)"] in {
def : RVVBuiltin<"v" # dst_lmul # "v", dst_lmul # "v" # dst_lmul # "vKzv", "csilxfd">;
+ let RequiredFeatures = ["Zvfbfmin"] in
+ def : RVVBuiltin<"v" # dst_lmul # "v", dst_lmul # "v" # dst_lmul # "vKzv", "y">;
def : RVVBuiltin<"Uv" # dst_lmul # "Uv", dst_lmul # "Uv" # dst_lmul #"UvKzUv", "csil">;
}
foreach nf = NFList in {
defvar T = "(Tuple:" # nf # ")";
def : RVVBuiltin<"v" # T # "v", T # "v" # T # "vKzv", "csilxfd">;
+ let RequiredFeatures = ["Zvfbfmin"] in
+ def : RVVBuiltin<"v" # T # "v", T # "v" # T # "vKzv", "y">;
def : RVVBuiltin<"Uv" # T # "Uv", T # "Uv" # T # "UvKzUv", "csil">;
}
}
@@ -2523,7 +2573,9 @@ let HasMasked = false, HasVL = false, IRName = "" in {
defvar T = "(Tuple:" # nf # ")";
defvar V = VString<nf, /*signed=*/true>.S;
defvar UV = VString<nf, /*signed=*/false>.S;
- def : RVVBuiltin<T # "v", T # "v" # V, "csilxfd">;
+ def : RVVBuiltin<T # "v", T # "v" # V, "csilxfdy">;
+ let RequiredFeatures = ["Zvfbfmin"] in
+ def : RVVBuiltin<T # "v", T # "v" # V, "y">;
def : RVVBuiltin<T # "Uv", T # "Uv" # UV, "csil">;
}
}
diff --git a/clang/include/clang/Support/RISCVVIntrinsicUtils.h b/clang/include/clang/Support/RISCVVIntrinsicUtils.h
index ef9d6c15724b68..97493bae5656e9 100644
--- a/clang/include/clang/Support/RISCVVIntrinsicUtils.h
+++ b/clang/include/clang/Support/RISCVVIntrinsicUtils.h
@@ -502,7 +502,8 @@ enum RVVRequire : uint32_t {
RVV_REQ_Zvksed = 1 << 14,
RVV_REQ_Zvksh = 1 << 15,
RVV_REQ_Zvfbfwma = 1 << 16,
- RVV_REQ_Experimental = 1 << 17,
+ RVV_REQ_Zvfbfmin = 1 << 17,
+ RVV_REQ_Experimental = 1 << 18,
LLVM_MARK_AS_BITMASK_ENUM(RVV_REQ_Experimental)
};
diff --git a/clang/lib/Sema/SemaRISCVVectorLookup.cpp b/clang/lib/Sema/SemaRISCVVectorLookup.cpp
index bf89a4ac51afd2..26e13e87b1d6b6 100644
--- a/clang/lib/Sema/SemaRISCVVectorLookup.cpp
+++ b/clang/lib/Sema/SemaRISCVVectorLookup.cpp
@@ -216,6 +216,7 @@ void RISCVIntrinsicManagerImpl::ConstructRVVIntrinsics(
{"zvksed", RVV_REQ_Zvksed},
{"zvksh", RVV_REQ_Zvksh},
{"zvfbfwma", RVV_REQ_Zvfbfwma},
+ {"zvfbfmin", RVV_REQ_Zvfbfmin},
{"experimental", RVV_REQ_Experimental}};
// Construction of RVVIntrinsicRecords need to sync with createRVVIntrinsics
diff --git a/clang/test/CodeGen/RISCV/rvv-intrinsics-autogenerated/non-policy/non-overloaded/bfloat16/vcreate.c b/clang/test/CodeGen/RISCV/rvv-intrinsics-autogenerated/non-policy/non-overloaded/bfloat16/vcreate.c
new file mode 100644
index 00000000000000..8c8e1cdfb65884
--- /dev/null
+++ b/clang/test/CodeGen/RISCV/rvv-intrinsics-autogenerated/non-policy/non-overloaded/bfloat16/vcreate.c
@@ -0,0 +1,477 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --version 4
+// REQUIRES: riscv-registered-target
+// RUN: %clang_cc1 -triple riscv64 -target-feature +v \
+// RUN: -target-feature +experimental-zvfbfmin \
+// RUN: -target-feature +experimental-zvfbfwma -disable-O0-optnone \
+// RUN: -emit-llvm %s -o - | opt -S -passes=mem2reg | \
+// RUN: FileCheck --check-prefix=CHECK-RV64 %s
+
+#include <riscv_vector.h>
+
+// CHECK-RV64-LABEL: define dso_local <vscale x 8 x bfloat> @test_vcreate_v_bf16m1_bf16m2(
+// CHECK-RV64-SAME: <vscale x 4 x bfloat> [[V0:%.*]], <vscale x 4 x bfloat> [[V1:%.*]]) #[[ATTR0:[0-9]+]] {
+// CHECK-RV64-NEXT: entry:
+// CHECK-RV64-NEXT: [[TMP0:%.*]] = call <vscale x 8 x bfloat> @llvm.vector.insert.nxv8bf16.nxv4bf16(<vscale x 8 x bfloat> poison, <vscale x 4 x bfloat> [[V0]], i64 0)
+// CHECK-RV64-NEXT: [[TMP1:%.*]] = call <vscale x 8 x bfloat> @llvm.vector.insert.nxv8bf16.nxv4bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 4 x bfloat> [[V1]], i64 4)
+// CHECK-RV64-NEXT: ret <vscale x 8 x bfloat> [[TMP1]]
+//
+vbfloat16m2_t test_vcreate_v_bf16m1_bf16m2(vbfloat16m1_t v0, vbfloat16m1_t v1) {
+ return __riscv_vcreate_v_bf16m1_bf16m2(v0, v1);
+}
+
+// CHECK-RV64-LABEL: define dso_local <vscale x 16 x bfloat> @test_vcreate_v_bf16m1_bf16m4(
+// CHECK-RV64-SAME: <vscale x 4 x bfloat> [[V0:%.*]], <vscale x 4 x bfloat> [[V1:%.*]], <vscale x 4 x bfloat> [[V2:%.*]], <vscale x 4 x bfloat> [[V3:%.*]]) #[[ATTR0]] {
+// CHECK-RV64-NEXT: entry:
+// CHECK-RV64-NEXT: [[TMP0:%.*]] = call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv4bf16(<vscale x 16 x bfloat> poison, <vscale x 4 x bfloat> [[V0]], i64 0)
+// CHECK-RV64-NEXT: [[TMP1:%.*]] = call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv4bf16(<vscale x 16 x bfloat> [[TMP0]], <vscale x 4 x bfloat> [[V1]], i64 4)
+// CHECK-RV64-NEXT: [[TMP2:%.*]] = call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv4bf16(<vscale x 16 x bfloat> [[TMP1]], <vscale x 4 x bfloat> [[V2]], i64 8)
+// CHECK-RV64-NEXT: [[TMP3:%.*]] = call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv4bf16(<vscale x 16 x bfloat> [[TMP2]], <vscale x 4 x bfloat> [[V3]], i64 12)
+// CHECK-RV64-NEXT: ret ...
[truncated]
|
4vtomat
requested review from
kito-cheng,
tclin914,
eopXD,
jacquesguan and
topperc
It follows the interface defined here: riscv-non-isa/rvv-intrinsic-doc#293
4vtomat
force-pushed
the
support_bfloat_intrinsics
branch
from
5ab8aa8
to
f683547
Compare
|
vfncvtbf16.c, vfwcvtbf16.c and vfwmaccbf16.c already in the LLVM repo, so I think those files could removed from this PR? |
|
Oh, I forgot to remove them. Or do you think they should be moved to bfloat folder to make them consistent? |
Remove files from this PR, that should be a separated NFC PR for moving those files, but I am fine to keep those file in same place :) |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
It follows the interface defined here:
riscv-non-isa/rvv-intrinsic-doc#293