[ValueTypes] Define MVTs for v6i32, v6f32, v7i32, v7f32

[critson]

Summary

For use in AMDGPU selection DAG.

[frasercrmck]

Awfully brave to add new non-pow2 types; I like it. Unfortunate that we've both filed patches to this same file at the same time (D103884). One of us will have a fun time.

[RKSimon]

In D103881#2805122, @frasercrmck wrote:

Awfully brave to add new non-pow2 types; I like it. Unfortunate that we've both filed patches to this same file at the same time (D103884). One of us will have a fun time.

Technically there's precedent with MVT::v3i32 but where does this end? What happens when someone next asks for v8f32 -> v15f32? How tricky would it be for a custom AMDGPU DAG node to encode the vector width instead and it just uses the next pow2 type?

[critson]

In D103881#2805122, @frasercrmck wrote:

Awfully brave to add new non-pow2 types; I like it. Unfortunate that we've both filed patches to this same file at the same time (D103884). One of us will have a fun time.

Feel free to commit. I'll rebase this on top of the new state.

In D103881#2805418, @RKSimon wrote:

Technically there's precedent with MVT::v3i32 but where does this end? What happens when someone next asks for v8f32 -> v15f32? How tricky would it be for a custom AMDGPU DAG node to encode the vector width instead and it just uses the next pow2 type?

Yes, the precedent is v3i32/v5i32.
I agree its a dangerous slope, but I would be happy if we declared v7 as the hard limit for non-pow2 types.
Declaring a hard limit may also benefit code that has to work with these?

On recent hardware we have features which mean there isn't a need for large contiguous vector types in common cases.
However we have to disable this beyond v5 on some common hardware for stability.
Supporting v6/v7 benefits all our hardware going back many generations (where benefit = reduced register pressure).
I can workaround the lack of v6/v7, but our consensus is that we should just add the MVT types (see D103800).

[foad]

In D103881#2805418, @RKSimon wrote:

Technically there's precedent with MVT::v3i32 but where does this end?

Where does it end for power-of-two types? It seems like they get added whenever any target has a need for them. I don't see why non-power-of-two types would be fundamentally different in that regard.

[critson]

• Rebase

[RKSimon]

OK, I get where everyone's coming from and I'm not going to block it - it just feels untidy to me. LGTM

[srj]

This commit seems to have injected a failure for x86-32 codegen in Halide; for one of our tests involving vector reductions, this commit causes a new assertion failure: MachineValueType.h:469: llvm::MVT llvm::MVT::changeVectorElementType(llvm::MVT) const: Assertion VecTy.SimpleTy != MVT::INVALID_SIMPLE_VALUE_TYPE && "Simple vector VT not representable by simple integer vector VT!"' failed.`

I'll be adding more details and adding a repro case as soon as I have a small one; posting this now to see if anyone else has noticed a regression/injection.

[srj]

In D103881#2817686, @srj wrote:

I'll be adding more details and adding a repro case as soon as I have a small one; posting this now to see if anyone else has noticed a regression/injection.

I don't have a standalone repro case for you yet, but the assertion failure in MVT::changeVectorElementType(EltVT) has this = llvm::MVT::v6i32, EltVT = llvm::MVT::i16.

[srj]

Debugging into this a little further, it's unclear to me how EVT::changeVectorElementType() can be expected to work reliably: it assumes that if EltVT.isSimple() == true, then MVT::changeVectorElementType(EltVT) will be legal, but this is clearly not the case, since there aren't equivalent widths of MVTs for all element types. (In this case, v6i32 -> v6i16 isn't possible since v6i16 doesn't exist).

But what really puzzles me is why this never broke before, since even before this change, this wasn't a valid assumption (e.g. v3i32 -> v3i8 would have failed similarly before) -- is there logic upstream of this call that intends to guard against this somehow?

[srj]

I've gotten a repro case. Created https://bugs.llvm.org/show_bug.cgi?id=50709 with details. Please let me know if I should replicate details here instead.

[srj]

Since there is a bug and a repro case, I'd like to ask if this change can be rolled back until a proper fix is in order.

[RKSimon]

In D103881#2818322, @srj wrote:

Since there is a bug and a repro case, I'd like to ask if this change can be rolled back until a proper fix is in order.

It looks like @craig.topper fixed the bug with rG4017d0335a35334835bfae6fc3e258adcd9ed2dc so probably no need for a revert

[critson]

In D103881#2818801, @RKSimon wrote:

In D103881#2818322, @srj wrote:

Since there is a bug and a repro case, I'd like to ask if this change can be rolled back until a proper fix is in order.

It looks like @craig.topper fixed the bug with rG4017d0335a35334835bfae6fc3e258adcd9ed2dc so probably no need for a revert

Let me know, I am happy to revert if we need to investigate further as I have not committed anything depend on this yet.
The v6 types in particular do create some odd issues.

llvm/include/llvm/CodeGen/ValueTypes.td

 def v128i16 : ValueType<2048, 43>;  // 128 x i16 vector value
 def v256i16 : ValueType<4096, 44>;  // 256 x i16 vector value
 
 def v1i32    : ValueType<32,    45>;  //    1 x i32 vector value
 def v2i32    : ValueType<64,    46>;  //    2 x i32 vector value
 def v3i32    : ValueType<96,    47>;  //    3 x i32 vector value
 def v4i32    : ValueType<128,   48>;  //    4 x i32 vector value
 def v5i32    : ValueType<160,   49>;  //    5 x i32 vector value
-def v8i32    : ValueType<256,   50>;  //    8 x i32 vector value
-def v16i32   : ValueType<512,   51>;  //   16 x i32 vector value
-def v32i32   : ValueType<1024,  52>;  //   32 x i32 vector value
-def v64i32   : ValueType<2048,  53>;  //   64 x i32 vector value
-def v128i32  : ValueType<4096,  54>;  //  128 x i32 vector value
-def v256i32  : ValueType<8192,  55>;  //  256 x i32 vector value
-def v512i32  : ValueType<16384, 56>;  //  512 x i32 vector value
-def v1024i32 : ValueType<32768, 57>;  // 1024 x i32 vector value
-def v2048i32 : ValueType<65536, 58>;  // 2048 x i32 vector value
-
-def v1i64   : ValueType<64,    59>;  //   1 x i64 vector value
-def v2i64   : ValueType<128,   60>;  //   2 x i64 vector value
-def v4i64   : ValueType<256,   61>;  //   4 x i64 vector value
-def v8i64   : ValueType<512,   62>;  //   8 x i64 vector value
-def v16i64  : ValueType<1024,  63>;  //  16 x i64 vector value
-def v32i64  : ValueType<2048,  64>;  //  32 x i64 vector value
-def v64i64  : ValueType<4096,  65>;  //  64 x i64 vector value
-def v128i64 : ValueType<8192,  66>;  // 128 x i64 vector value
-def v256i64 : ValueType<16384, 67>;  // 256 x i64 vector value
-
-def v1i128 : ValueType<128, 68>;  //  1 x i128 vector value
-
-def v1f16    : ValueType<16,     69>;  //    1 x f16 vector value
-def v2f16    : ValueType<32,     70>;  //    2 x f16 vector value
-def v3f16    : ValueType<48,     71>;  //    3 x f16 vector value
-def v4f16    : ValueType<64,     72>;  //    4 x f16 vector value
-def v8f16    : ValueType<128,    73>;  //    8 x f16 vector value
-def v16f16   : ValueType<256,    74>;  //   16 x f16 vector value
-def v32f16   : ValueType<512,    75>;  //   32 x f16 vector value
-def v64f16   : ValueType<1024,   76>;  //   64 x f16 vector value
-def v128f16  : ValueType<2048,   77>;  //  128 x f16 vector value
-def v256f16  : ValueType<4096,   78>;  //  256 x f16 vector value
-
-def v2bf16   : ValueType<32,     79>;  //    2 x bf16 vector value
-def v3bf16   : ValueType<48,     80>;  //    3 x bf16 vector value
-def v4bf16   : ValueType<64,     81>;  //    4 x bf16 vector value
-def v8bf16   : ValueType<128,    82>;  //    8 x bf16 vector value
-def v16bf16  : ValueType<256,    83>;  //   16 x bf16 vector value
-def v32bf16  : ValueType<512,    84>;  //   32 x bf16 vector value
-def v64bf16  : ValueType<1024,   85>;  //   64 x bf16 vector value
-def v128bf16 : ValueType<2048,   86>;  //  128 x bf16 vector value
-
-def v1f32    : ValueType<32,     87>;  //    1 x f32 vector value
-def v2f32    : ValueType<64,     88>;  //    2 x f32 vector value
-def v3f32    : ValueType<96,     89>;  //    3 x f32 vector value
-def v4f32    : ValueType<128,    90>;  //    4 x f32 vector value
-def v5f32    : ValueType<160,    91>;  //    5 x f32 vector value
-def v8f32    : ValueType<256,    92>;  //    8 x f32 vector value
-def v16f32   : ValueType<512,    93>;  //   16 x f32 vector value
-def v32f32   : ValueType<1024,   94>;  //   32 x f32 vector value
-def v64f32   : ValueType<2048,   95>;  //   64 x f32 vector value
-def v128f32  : ValueType<4096,   96>;  //  128 x f32 vector value
-def v256f32  : ValueType<8192,   97>;  //  256 x f32 vector value
-def v512f32  : ValueType<16384,  98>;  //  512 x f32 vector value
-def v1024f32 : ValueType<32768,  99>;  // 1024 x f32 vector value
-def v2048f32 : ValueType<65536, 100>;  // 2048 x f32 vector value
-
-def v1f64    : ValueType<64,    101>;  //    1 x f64 vector value
-def v2f64    : ValueType<128,   102>;  //    2 x f64 vector value
-def v4f64    : ValueType<256,   103>;  //    4 x f64 vector value
-def v8f64    : ValueType<512,   104>;  //    8 x f64 vector value
-def v16f64   : ValueType<1024,  105>;  //   16 x f64 vector value
-def v32f64   : ValueType<2048,  106>;  //   32 x f64 vector value
-def v64f64   : ValueType<4096,  107>;  //   64 x f64 vector value
-def v128f64  : ValueType<8192,  108>;  //  128 x f64 vector value
-def v256f64  : ValueType<16384, 109>;  //  256 x f64 vector value
-
-def nxv1i1  : ValueType<1,  110>;  // n x  1 x i1  vector value
-def nxv2i1  : ValueType<2,  111>;  // n x  2 x i1  vector value
-def nxv4i1  : ValueType<4,  112>;  // n x  4 x i1  vector value
-def nxv8i1  : ValueType<8,  113>;  // n x  8 x i1  vector value
-def nxv16i1 : ValueType<16, 114>;  // n x 16 x i1  vector value
-def nxv32i1 : ValueType<32, 115>;  // n x 32 x i1  vector value
-def nxv64i1 : ValueType<64, 116>;  // n x 64 x i1  vector value
-
-def nxv1i8  : ValueType<8,   117>;  // n x  1 x i8  vector value
-def nxv2i8  : ValueType<16,  118>;  // n x  2 x i8  vector value
-def nxv4i8  : ValueType<32,  119>;  // n x  4 x i8  vector value
-def nxv8i8  : ValueType<64,  120>;  // n x  8 x i8  vector value
-def nxv16i8 : ValueType<128, 121>;  // n x 16 x i8  vector value
-def nxv32i8 : ValueType<256, 122>;  // n x 32 x i8  vector value
-def nxv64i8 : ValueType<512, 123>;  // n x 64 x i8  vector value
-
-def nxv1i16  : ValueType<16,  124>;  // n x  1 x i16 vector value
-def nxv2i16  : ValueType<32,  125>;  // n x  2 x i16 vector value
-def nxv4i16  : ValueType<64,  126>;  // n x  4 x i16 vector value
-def nxv8i16  : ValueType<128, 127>;  // n x  8 x i16 vector value
-def nxv16i16 : ValueType<256, 128>;  // n x 16 x i16 vector value
-def nxv32i16 : ValueType<512, 129>;  // n x 32 x i16 vector value
-
-def nxv1i32  : ValueType<32,   130>;  // n x  1 x i32 vector value
-def nxv2i32  : ValueType<64,   131>;  // n x  2 x i32 vector value
-def nxv4i32  : ValueType<128,  132>;  // n x  4 x i32 vector value
-def nxv8i32  : ValueType<256,  133>;  // n x  8 x i32 vector value
-def nxv16i32 : ValueType<512,  134>;  // n x 16 x i32 vector value
-def nxv32i32 : ValueType<1024, 135>;  // n x 32 x i32 vector value
-
-def nxv1i64  : ValueType<64,   136>;  // n x  1 x i64 vector value
-def nxv2i64  : ValueType<128,  137>;  // n x  2 x i64 vector value
-def nxv4i64  : ValueType<256,  138>;  // n x  4 x i64 vector value
-def nxv8i64  : ValueType<512,  139>;  // n x  8 x i64 vector value
-def nxv16i64 : ValueType<1024, 140>;  // n x 16 x i64 vector value
-def nxv32i64 : ValueType<2048, 141>;  // n x 32 x i64 vector value
-
-def nxv1f16  : ValueType<16,  142>;  // n x  1 x  f16 vector value
-def nxv2f16  : ValueType<32,  143>;  // n x  2 x  f16 vector value
-def nxv4f16  : ValueType<64,  144>;  // n x  4 x  f16 vector value
-def nxv8f16  : ValueType<128, 145>;  // n x  8 x  f16 vector value
-def nxv16f16 : ValueType<256, 146>;  // n x 16 x  f16 vector value
-def nxv32f16 : ValueType<512, 147>;  // n x 32 x  f16 vector value
-
-def nxv1bf16 : ValueType<16,  148>;  // n x  1 x bf16 vector value
-def nxv2bf16 : ValueType<32,  149>;  // n x  2 x bf16 vector value
-def nxv4bf16 : ValueType<64,  150>;  // n x  4 x bf16 vector value
-def nxv8bf16 : ValueType<128, 151>;  // n x  8 x bf16 vector value
-
-def nxv1f32  : ValueType<32,  152>;  // n x  1 x  f32 vector value
-def nxv2f32  : ValueType<64,  153>;  // n x  2 x  f32 vector value
-def nxv4f32  : ValueType<128, 154>;  // n x  4 x  f32 vector value
-def nxv8f32  : ValueType<256, 155>;  // n x  8 x  f32 vector value
-def nxv16f32 : ValueType<512, 156>;  // n x 16 x  f32 vector value
-
-def nxv1f64  : ValueType<64,  157>;  // n x  1 x  f64 vector value
-def nxv2f64  : ValueType<128, 158>;  // n x  2 x  f64 vector value
-def nxv4f64  : ValueType<256, 159>;  // n x  4 x  f64 vector value
-def nxv8f64  : ValueType<512, 160>;  // n x  8 x  f64 vector value
-
-def x86mmx    : ValueType<64,   161>;  // X86 MMX value
-def FlagVT    : ValueType<0,    162>;  // Pre-RA sched glue
-def isVoid    : ValueType<0,    163>;  // Produces no value
-def untyped   : ValueType<8,    164>;  // Produces an untyped value
-def funcref   : ValueType<0,    165>;  // WebAssembly's funcref type
-def externref : ValueType<0,    166>;  // WebAssembly's externref type
-def x86amx    : ValueType<8192, 167>;  // X86 AMX value
+def v6i32    : ValueType<192,   50>;  //    6 x i32 vector value
+def v7i32    : ValueType<224,   51>;  //    7 x i32 vector value
+def v8i32    : ValueType<256,   52>;  //    8 x i32 vector value
+def v16i32   : ValueType<512,   53>;  //   16 x i32 vector value
+def v32i32   : ValueType<1024,  54>;  //   32 x i32 vector value
+def v64i32   : ValueType<2048,  55>;  //   64 x i32 vector value
+def v128i32  : ValueType<4096,  56>;  //  128 x i32 vector value
+def v256i32  : ValueType<8192,  57>;  //  256 x i32 vector value
+def v512i32  : ValueType<16384, 58>;  //  512 x i32 vector value
+def v1024i32 : ValueType<32768, 59>;  // 1024 x i32 vector value
+def v2048i32 : ValueType<65536, 60>;  // 2048 x i32 vector value
+
+def v1i64   : ValueType<64,    61>;  //   1 x i64 vector value
+def v2i64   : ValueType<128,   62>;  //   2 x i64 vector value
+def v4i64   : ValueType<256,   63>;  //   4 x i64 vector value
+def v8i64   : ValueType<512,   64>;  //   8 x i64 vector value
+def v16i64  : ValueType<1024,  65>;  //  16 x i64 vector value
+def v32i64  : ValueType<2048,  66>;  //  32 x i64 vector value
+def v64i64  : ValueType<4096,  67>;  //  64 x i64 vector value
+def v128i64 : ValueType<8192,  68>;  // 128 x i64 vector value
+def v256i64 : ValueType<16384, 69>;  // 256 x i64 vector value
+
+def v1i128 : ValueType<128, 70>;  //  1 x i128 vector value
+
+def v1f16    : ValueType<16,     71>;  //    1 x f16 vector value
+def v2f16    : ValueType<32,     72>;  //    2 x f16 vector value
+def v3f16    : ValueType<48,     73>;  //    3 x f16 vector value
+def v4f16    : ValueType<64,     74>;  //    4 x f16 vector value
+def v8f16    : ValueType<128,    75>;  //    8 x f16 vector value
+def v16f16   : ValueType<256,    76>;  //   16 x f16 vector value
+def v32f16   : ValueType<512,    77>;  //   32 x f16 vector value
+def v64f16   : ValueType<1024,   78>;  //   64 x f16 vector value
+def v128f16  : ValueType<2048,   79>;  //  128 x f16 vector value
+def v256f16  : ValueType<4096,   80>;  //  256 x f16 vector value
+
+def v2bf16   : ValueType<32,     81>;  //    2 x bf16 vector value
+def v3bf16   : ValueType<48,     82>;  //    3 x bf16 vector value
+def v4bf16   : ValueType<64,     83>;  //    4 x bf16 vector value
+def v8bf16   : ValueType<128,    84>;  //    8 x bf16 vector value
+def v16bf16  : ValueType<256,    85>;  //   16 x bf16 vector value
+def v32bf16  : ValueType<512,    86>;  //   32 x bf16 vector value
+def v64bf16  : ValueType<1024,   87>;  //   64 x bf16 vector value
+def v128bf16 : ValueType<2048,   88>;  //  128 x bf16 vector value
+
+def v1f32    : ValueType<32,     89>;  //    1 x f32 vector value
+def v2f32    : ValueType<64,     90>;  //    2 x f32 vector value
+def v3f32    : ValueType<96,     91>;  //    3 x f32 vector value
+def v4f32    : ValueType<128,    92>;  //    4 x f32 vector value
+def v5f32    : ValueType<160,    93>;  //    5 x f32 vector value
+def v6f32    : ValueType<192,    94>;  //    6 x f32 vector value
+def v7f32    : ValueType<224,    95>;  //    7 x f32 vector value
+def v8f32    : ValueType<256,    96>;  //    8 x f32 vector value
+def v16f32   : ValueType<512,    97>;  //   16 x f32 vector value
+def v32f32   : ValueType<1024,   98>;  //   32 x f32 vector value
+def v64f32   : ValueType<2048,   99>;  //   64 x f32 vector value
+def v128f32  : ValueType<4096,  100>;  //  128 x f32 vector value
+def v256f32  : ValueType<8192,  101>;  //  256 x f32 vector value
+def v512f32  : ValueType<16384, 102>;  //  512 x f32 vector value
+def v1024f32 : ValueType<32768, 103>;  // 1024 x f32 vector value
+def v2048f32 : ValueType<65536, 104>;  // 2048 x f32 vector value
+
+def v1f64    : ValueType<64,    105>;  //    1 x f64 vector value
+def v2f64    : ValueType<128,   106>;  //    2 x f64 vector value
+def v4f64    : ValueType<256,   107>;  //    4 x f64 vector value
+def v8f64    : ValueType<512,   108>;  //    8 x f64 vector value
+def v16f64   : ValueType<1024,  109>;  //   16 x f64 vector value
+def v32f64   : ValueType<2048,  110>;  //   32 x f64 vector value
+def v64f64   : ValueType<4096,  111>;  //   64 x f64 vector value
+def v128f64  : ValueType<8192,  112>;  //  128 x f64 vector value
+def v256f64  : ValueType<16384, 113>;  //  256 x f64 vector value
+
+def nxv1i1  : ValueType<1,  114>;  // n x  1 x i1  vector value
+def nxv2i1  : ValueType<2,  115>;  // n x  2 x i1  vector value
+def nxv4i1  : ValueType<4,  116>;  // n x  4 x i1  vector value
+def nxv8i1  : ValueType<8,  117>;  // n x  8 x i1  vector value
+def nxv16i1 : ValueType<16, 118>;  // n x 16 x i1  vector value
+def nxv32i1 : ValueType<32, 119>;  // n x 32 x i1  vector value
+def nxv64i1 : ValueType<64, 120>;  // n x 64 x i1  vector value
+
+def nxv1i8  : ValueType<8,   121>;  // n x  1 x i8  vector value
+def nxv2i8  : ValueType<16,  122>;  // n x  2 x i8  vector value
+def nxv4i8  : ValueType<32,  123>;  // n x  4 x i8  vector value
+def nxv8i8  : ValueType<64,  124>;  // n x  8 x i8  vector value
+def nxv16i8 : ValueType<128, 125>;  // n x 16 x i8  vector value
+def nxv32i8 : ValueType<256, 126>;  // n x 32 x i8  vector value
+def nxv64i8 : ValueType<512, 127>;  // n x 64 x i8  vector value
+
+def nxv1i16  : ValueType<16,  128>;  // n x  1 x i16 vector value
+def nxv2i16  : ValueType<32,  129>;  // n x  2 x i16 vector value
+def nxv4i16  : ValueType<64,  130>;  // n x  4 x i16 vector value
+def nxv8i16  : ValueType<128, 131>;  // n x  8 x i16 vector value
+def nxv16i16 : ValueType<256, 132>;  // n x 16 x i16 vector value
+def nxv32i16 : ValueType<512, 133>;  // n x 32 x i16 vector value
+
+def nxv1i32  : ValueType<32,   134>;  // n x  1 x i32 vector value
+def nxv2i32  : ValueType<64,   135>;  // n x  2 x i32 vector value
+def nxv4i32  : ValueType<128,  136>;  // n x  4 x i32 vector value
+def nxv8i32  : ValueType<256,  137>;  // n x  8 x i32 vector value
+def nxv16i32 : ValueType<512,  138>;  // n x 16 x i32 vector value
+def nxv32i32 : ValueType<1024, 139>;  // n x 32 x i32 vector value
+
+def nxv1i64  : ValueType<64,   140>;  // n x  1 x i64 vector value
+def nxv2i64  : ValueType<128,  141>;  // n x  2 x i64 vector value
+def nxv4i64  : ValueType<256,  142>;  // n x  4 x i64 vector value
+def nxv8i64  : ValueType<512,  143>;  // n x  8 x i64 vector value
+def nxv16i64 : ValueType<1024, 144>;  // n x 16 x i64 vector value
+def nxv32i64 : ValueType<2048, 145>;  // n x 32 x i64 vector value
+
+def nxv1f16  : ValueType<16,  146>;  // n x  1 x  f16 vector value
+def nxv2f16  : ValueType<32,  147>;  // n x  2 x  f16 vector value
+def nxv4f16  : ValueType<64,  148>;  // n x  4 x  f16 vector value
+def nxv8f16  : ValueType<128, 149>;  // n x  8 x  f16 vector value
+def nxv16f16 : ValueType<256, 150>;  // n x 16 x  f16 vector value
+def nxv32f16 : ValueType<512, 151>;  // n x 32 x  f16 vector value
+
+def nxv1bf16 : ValueType<16,  152>;  // n x  1 x bf16 vector value
+def nxv2bf16 : ValueType<32,  153>;  // n x  2 x bf16 vector value
+def nxv4bf16 : ValueType<64,  154>;  // n x  4 x bf16 vector value
+def nxv8bf16 : ValueType<128, 155>;  // n x  8 x bf16 vector value
+
+def nxv1f32  : ValueType<32,  156>;  // n x  1 x  f32 vector value
+def nxv2f32  : ValueType<64,  157>;  // n x  2 x  f32 vector value
+def nxv4f32  : ValueType<128, 158>;  // n x  4 x  f32 vector value
+def nxv8f32  : ValueType<256, 159>;  // n x  8 x  f32 vector value
+def nxv16f32 : ValueType<512, 160>;  // n x 16 x  f32 vector value
+
+def nxv1f64  : ValueType<64,  161>;  // n x  1 x  f64 vector value
+def nxv2f64  : ValueType<128, 162>;  // n x  2 x  f64 vector value
+def nxv4f64  : ValueType<256, 163>;  // n x  4 x  f64 vector value
+def nxv8f64  : ValueType<512, 164>;  // n x  8 x  f64 vector value
+
+def x86mmx    : ValueType<64,   165>;  // X86 MMX value
+def FlagVT    : ValueType<0,    166>;  // Pre-RA sched glue
+def isVoid    : ValueType<0,    167>;  // Produces no value
+def untyped   : ValueType<8,    168>;  // Produces an untyped value
+def funcref   : ValueType<0,    169>;  // WebAssembly's funcref type
+def externref : ValueType<0,    170>;  // WebAssembly's externref type
+def x86amx    : ValueType<8192, 171>;  // X86 AMX value
 
 
 def token      : ValueType<0, 248>;  // TokenTy
 def MetadataVT : ValueType<0, 249>;  // Metadata
 
 // Pseudo valuetype mapped to the current pointer size to any address space.
 // Should only be used in TableGen.
 def iPTRAny    : ValueType<0, 250>;

llvm/include/llvm/Support/MachineValueType.h

 //===- Support/MachineValueType.h - Machine-Level types ---------*- C++ -*-===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
 // This file defines the set of machine-level target independent types which
enum SimpleValueType : uint8_t {
       v128i16 = 43, // 128 x i16
       v256i16 = 44, // 256 x i16
 
       v1i32 = 45,    //    1 x i32
       v2i32 = 46,    //    2 x i32
       v3i32 = 47,    //    3 x i32
       v4i32 = 48,    //    4 x i32
       v5i32 = 49,    //    5 x i32
-      v8i32 = 50,    //    8 x i32
-      v16i32 = 51,   //   16 x i32
-      v32i32 = 52,   //   32 x i32
-      v64i32 = 53,   //   64 x i32
-      v128i32 = 54,  //  128 x i32
-      v256i32 = 55,  //  256 x i32
-      v512i32 = 56,  //  512 x i32
-      v1024i32 = 57, // 1024 x i32
-      v2048i32 = 58, // 2048 x i32
-
-      v1i64 = 59,   //   1 x i64
-      v2i64 = 60,   //   2 x i64
-      v4i64 = 61,   //   4 x i64
-      v8i64 = 62,   //   8 x i64
-      v16i64 = 63,  //  16 x i64
-      v32i64 = 64,  //  32 x i64
-      v64i64 = 65,  //  64 x i64
-      v128i64 = 66, // 128 x i64
-      v256i64 = 67, // 256 x i64
+      v6i32 = 50,    //    6 x i32
+      v7i32 = 51,    //    7 x i32
+      v8i32 = 52,    //    8 x i32
+      v16i32 = 53,   //   16 x i32
+      v32i32 = 54,   //   32 x i32
+      v64i32 = 55,   //   64 x i32
+      v128i32 = 56,  //  128 x i32
+      v256i32 = 57,  //  256 x i32
+      v512i32 = 58,  //  512 x i32
+      v1024i32 = 59, // 1024 x i32
+      v2048i32 = 60, // 2048 x i32
+
+      v1i64 = 61,   //   1 x i64
+      v2i64 = 62,   //   2 x i64
+      v4i64 = 63,   //   4 x i64
+      v8i64 = 64,   //   8 x i64
+      v16i64 = 65,  //  16 x i64
+      v32i64 = 66,  //  32 x i64
+      v64i64 = 67,  //  64 x i64
+      v128i64 = 68, // 128 x i64
+      v256i64 = 69, // 256 x i64
 
-      v1i128 = 68, //  1 x i128
+      v1i128 = 70, //  1 x i128
 
       FIRST_INTEGER_FIXEDLEN_VECTOR_VALUETYPE = v1i1,
       LAST_INTEGER_FIXEDLEN_VECTOR_VALUETYPE = v1i128,
 
-      v1f16 = 69,   //    1 x f16
-      v2f16 = 70,   //    2 x f16
-      v3f16 = 71,   //    3 x f16
-      v4f16 = 72,   //    4 x f16
-      v8f16 = 73,   //    8 x f16
-      v16f16 = 74,  //   16 x f16
-      v32f16 = 75,  //   32 x f16
-      v64f16 = 76,  //   64 x f16
-      v128f16 = 77, //  128 x f16
-      v256f16 = 78, //  256 x f16
-
-      v2bf16 = 79,   //    2 x bf16
-      v3bf16 = 80,   //    3 x bf16
-      v4bf16 = 81,   //    4 x bf16
-      v8bf16 = 82,   //    8 x bf16
-      v16bf16 = 83,  //   16 x bf16
-      v32bf16 = 84,  //   32 x bf16
-      v64bf16 = 85,  //   64 x bf16
-      v128bf16 = 86, //  128 x bf16
-
-      v1f32 = 87,     //    1 x f32
-      v2f32 = 88,     //    2 x f32
-      v3f32 = 89,     //    3 x f32
-      v4f32 = 90,     //    4 x f32
-      v5f32 = 91,     //    5 x f32
-      v8f32 = 92,     //    8 x f32
-      v16f32 = 93,    //   16 x f32
-      v32f32 = 94,    //   32 x f32
-      v64f32 = 95,    //   64 x f32
-      v128f32 = 96,   //  128 x f32
-      v256f32 = 97,   //  256 x f32
-      v512f32 = 98,   //  512 x f32
-      v1024f32 = 99,  // 1024 x f32
-      v2048f32 = 100, // 2048 x f32
-
-      v1f64 = 101,   //    1 x f64
-      v2f64 = 102,   //    2 x f64
-      v4f64 = 103,   //    4 x f64
-      v8f64 = 104,   //    8 x f64
-      v16f64 = 105,  //   16 x f64
-      v32f64 = 106,  //   32 x f64
-      v64f64 = 107,  //   64 x f64
-      v128f64 = 108, //  128 x f64
-      v256f64 = 109, //  256 x f64
+      v1f16 = 71,   //    1 x f16
+      v2f16 = 72,   //    2 x f16
+      v3f16 = 73,   //    3 x f16
+      v4f16 = 74,   //    4 x f16
+      v8f16 = 75,   //    8 x f16
+      v16f16 = 76,  //   16 x f16
+      v32f16 = 77,  //   32 x f16
+      v64f16 = 78,  //   64 x f16
+      v128f16 = 79, //  128 x f16
+      v256f16 = 80, //  256 x f16
+
+      v2bf16 = 81,   //    2 x bf16
+      v3bf16 = 82,   //    3 x bf16
+      v4bf16 = 83,   //    4 x bf16
+      v8bf16 = 84,   //    8 x bf16
+      v16bf16 = 85,  //   16 x bf16
+      v32bf16 = 86,  //   32 x bf16
+      v64bf16 = 87,  //   64 x bf16
+      v128bf16 = 88, //  128 x bf16
+
+      v1f32 = 89,     //    1 x f32
+      v2f32 = 90,     //    2 x f32
+      v3f32 = 91,     //    3 x f32
+      v4f32 = 92,     //    4 x f32
+      v5f32 = 93,     //    5 x f32
+      v6f32 = 94,     //    6 x f32
+      v7f32 = 95,     //    7 x f32
+      v8f32 = 96,     //    8 x f32
+      v16f32 = 97,    //   16 x f32
+      v32f32 = 98,    //   32 x f32
+      v64f32 = 99,    //   64 x f32
+      v128f32 = 100,  //  128 x f32
+      v256f32 = 101,  //  256 x f32
+      v512f32 = 102,  //  512 x f32
+      v1024f32 = 103, // 1024 x f32
+      v2048f32 = 104, // 2048 x f32
+
+      v1f64 = 105,   //    1 x f64
+      v2f64 = 106,   //    2 x f64
+      v4f64 = 107,   //    4 x f64
+      v8f64 = 108,   //    8 x f64
+      v16f64 = 109,  //   16 x f64
+      v32f64 = 110,  //   32 x f64
+      v64f64 = 111,  //   64 x f64
+      v128f64 = 112, //  128 x f64
+      v256f64 = 113, //  256 x f64
 
       FIRST_FP_FIXEDLEN_VECTOR_VALUETYPE = v1f16,
       LAST_FP_FIXEDLEN_VECTOR_VALUETYPE = v256f64,
 
       FIRST_FIXEDLEN_VECTOR_VALUETYPE = v1i1,
       LAST_FIXEDLEN_VECTOR_VALUETYPE = v256f64,
 
-      nxv1i1 = 110,  // n x  1 x i1
-      nxv2i1 = 111,  // n x  2 x i1
-      nxv4i1 = 112,  // n x  4 x i1
-      nxv8i1 = 113,  // n x  8 x i1
-      nxv16i1 = 114, // n x 16 x i1
-      nxv32i1 = 115, // n x 32 x i1
-      nxv64i1 = 116, // n x 64 x i1
-
-      nxv1i8 = 117,  // n x  1 x i8
-      nxv2i8 = 118,  // n x  2 x i8
-      nxv4i8 = 119,  // n x  4 x i8
-      nxv8i8 = 120,  // n x  8 x i8
-      nxv16i8 = 121, // n x 16 x i8
-      nxv32i8 = 122, // n x 32 x i8
-      nxv64i8 = 123, // n x 64 x i8
-
-      nxv1i16 = 124,  // n x  1 x i16
-      nxv2i16 = 125,  // n x  2 x i16
-      nxv4i16 = 126,  // n x  4 x i16
-      nxv8i16 = 127,  // n x  8 x i16
-      nxv16i16 = 128, // n x 16 x i16
-      nxv32i16 = 129, // n x 32 x i16
-
-      nxv1i32 = 130,  // n x  1 x i32
-      nxv2i32 = 131,  // n x  2 x i32
-      nxv4i32 = 132,  // n x  4 x i32
-      nxv8i32 = 133,  // n x  8 x i32
-      nxv16i32 = 134, // n x 16 x i32
-      nxv32i32 = 135, // n x 32 x i32
-
-      nxv1i64 = 136,  // n x  1 x i64
-      nxv2i64 = 137,  // n x  2 x i64
-      nxv4i64 = 138,  // n x  4 x i64
-      nxv8i64 = 139,  // n x  8 x i64
-      nxv16i64 = 140, // n x 16 x i64
-      nxv32i64 = 141, // n x 32 x i64
+      nxv1i1 = 114,  // n x  1 x i1
+      nxv2i1 = 115,  // n x  2 x i1
+      nxv4i1 = 116,  // n x  4 x i1
+      nxv8i1 = 117,  // n x  8 x i1
+      nxv16i1 = 118, // n x 16 x i1
+      nxv32i1 = 119, // n x 32 x i1
+      nxv64i1 = 120, // n x 64 x i1
+
+      nxv1i8 = 121,  // n x  1 x i8
+      nxv2i8 = 122,  // n x  2 x i8
+      nxv4i8 = 123,  // n x  4 x i8
+      nxv8i8 = 124,  // n x  8 x i8
+      nxv16i8 = 125, // n x 16 x i8
+      nxv32i8 = 126, // n x 32 x i8
+      nxv64i8 = 127, // n x 64 x i8
+
+      nxv1i16 = 128,  // n x  1 x i16
+      nxv2i16 = 129,  // n x  2 x i16
+      nxv4i16 = 130,  // n x  4 x i16
+      nxv8i16 = 131,  // n x  8 x i16
+      nxv16i16 = 132, // n x 16 x i16
+      nxv32i16 = 133, // n x 32 x i16
+
+      nxv1i32 = 134,  // n x  1 x i32
+      nxv2i32 = 135,  // n x  2 x i32
+      nxv4i32 = 136,  // n x  4 x i32
+      nxv8i32 = 137,  // n x  8 x i32
+      nxv16i32 = 138, // n x 16 x i32
+      nxv32i32 = 139, // n x 32 x i32
+
+      nxv1i64 = 140,  // n x  1 x i64
+      nxv2i64 = 141,  // n x  2 x i64
+      nxv4i64 = 142,  // n x  4 x i64
+      nxv8i64 = 143,  // n x  8 x i64
+      nxv16i64 = 144, // n x 16 x i64
+      nxv32i64 = 145, // n x 32 x i64
 
       FIRST_INTEGER_SCALABLE_VECTOR_VALUETYPE = nxv1i1,
       LAST_INTEGER_SCALABLE_VECTOR_VALUETYPE = nxv32i64,
 
-      nxv1f16 = 142,  // n x  1 x f16
-      nxv2f16 = 143,  // n x  2 x f16
-      nxv4f16 = 144,  // n x  4 x f16
-      nxv8f16 = 145,  // n x  8 x f16
-      nxv16f16 = 146, // n x 16 x f16
-      nxv32f16 = 147, // n x 32 x f16
-
-      nxv1bf16 = 148, // n x  1 x bf16
-      nxv2bf16 = 149, // n x  2 x bf16
-      nxv4bf16 = 150, // n x  4 x bf16
-      nxv8bf16 = 151, // n x  8 x bf16
-
-      nxv1f32 = 152,  // n x  1 x f32
-      nxv2f32 = 153,  // n x  2 x f32
-      nxv4f32 = 154,  // n x  4 x f32
-      nxv8f32 = 155,  // n x  8 x f32
-      nxv16f32 = 156, // n x 16 x f32
-
-      nxv1f64 = 157, // n x  1 x f64
-      nxv2f64 = 158, // n x  2 x f64
-      nxv4f64 = 159, // n x  4 x f64
-      nxv8f64 = 160, // n x  8 x f64
+      nxv1f16 = 146,  // n x  1 x f16
+      nxv2f16 = 147,  // n x  2 x f16
+      nxv4f16 = 148,  // n x  4 x f16
+      nxv8f16 = 149,  // n x  8 x f16
+      nxv16f16 = 150, // n x 16 x f16
+      nxv32f16 = 151, // n x 32 x f16
+
+      nxv1bf16 = 152, // n x  1 x bf16
+      nxv2bf16 = 153, // n x  2 x bf16
+      nxv4bf16 = 154, // n x  4 x bf16
+      nxv8bf16 = 155, // n x  8 x bf16
+
+      nxv1f32 = 156,  // n x  1 x f32
+      nxv2f32 = 157,  // n x  2 x f32
+      nxv4f32 = 158,  // n x  4 x f32
+      nxv8f32 = 159,  // n x  8 x f32
+      nxv16f32 = 160, // n x 16 x f32
+
+      nxv1f64 = 161, // n x  1 x f64
+      nxv2f64 = 162, // n x  2 x f64
+      nxv4f64 = 163, // n x  4 x f64
+      nxv8f64 = 164, // n x  8 x f64
 
       FIRST_FP_SCALABLE_VECTOR_VALUETYPE = nxv1f16,
       LAST_FP_SCALABLE_VECTOR_VALUETYPE = nxv8f64,
 
       FIRST_SCALABLE_VECTOR_VALUETYPE = nxv1i1,
       LAST_SCALABLE_VECTOR_VALUETYPE = nxv8f64,
 
       FIRST_VECTOR_VALUETYPE = v1i1,
       LAST_VECTOR_VALUETYPE = nxv8f64,
 
-      x86mmx = 161, // This is an X86 MMX value
+      x86mmx = 165, // This is an X86 MMX value
 
-      Glue = 162, // This glues nodes together during pre-RA sched
+      Glue = 166, // This glues nodes together during pre-RA sched
 
-      isVoid = 163, // This has no value
+      isVoid = 167, // This has no value
 
-      Untyped = 164, // This value takes a register, but has
+      Untyped = 168, // This value takes a register, but has
                      // unspecified type.  The register class
                      // will be determined by the opcode.
 
-      funcref = 165,   // WebAssembly's funcref type
-      externref = 166, // WebAssembly's externref type
-      x86amx = 167,    // This is an X86 AMX value
+      funcref = 169,   // WebAssembly's funcref type
+      externref = 170, // WebAssembly's externref type
+      x86amx = 171,    // This is an X86 AMX value
 
       FIRST_VALUETYPE = 1,     // This is always the beginning of the list.
       LAST_VALUETYPE = x86amx, // This always remains at the end of the list.
       VALUETYPE_SIZE = LAST_VALUETYPE + 1,
 
       // This is the current maximum for LAST_VALUETYPE.
       // MVT::MAX_ALLOWED_VALUETYPE is used for asserts and to size bit vectors
       // This value must be a multiple of 32.
MVT getVectorElementType() const {
       case nxv8i16:
       case nxv16i16:
       case nxv32i16: return i16;
       case v1i32:
       case v2i32:
       case v3i32:
       case v4i32:
       case v5i32:
+      case v6i32:
+      case v7i32:
       case v8i32:
       case v16i32:
       case v32i32:
       case v64i32:
       case v128i32:
       case v256i32:
       case v512i32:
       case v1024i32:
MVT getVectorElementType() const {
       case nxv2bf16:
       case nxv4bf16:
       case nxv8bf16: return bf16;
       case v1f32:
       case v2f32:
       case v3f32:
       case v4f32:
       case v5f32:
+      case v6f32:
+      case v7f32:
       case v8f32:
       case v16f32:
       case v32f32:
       case v64f32:
       case v128f32:
       case v256f32:
       case v512f32:
       case v1024f32:
unsigned getVectorMinNumElements() const {
       case nxv8i8:
       case nxv8i16:
       case nxv8i32:
       case nxv8i64:
       case nxv8f16:
       case nxv8bf16:
       case nxv8f32:
       case nxv8f64: return 8;
+      case v7i32:
+      case v7f32: return 7;
+      case v6i32:
+      case v6f32: return 6;
       case v5i32:
       case v5f32: return 5;
       case v4i1:
       case v4i8:
       case v4i16:
       case v4i32:
       case v4i64:
       case v4f16:
TypeSize getSizeInBits() const {
       case nxv4i32:
       case nxv2i64:
       case nxv8f16:
       case nxv8bf16:
       case nxv4f32:
       case nxv2f64: return TypeSize::Scalable(128);
       case v5i32:
       case v5f32: return TypeSize::Fixed(160);
+      case v6i32:
+      case v6f32: return TypeSize::Fixed(192);
+      case v7i32:
+      case v7f32: return TypeSize::Fixed(224);
       case v256i1:
       case v32i8:
       case v16i16:
       case v8i32:
       case v4i64:
       case v16f16:
       case v16bf16:
       case v8f32:
static MVT getVectorVT(MVT VT, unsigned NumElements) {
         if (NumElements == 256) return MVT::v256i16;
         break;
       case MVT::i32:
         if (NumElements == 1)    return MVT::v1i32;
         if (NumElements == 2)    return MVT::v2i32;
         if (NumElements == 3)    return MVT::v3i32;
         if (NumElements == 4)    return MVT::v4i32;
         if (NumElements == 5)    return MVT::v5i32;
+        if (NumElements == 6)    return MVT::v6i32;
+        if (NumElements == 7)    return MVT::v7i32;
         if (NumElements == 8)    return MVT::v8i32;
         if (NumElements == 16)   return MVT::v16i32;
         if (NumElements == 32)   return MVT::v32i32;
         if (NumElements == 64)   return MVT::v64i32;
         if (NumElements == 128)  return MVT::v128i32;
         if (NumElements == 256)  return MVT::v256i32;
         if (NumElements == 512)  return MVT::v512i32;
         if (NumElements == 1024) return MVT::v1024i32;
static MVT getVectorVT(MVT VT, unsigned NumElements) {
         if (NumElements == 128) return MVT::v128bf16;
         break;
       case MVT::f32:
         if (NumElements == 1)    return MVT::v1f32;
         if (NumElements == 2)    return MVT::v2f32;
         if (NumElements == 3)    return MVT::v3f32;
         if (NumElements == 4)    return MVT::v4f32;
         if (NumElements == 5)    return MVT::v5f32;
+        if (NumElements == 6)    return MVT::v6f32;
+        if (NumElements == 7)    return MVT::v7f32;
         if (NumElements == 8)    return MVT::v8f32;
         if (NumElements == 16)   return MVT::v16f32;
         if (NumElements == 32)   return MVT::v32f32;
         if (NumElements == 64)   return MVT::v64f32;
         if (NumElements == 128)  return MVT::v128f32;
         if (NumElements == 256)  return MVT::v256f32;
         if (NumElements == 512)  return MVT::v512f32;
         if (NumElements == 1024) return MVT::v1024f32;

llvm/lib/CodeGen/ValueTypes.cpp

Type *EVT::getTypeForEVT(LLVMContext &Context) const {
   case MVT::v2i32:
     return FixedVectorType::get(Type::getInt32Ty(Context), 2);
   case MVT::v3i32:
     return FixedVectorType::get(Type::getInt32Ty(Context), 3);
   case MVT::v4i32:
     return FixedVectorType::get(Type::getInt32Ty(Context), 4);
   case MVT::v5i32:
     return FixedVectorType::get(Type::getInt32Ty(Context), 5);
+  case MVT::v6i32:
+    return FixedVectorType::get(Type::getInt32Ty(Context), 6);
+  case MVT::v7i32:
+    return FixedVectorType::get(Type::getInt32Ty(Context), 7);
   case MVT::v8i32:
     return FixedVectorType::get(Type::getInt32Ty(Context), 8);
   case MVT::v16i32:
     return FixedVectorType::get(Type::getInt32Ty(Context), 16);
   case MVT::v32i32:
     return FixedVectorType::get(Type::getInt32Ty(Context), 32);
   case MVT::v64i32:
     return FixedVectorType::get(Type::getInt32Ty(Context), 64);
Type *EVT::getTypeForEVT(LLVMContext &Context) const {
   case MVT::v2f32:
     return FixedVectorType::get(Type::getFloatTy(Context), 2);
   case MVT::v3f32:
     return FixedVectorType::get(Type::getFloatTy(Context), 3);
   case MVT::v4f32:
     return FixedVectorType::get(Type::getFloatTy(Context), 4);
   case MVT::v5f32:
     return FixedVectorType::get(Type::getFloatTy(Context), 5);
+  case MVT::v6f32:
+    return FixedVectorType::get(Type::getFloatTy(Context), 6);
+  case MVT::v7f32:
+    return FixedVectorType::get(Type::getFloatTy(Context), 7);
   case MVT::v8f32:
     return FixedVectorType::get(Type::getFloatTy(Context), 8);
   case MVT::v16f32:
     return FixedVectorType::get(Type::getFloatTy(Context), 16);
   case MVT::v32f32:
     return FixedVectorType::get(Type::getFloatTy(Context), 32);
   case MVT::v64f32:
     return FixedVectorType::get(Type::getFloatTy(Context), 64);

llvm/utils/TableGen/CodeGenTarget.cpp

 //===- CodeGenTarget.cpp - CodeGen Target Class Wrapper -------------------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 //
 // This class wraps target description classes used by the various code
StringRef llvm::getEnumName(MVT::SimpleValueType T) {
   case MVT::v32i16:   return "MVT::v32i16";
   case MVT::v64i16:   return "MVT::v64i16";
   case MVT::v128i16:  return "MVT::v128i16";
   case MVT::v1i32:    return "MVT::v1i32";
   case MVT::v2i32:    return "MVT::v2i32";
   case MVT::v3i32:    return "MVT::v3i32";
   case MVT::v4i32:    return "MVT::v4i32";
   case MVT::v5i32:    return "MVT::v5i32";
+  case MVT::v6i32:    return "MVT::v6i32";
+  case MVT::v7i32:    return "MVT::v7i32";
   case MVT::v8i32:    return "MVT::v8i32";
   case MVT::v16i32:   return "MVT::v16i32";
   case MVT::v32i32:   return "MVT::v32i32";
   case MVT::v64i32:   return "MVT::v64i32";
   case MVT::v128i32:  return "MVT::v128i32";
   case MVT::v256i32:  return "MVT::v256i32";
   case MVT::v512i32:  return "MVT::v512i32";
   case MVT::v1024i32: return "MVT::v1024i32";
StringRef llvm::getEnumName(MVT::SimpleValueType T) {
   case MVT::v32bf16:  return "MVT::v32bf16";
   case MVT::v64bf16:  return "MVT::v64bf16";
   case MVT::v128bf16: return "MVT::v128bf16";
   case MVT::v1f32:    return "MVT::v1f32";
   case MVT::v2f32:    return "MVT::v2f32";
   case MVT::v3f32:    return "MVT::v3f32";
   case MVT::v4f32:    return "MVT::v4f32";
   case MVT::v5f32:    return "MVT::v5f32";
+  case MVT::v6f32:    return "MVT::v6f32";
+  case MVT::v7f32:    return "MVT::v7f32";
   case MVT::v8f32:    return "MVT::v8f32";
   case MVT::v16f32:   return "MVT::v16f32";
   case MVT::v32f32:   return "MVT::v32f32";
   case MVT::v64f32:   return "MVT::v64f32";
   case MVT::v128f32:  return "MVT::v128f32";
   case MVT::v256f32:  return "MVT::v256f32";
   case MVT::v512f32:  return "MVT::v512f32";
   case MVT::v1024f32: return "MVT::v1024f32";