ukernels: stop abusing signless as signed (#15338)

In MLIR we are chronically abusing signless types as signed. Ongoing
discussion at #15241, etc. At least in ukernels we are insulated enough
from the compiler code that we don't have to replicate the same
confusion. So, just because `linalg` named matmul ops want `int8 x int8
-> int32` matmuls to be represented as `i8 x i8 -> i32` even though
these are signless types and this is silently interpreting these as
signed, we didn't have to let that lower to a `_i8i8i32` ukernel, we
could instead more properly represent that as a `_s8s8s32` ukernel.

Fixing this now paves the way for unsigned LHS/RHS support in ukernels,
a step in #15158, without adding more tech debt.

The bulk of is PR is just a regex substitution:

```
find runtime/src/ compiler/src/ -type f | xargs sed -i 's/IREE_UK_FLAG_MMT4D_TYPE_I8I8I32/IREE_UK_FLAG_MMT4D_TYPE_S8S8S32/g;s/iree_uk_mmt4d_type_i8i8i32/iree_uk_mmt4d_type_s8s8s32/g;s/iree_uk_mmt4d_tile_i8i8i32/iree_uk_mmt4d_tile_s8s8s32/g;s/iree_mmt4d_reference_innerloop_i8i8i32/iree_mmt4d_reference_innerloop_s8s8s32/g'
```

compiler/src/iree/compiler/Codegen/LLVMCPU/LLVMCPULowerToUKernels.cpp

@@ -120,7 +120,7 @@ matchDAGForUKernel(RewriterBase &rewriter, linalg::Mmt4DOp op,
   uint32_t flags = 0;
   if (lhsElemType.isSignlessInteger(8) && rhsElemType.isSignlessInteger(8) &&
       outElemType.isSignlessInteger(32)) {
-    flags = IREE_UK_FLAG_MMT4D_TYPE_I8I8I32;
+    flags = IREE_UK_FLAG_MMT4D_TYPE_S8S8S32;
   } else if (lhsElemType.isF32() && rhsElemType.isF32() &&
              outElemType.isF32()) {
     flags = IREE_UK_FLAG_MMT4D_TYPE_F32F32F32;

runtime/src/iree/builtins/ukernel/arch/arm_64/mmt4d_arm_64.c

@@ -200,7 +200,7 @@ IREE_UK_MMT4D_TILE_FUNC_IMPL_FOR_M0_1_2_4_8(
     iree_uk_mmt4d_tile_f16f16f16_4x8x1_arm_64,
     iree_uk_mmt4d_tile_f16f16f16_8x8x1_arm_64)
 
-static inline void iree_uk_mmt4d_tile_i8i8i32_1x8x1_to_8x8x1_arm_64(
+static inline void iree_uk_mmt4d_tile_s8s8s32_1x8x1_to_8x8x1_arm_64(
     void* IREE_UK_RESTRICT out_tile, const void* IREE_UK_RESTRICT lhs_panel,
     const void* IREE_UK_RESTRICT rhs_panel,
     const iree_uk_mmt4d_params_t* params, int M0) {
@@ -262,8 +262,8 @@ static inline void iree_uk_mmt4d_tile_i8i8i32_1x8x1_to_8x8x1_arm_64(
 }
 
 IREE_UK_MMT4D_TILE_FUNC_IMPL_FOR_M0_1_2_4_8(
-    iree_uk_mmt4d_tile_i8i8i32_1x8x1_to_8x8x1_arm_64,
-    iree_uk_mmt4d_tile_i8i8i32_1x8x1_arm_64,
-    iree_uk_mmt4d_tile_i8i8i32_2x8x1_arm_64,
-    iree_uk_mmt4d_tile_i8i8i32_4x8x1_arm_64,
-    iree_uk_mmt4d_tile_i8i8i32_8x8x1_arm_64)
+    iree_uk_mmt4d_tile_s8s8s32_1x8x1_to_8x8x1_arm_64,
+    iree_uk_mmt4d_tile_s8s8s32_1x8x1_arm_64,
+    iree_uk_mmt4d_tile_s8s8s32_2x8x1_arm_64,
+    iree_uk_mmt4d_tile_s8s8s32_4x8x1_arm_64,
+    iree_uk_mmt4d_tile_s8s8s32_8x8x1_arm_64)

runtime/src/iree/builtins/ukernel/arch/arm_64/mmt4d_arm_64_dotprod.c

@@ -7,7 +7,7 @@
 #include "iree/builtins/ukernel/arch/arm_64/common_arm_64.h"
 #include "iree/builtins/ukernel/arch/arm_64/mmt4d_arm_64_internal.h"
 
-static inline void iree_uk_mmt4d_tile_i8i8i32_1x8x4_to_8x8x4_arm_64_dotprod(
+static inline void iree_uk_mmt4d_tile_s8s8s32_1x8x4_to_8x8x4_arm_64_dotprod(
     void* IREE_UK_RESTRICT out_tile, const void* IREE_UK_RESTRICT lhs_panel,
     const void* IREE_UK_RESTRICT rhs_panel,
     const iree_uk_mmt4d_params_t* params, int M0) {
@@ -71,8 +71,8 @@ static inline void iree_uk_mmt4d_tile_i8i8i32_1x8x4_to_8x8x4_arm_64_dotprod(
 }
 
 IREE_UK_MMT4D_TILE_FUNC_IMPL_FOR_M0_1_2_4_8(
-    iree_uk_mmt4d_tile_i8i8i32_1x8x4_to_8x8x4_arm_64_dotprod,
-    iree_uk_mmt4d_tile_i8i8i32_1x8x4_arm_64_dotprod,
-    iree_uk_mmt4d_tile_i8i8i32_2x8x4_arm_64_dotprod,
-    iree_uk_mmt4d_tile_i8i8i32_4x8x4_arm_64_dotprod,
-    iree_uk_mmt4d_tile_i8i8i32_8x8x4_arm_64_dotprod)
+    iree_uk_mmt4d_tile_s8s8s32_1x8x4_to_8x8x4_arm_64_dotprod,
+    iree_uk_mmt4d_tile_s8s8s32_1x8x4_arm_64_dotprod,
+    iree_uk_mmt4d_tile_s8s8s32_2x8x4_arm_64_dotprod,
+    iree_uk_mmt4d_tile_s8s8s32_4x8x4_arm_64_dotprod,
+    iree_uk_mmt4d_tile_s8s8s32_8x8x4_arm_64_dotprod)

runtime/src/iree/builtins/ukernel/arch/arm_64/mmt4d_arm_64_entry_point.c

@@ -75,32 +75,32 @@ static iree_uk_mmt4d_tile_func_t
 iree_uk_mmt4d_select_tile_func_arm_64_f16f16f16_M0x8x1(
     const iree_uk_mmt4d_params_t* params) {
 #ifdef IREE_UK_BUILD_ARM_64_FP16
-    if (iree_uk_cpu_supports_fp16(params->cpu_data)) {
-      switch (params->M0) {
-        case 1:
-          return iree_uk_mmt4d_tile_f16f16f16_1x8x1_arm_64_fp16;
-        case 2:
-          return iree_uk_mmt4d_tile_f16f16f16_2x8x1_arm_64_fp16;
-        case 4:
-          return iree_uk_mmt4d_tile_f16f16f16_4x8x1_arm_64_fp16;
-        case 8:
-          return iree_uk_mmt4d_tile_f16f16f16_8x8x1_arm_64_fp16;
-      }
+  if (iree_uk_cpu_supports_fp16(params->cpu_data)) {
+    switch (params->M0) {
+      case 1:
+        return iree_uk_mmt4d_tile_f16f16f16_1x8x1_arm_64_fp16;
+      case 2:
+        return iree_uk_mmt4d_tile_f16f16f16_2x8x1_arm_64_fp16;
+      case 4:
+        return iree_uk_mmt4d_tile_f16f16f16_4x8x1_arm_64_fp16;
+      case 8:
+        return iree_uk_mmt4d_tile_f16f16f16_8x8x1_arm_64_fp16;
     }
+  }
 #endif
-    if (params->flags & IREE_UK_FLAG_MMT4D_SKIP_INTERMEDIATE_ROUNDINGS) {
-      switch (params->M0) {
-        case 1:
-          return iree_uk_mmt4d_tile_f16f16f16_1x8x1_arm_64;
-        case 2:
-          return iree_uk_mmt4d_tile_f16f16f16_2x8x1_arm_64;
-        case 4:
-          return iree_uk_mmt4d_tile_f16f16f16_4x8x1_arm_64;
-        case 8:
-          return iree_uk_mmt4d_tile_f16f16f16_8x8x1_arm_64;
-      }
+  if (params->flags & IREE_UK_FLAG_MMT4D_SKIP_INTERMEDIATE_ROUNDINGS) {
+    switch (params->M0) {
+      case 1:
+        return iree_uk_mmt4d_tile_f16f16f16_1x8x1_arm_64;
+      case 2:
+        return iree_uk_mmt4d_tile_f16f16f16_2x8x1_arm_64;
+      case 4:
+        return iree_uk_mmt4d_tile_f16f16f16_4x8x1_arm_64;
+      case 8:
+        return iree_uk_mmt4d_tile_f16f16f16_8x8x1_arm_64;
     }
-    return 0;
+  }
+  return 0;
 }
 
 static iree_uk_mmt4d_tile_func_t
@@ -116,20 +116,20 @@ static iree_uk_mmt4d_tile_func_t
 iree_uk_mmt4d_select_tile_func_arm_64_bf16bf16f32_M0x8x4(
     const iree_uk_mmt4d_params_t* params) {
 #ifdef IREE_UK_BUILD_ARM_64_BF16
-    if (iree_uk_cpu_supports_bf16(params->cpu_data)) {
-      switch (params->M0) {
-        case 1:
-          return iree_uk_mmt4d_tile_bf16bf16f32_1x8x4_arm_64_bf16;
-        case 2:
-          return iree_uk_mmt4d_tile_bf16bf16f32_2x8x4_arm_64_bf16;
-        case 4:
-          return iree_uk_mmt4d_tile_bf16bf16f32_4x8x4_arm_64_bf16;
-        case 8:
-          return iree_uk_mmt4d_tile_bf16bf16f32_8x8x4_arm_64_bf16;
-      }
+  if (iree_uk_cpu_supports_bf16(params->cpu_data)) {
+    switch (params->M0) {
+      case 1:
+        return iree_uk_mmt4d_tile_bf16bf16f32_1x8x4_arm_64_bf16;
+      case 2:
+        return iree_uk_mmt4d_tile_bf16bf16f32_2x8x4_arm_64_bf16;
+      case 4:
+        return iree_uk_mmt4d_tile_bf16bf16f32_4x8x4_arm_64_bf16;
+      case 8:
+        return iree_uk_mmt4d_tile_bf16bf16f32_8x8x4_arm_64_bf16;
     }
+  }
 #endif
-    return 0;
+  return 0;
 }
 
 static iree_uk_mmt4d_tile_func_t
@@ -152,13 +152,13 @@ iree_uk_mmt4d_select_tile_func_arm_64_i8i8i32_M0x8x1(
     const iree_uk_mmt4d_params_t* params) {
   switch (params->M0) {
     case 1:
-      return iree_uk_mmt4d_tile_i8i8i32_1x8x1_arm_64;
+      return iree_uk_mmt4d_tile_s8s8s32_1x8x1_arm_64;
     case 2:
-      return iree_uk_mmt4d_tile_i8i8i32_2x8x1_arm_64;
+      return iree_uk_mmt4d_tile_s8s8s32_2x8x1_arm_64;
     case 4:
-      return iree_uk_mmt4d_tile_i8i8i32_4x8x1_arm_64;
+      return iree_uk_mmt4d_tile_s8s8s32_4x8x1_arm_64;
     case 8:
-      return iree_uk_mmt4d_tile_i8i8i32_8x8x1_arm_64;
+      return iree_uk_mmt4d_tile_s8s8s32_8x8x1_arm_64;
   }
   return 0;
 }
@@ -170,13 +170,13 @@ iree_uk_mmt4d_select_tile_func_arm_64_i8i8i32_M0x8x4(
   if (iree_uk_cpu_supports_dotprod(params->cpu_data)) {
     switch (params->M0) {
       case 1:
-        return iree_uk_mmt4d_tile_i8i8i32_1x8x4_arm_64_dotprod;
+        return iree_uk_mmt4d_tile_s8s8s32_1x8x4_arm_64_dotprod;
       case 2:
-        return iree_uk_mmt4d_tile_i8i8i32_2x8x4_arm_64_dotprod;
+        return iree_uk_mmt4d_tile_s8s8s32_2x8x4_arm_64_dotprod;
       case 4:
-        return iree_uk_mmt4d_tile_i8i8i32_4x8x4_arm_64_dotprod;
+        return iree_uk_mmt4d_tile_s8s8s32_4x8x4_arm_64_dotprod;
       case 8:
-        return iree_uk_mmt4d_tile_i8i8i32_8x8x4_arm_64_dotprod;
+        return iree_uk_mmt4d_tile_s8s8s32_8x8x4_arm_64_dotprod;
     }
   }
 #endif
@@ -190,13 +190,13 @@ iree_uk_mmt4d_select_tile_func_arm_64_i8i8i32_M0x8x8(
   if (iree_uk_cpu_supports_i8mm(params->cpu_data)) {
     switch (params->M0) {
       case 1:
-        return iree_uk_mmt4d_tile_i8i8i32_1x8x8_arm_64_i8mm;
+        return iree_uk_mmt4d_tile_s8s8s32_1x8x8_arm_64_i8mm;
       case 2:
-        return iree_uk_mmt4d_tile_i8i8i32_2x8x8_arm_64_i8mm;
+        return iree_uk_mmt4d_tile_s8s8s32_2x8x8_arm_64_i8mm;
       case 4:
-        return iree_uk_mmt4d_tile_i8i8i32_4x8x8_arm_64_i8mm;
+        return iree_uk_mmt4d_tile_s8s8s32_4x8x8_arm_64_i8mm;
       case 8:
-        return iree_uk_mmt4d_tile_i8i8i32_8x8x8_arm_64_i8mm;
+        return iree_uk_mmt4d_tile_s8s8s32_8x8x8_arm_64_i8mm;
     }
   }
 #endif
@@ -230,7 +230,7 @@ iree_uk_mmt4d_tile_func_t iree_uk_mmt4d_select_tile_func_arch(
       return iree_uk_mmt4d_select_tile_func_arm_64_bf16bf16f32(params);
     case iree_uk_mmt4d_type_bf16bf16bf16:
       return iree_uk_mmt4d_select_tile_func_arm_64_bf16bf16bf16(params);
-    case iree_uk_mmt4d_type_i8i8i32:
+    case iree_uk_mmt4d_type_s8s8s32:
       return iree_uk_mmt4d_select_tile_func_arm_64_i8i8i32(params);
     default:
       IREE_UK_ASSUME_UNREACHABLE;

runtime/src/iree/builtins/ukernel/arch/arm_64/mmt4d_arm_64_i8mm.c

@@ -27,7 +27,7 @@ static inline int32x4_t iree_uk_neon_uzp2_s32_as_s64(int32x4_t a, int32x4_t b) {
       vuzp2q_s64(vreinterpretq_s64_s32(a), vreinterpretq_s64_s32(b)));
 }
 
-void iree_uk_mmt4d_tile_i8i8i32_1x8x8_to_8x8x8_arm_64_i8mm(
+void iree_uk_mmt4d_tile_s8s8s32_1x8x8_to_8x8x8_arm_64_i8mm(
     void* IREE_UK_RESTRICT out_tile, const void* IREE_UK_RESTRICT lhs_panel,
     const void* IREE_UK_RESTRICT rhs_panel,
     const iree_uk_mmt4d_params_t* params, int M0) {
@@ -99,8 +99,8 @@ void iree_uk_mmt4d_tile_i8i8i32_1x8x8_to_8x8x8_arm_64_i8mm(
 }
 
 IREE_UK_MMT4D_TILE_FUNC_IMPL_FOR_M0_1_2_4_8(
-    iree_uk_mmt4d_tile_i8i8i32_1x8x8_to_8x8x8_arm_64_i8mm,
-    iree_uk_mmt4d_tile_i8i8i32_1x8x8_arm_64_i8mm,
-    iree_uk_mmt4d_tile_i8i8i32_2x8x8_arm_64_i8mm,
-    iree_uk_mmt4d_tile_i8i8i32_4x8x8_arm_64_i8mm,
-    iree_uk_mmt4d_tile_i8i8i32_8x8x8_arm_64_i8mm)
+    iree_uk_mmt4d_tile_s8s8s32_1x8x8_to_8x8x8_arm_64_i8mm,
+    iree_uk_mmt4d_tile_s8s8s32_1x8x8_arm_64_i8mm,
+    iree_uk_mmt4d_tile_s8s8s32_2x8x8_arm_64_i8mm,
+    iree_uk_mmt4d_tile_s8s8s32_4x8x8_arm_64_i8mm,
+    iree_uk_mmt4d_tile_s8s8s32_8x8x8_arm_64_i8mm)

runtime/src/iree/builtins/ukernel/arch/arm_64/mmt4d_arm_64_internal.h

@@ -33,17 +33,17 @@ IREE_UK_MMT4D_TILE_FUNC_DECL(iree_uk_mmt4d_tile_bf16bf16f32_1x8x4_arm_64_bf16)
 IREE_UK_MMT4D_TILE_FUNC_DECL(iree_uk_mmt4d_tile_bf16bf16f32_2x8x4_arm_64_bf16)
 IREE_UK_MMT4D_TILE_FUNC_DECL(iree_uk_mmt4d_tile_bf16bf16f32_4x8x4_arm_64_bf16)
 IREE_UK_MMT4D_TILE_FUNC_DECL(iree_uk_mmt4d_tile_bf16bf16f32_8x8x4_arm_64_bf16)
-IREE_UK_MMT4D_TILE_FUNC_DECL(iree_uk_mmt4d_tile_i8i8i32_1x8x1_arm_64)
-IREE_UK_MMT4D_TILE_FUNC_DECL(iree_uk_mmt4d_tile_i8i8i32_2x8x1_arm_64)
-IREE_UK_MMT4D_TILE_FUNC_DECL(iree_uk_mmt4d_tile_i8i8i32_4x8x1_arm_64)
-IREE_UK_MMT4D_TILE_FUNC_DECL(iree_uk_mmt4d_tile_i8i8i32_8x8x1_arm_64)
-IREE_UK_MMT4D_TILE_FUNC_DECL(iree_uk_mmt4d_tile_i8i8i32_1x8x4_arm_64_dotprod)
-IREE_UK_MMT4D_TILE_FUNC_DECL(iree_uk_mmt4d_tile_i8i8i32_2x8x4_arm_64_dotprod)
-IREE_UK_MMT4D_TILE_FUNC_DECL(iree_uk_mmt4d_tile_i8i8i32_4x8x4_arm_64_dotprod)
-IREE_UK_MMT4D_TILE_FUNC_DECL(iree_uk_mmt4d_tile_i8i8i32_8x8x4_arm_64_dotprod)
-IREE_UK_MMT4D_TILE_FUNC_DECL(iree_uk_mmt4d_tile_i8i8i32_1x8x8_arm_64_i8mm)
-IREE_UK_MMT4D_TILE_FUNC_DECL(iree_uk_mmt4d_tile_i8i8i32_2x8x8_arm_64_i8mm)
-IREE_UK_MMT4D_TILE_FUNC_DECL(iree_uk_mmt4d_tile_i8i8i32_4x8x8_arm_64_i8mm)
-IREE_UK_MMT4D_TILE_FUNC_DECL(iree_uk_mmt4d_tile_i8i8i32_8x8x8_arm_64_i8mm)
+IREE_UK_MMT4D_TILE_FUNC_DECL(iree_uk_mmt4d_tile_s8s8s32_1x8x1_arm_64)
+IREE_UK_MMT4D_TILE_FUNC_DECL(iree_uk_mmt4d_tile_s8s8s32_2x8x1_arm_64)
+IREE_UK_MMT4D_TILE_FUNC_DECL(iree_uk_mmt4d_tile_s8s8s32_4x8x1_arm_64)
+IREE_UK_MMT4D_TILE_FUNC_DECL(iree_uk_mmt4d_tile_s8s8s32_8x8x1_arm_64)
+IREE_UK_MMT4D_TILE_FUNC_DECL(iree_uk_mmt4d_tile_s8s8s32_1x8x4_arm_64_dotprod)
+IREE_UK_MMT4D_TILE_FUNC_DECL(iree_uk_mmt4d_tile_s8s8s32_2x8x4_arm_64_dotprod)
+IREE_UK_MMT4D_TILE_FUNC_DECL(iree_uk_mmt4d_tile_s8s8s32_4x8x4_arm_64_dotprod)
+IREE_UK_MMT4D_TILE_FUNC_DECL(iree_uk_mmt4d_tile_s8s8s32_8x8x4_arm_64_dotprod)
+IREE_UK_MMT4D_TILE_FUNC_DECL(iree_uk_mmt4d_tile_s8s8s32_1x8x8_arm_64_i8mm)
+IREE_UK_MMT4D_TILE_FUNC_DECL(iree_uk_mmt4d_tile_s8s8s32_2x8x8_arm_64_i8mm)
+IREE_UK_MMT4D_TILE_FUNC_DECL(iree_uk_mmt4d_tile_s8s8s32_4x8x8_arm_64_i8mm)
+IREE_UK_MMT4D_TILE_FUNC_DECL(iree_uk_mmt4d_tile_s8s8s32_8x8x8_arm_64_i8mm)
 
 #endif  // IREE_BUILTINS_UKERNEL_ARCH_ARM_64_MMT4D_ARM_64_INTERNAL_H_

runtime/src/iree/builtins/ukernel/arch/x86_64/mmt4d_x86_64_avx2_fma.c

@@ -127,7 +127,7 @@ IREE_UK_MMT4D_TILE_FUNC_IMPL_FOR_M0_1_2_4_8(
     iree_uk_mmt4d_tile_f16f16f16_4x8x1_x86_64_avx2_fma,
     iree_uk_mmt4d_tile_f16f16f16_8x8x1_x86_64_avx2_fma)
 
-static inline void iree_uk_mmt4d_tile_i8i8i32_1x8x2_to_8x8x2_x86_64_avx2_fma(
+static inline void iree_uk_mmt4d_tile_s8s8s32_1x8x2_to_8x8x2_x86_64_avx2_fma(
     void* IREE_UK_RESTRICT out_tile, const void* IREE_UK_RESTRICT lhs_panel,
     const void* IREE_UK_RESTRICT rhs_panel,
     const iree_uk_mmt4d_params_t* params, int M0) {
@@ -216,8 +216,8 @@ static inline void iree_uk_mmt4d_tile_i8i8i32_1x8x2_to_8x8x2_x86_64_avx2_fma(
 }
 
 IREE_UK_MMT4D_TILE_FUNC_IMPL_FOR_M0_1_2_4_8(
-    iree_uk_mmt4d_tile_i8i8i32_1x8x2_to_8x8x2_x86_64_avx2_fma,
-    iree_uk_mmt4d_tile_i8i8i32_1x8x2_x86_64_avx2_fma,
-    iree_uk_mmt4d_tile_i8i8i32_2x8x2_x86_64_avx2_fma,
-    iree_uk_mmt4d_tile_i8i8i32_4x8x2_x86_64_avx2_fma,
-    iree_uk_mmt4d_tile_i8i8i32_8x8x2_x86_64_avx2_fma)
+    iree_uk_mmt4d_tile_s8s8s32_1x8x2_to_8x8x2_x86_64_avx2_fma,
+    iree_uk_mmt4d_tile_s8s8s32_1x8x2_x86_64_avx2_fma,
+    iree_uk_mmt4d_tile_s8s8s32_2x8x2_x86_64_avx2_fma,
+    iree_uk_mmt4d_tile_s8s8s32_4x8x2_x86_64_avx2_fma,
+    iree_uk_mmt4d_tile_s8s8s32_8x8x2_x86_64_avx2_fma)

runtime/src/iree/builtins/ukernel/arch/x86_64/mmt4d_x86_64_avx512_base.c

@@ -179,7 +179,7 @@ IREE_UK_MMT4D_TILE_FUNC_IMPL_FOR_M0_1_2_4_8_16(
     iree_uk_mmt4d_tile_f16f16f16_16x16x1_x86_64_avx512_base)
 
 static inline void
-iree_uk_mmt4d_tile_i8i8i32_1x16x2_to_16x16x2_x86_64_avx512_base(
+iree_uk_mmt4d_tile_s8s8s32_1x16x2_to_16x16x2_x86_64_avx512_base(
     void* IREE_UK_RESTRICT out_tile, const void* IREE_UK_RESTRICT lhs_panel,
     const void* IREE_UK_RESTRICT rhs_panel,
     const iree_uk_mmt4d_params_t* params, int M0) {
@@ -300,9 +300,9 @@ iree_uk_mmt4d_tile_i8i8i32_1x16x2_to_16x16x2_x86_64_avx512_base(
 }
 
 IREE_UK_MMT4D_TILE_FUNC_IMPL_FOR_M0_1_2_4_8_16(
-    iree_uk_mmt4d_tile_i8i8i32_1x16x2_to_16x16x2_x86_64_avx512_base,
-    iree_uk_mmt4d_tile_i8i8i32_1x16x2_x86_64_avx512_base,
-    iree_uk_mmt4d_tile_i8i8i32_2x16x2_x86_64_avx512_base,
-    iree_uk_mmt4d_tile_i8i8i32_4x16x2_x86_64_avx512_base,
-    iree_uk_mmt4d_tile_i8i8i32_8x16x2_x86_64_avx512_base,
-    iree_uk_mmt4d_tile_i8i8i32_16x16x2_x86_64_avx512_base)
+    iree_uk_mmt4d_tile_s8s8s32_1x16x2_to_16x16x2_x86_64_avx512_base,
+    iree_uk_mmt4d_tile_s8s8s32_1x16x2_x86_64_avx512_base,
+    iree_uk_mmt4d_tile_s8s8s32_2x16x2_x86_64_avx512_base,
+    iree_uk_mmt4d_tile_s8s8s32_4x16x2_x86_64_avx512_base,
+    iree_uk_mmt4d_tile_s8s8s32_8x16x2_x86_64_avx512_base,
+    iree_uk_mmt4d_tile_s8s8s32_16x16x2_x86_64_avx512_base)

runtime/src/iree/builtins/ukernel/arch/x86_64/mmt4d_x86_64_avx512_vnni.c

@@ -8,7 +8,7 @@
 #include "iree/builtins/ukernel/arch/x86_64/mmt4d_x86_64_internal.h"
 
 static inline void
-iree_uk_mmt4d_tile_i8i8i32_1x16x2_to_16x16x2_x86_64_avx512_vnni(
+iree_uk_mmt4d_tile_s8s8s32_1x16x2_to_16x16x2_x86_64_avx512_vnni(
     void* IREE_UK_RESTRICT out_tile, const void* IREE_UK_RESTRICT lhs_panel,
     const void* IREE_UK_RESTRICT rhs_panel,
     const iree_uk_mmt4d_params_t* params, int M0) {
@@ -129,9 +129,9 @@ iree_uk_mmt4d_tile_i8i8i32_1x16x2_to_16x16x2_x86_64_avx512_vnni(
 }
 
 IREE_UK_MMT4D_TILE_FUNC_IMPL_FOR_M0_1_2_4_8_16(
-    iree_uk_mmt4d_tile_i8i8i32_1x16x2_to_16x16x2_x86_64_avx512_vnni,
-    iree_uk_mmt4d_tile_i8i8i32_1x16x2_x86_64_avx512_vnni,
-    iree_uk_mmt4d_tile_i8i8i32_2x16x2_x86_64_avx512_vnni,
-    iree_uk_mmt4d_tile_i8i8i32_4x16x2_x86_64_avx512_vnni,
-    iree_uk_mmt4d_tile_i8i8i32_8x16x2_x86_64_avx512_vnni,
-    iree_uk_mmt4d_tile_i8i8i32_16x16x2_x86_64_avx512_vnni)
+    iree_uk_mmt4d_tile_s8s8s32_1x16x2_to_16x16x2_x86_64_avx512_vnni,
+    iree_uk_mmt4d_tile_s8s8s32_1x16x2_x86_64_avx512_vnni,
+    iree_uk_mmt4d_tile_s8s8s32_2x16x2_x86_64_avx512_vnni,
+    iree_uk_mmt4d_tile_s8s8s32_4x16x2_x86_64_avx512_vnni,
+    iree_uk_mmt4d_tile_s8s8s32_8x16x2_x86_64_avx512_vnni,
+    iree_uk_mmt4d_tile_s8s8s32_16x16x2_x86_64_avx512_vnni)