[Fix] [μKernels]: Reverting the batch=256 for ARM vector bench (#1085)

This `PR` updates the `batch` back to `256` for `ARM` micro-kernel
lowering. Seems like the `ARM` micro-kernel lowering is broken, lowers
to `scalar` code, and triggers error for `512`. Raised issue:
#1086.

Author: arun-thmn

benchmarks/config/base/base.json

@@ -52,7 +52,7 @@
     },
     "gemm_fp32_mlir_vector_sve": {
       "type": "IR-GEN",
-      "benchmark": [ "mlir-gen", "--kernel=const --float-type=f32 --batch=512 --layers=1024,1024,1024,1024 --tiles=32,32,32" ],
+      "benchmark": [ "mlir-gen", "--kernel=const --float-type=f32 --batch=256 --layers=1024,1024,1024,1024 --tiles=32,32,32" ],
       "environment": {},
       "flags": [ "-n", "100",  "-run-args='--vector-to-kernels --registerBlocking=4,32,1'" ],
       "extensions": ["asimd"]
@@ -129,7 +129,7 @@
     },
     "mlp_fp32_mlir_vector_sve": {
       "type": "IR-GEN",
-      "benchmark": [ "mlir-gen", "--kernel=const --bias --relu --float-type=f32 --batch=512 --layers=1024,1024,1024,1024 --tiles=32,32,32" ],
+      "benchmark": [ "mlir-gen", "--kernel=const --bias --relu --float-type=f32 --batch=256 --layers=1024,1024,1024,1024 --tiles=32,32,32" ],
       "environment": {},
       "flags": [ "-n", "100",  "-run-args='--def-parallel --vector-to-kernels --registerBlocking=4,32,1'" ],
       "extensions": ["asimd"]

benchmarks/config/omp/mlir-fp32-vector-to-kernel.json

@@ -189,28 +189,28 @@
   "gemm_fp32_mlir_vector_kernel_32_sve": {
     "fp32_3x1024_omp_2_mlir": {
       "type": "IR-GEN",
-      "benchmark": [ "mlir-gen", "--kernel=const --float-type=f32 --batch=512 --layers=1024,1024,1024,1024 --tiles=32,32,32" ],
+      "benchmark": [ "mlir-gen", "--kernel=const --float-type=f32 --batch=256 --layers=1024,1024,1024,1024 --tiles=32,32,32" ],
       "environment": { "OMP_NUM_THREADS": "2", "KMP_AFFINITY": "granularity=fine,verbose,compact,1,0" },
       "flags": [ "-n", "100", "-run-args='--def-parallel --parallel-task-grid=8,16 --vector-to-kernels --registerBlocking=4,32,1   -aarch64-sve-vector-bits-min=256 -aarch64-sve-vector-bits-max=256'" ],
       "extensions": [ "asimd" ]
     },
     "fp32_3x1024_omp_4_mlir": {
       "type": "IR-GEN",
-      "benchmark": [ "mlir-gen", "--kernel=const --float-type=f32 --batch=512 --layers=1024,1024,1024,1024 --tiles=32,32,32" ],
+      "benchmark": [ "mlir-gen", "--kernel=const --float-type=f32 --batch=256 --layers=1024,1024,1024,1024 --tiles=32,32,32" ],
       "environment": { "OMP_NUM_THREADS": "4", "KMP_AFFINITY": "granularity=fine,verbose,compact,1,0" },
       "flags": [ "-n", "100", "-run-args='--def-parallel --parallel-task-grid=8,8 --vector-to-kernels --registerBlocking=4,32,1   -aarch64-sve-vector-bits-min=256 -aarch64-sve-vector-bits-max=256'" ],
       "extensions": [ "asimd" ]
     },
     "fp32_3x1024_omp_8_mlir": {
       "type": "IR-GEN",
-      "benchmark": [ "mlir-gen", "--kernel=const --float-type=f32 --batch=512 --layers=1024,1024,1024,1024 --tiles=32,32,32" ],
+      "benchmark": [ "mlir-gen", "--kernel=const --float-type=f32 --batch=256 --layers=1024,1024,1024,1024 --tiles=32,32,32" ],
       "environment": { "OMP_NUM_THREADS": "8", "KMP_AFFINITY": "granularity=fine,verbose,compact,1,0" },
       "flags": [ "-n", "100", "-run-args='--def-parallel --parallel-task-grid=4,8 --vector-to-kernels --registerBlocking=4,32,1   -aarch64-sve-vector-bits-min=256 -aarch64-sve-vector-bits-max=256'" ],
       "extensions": [ "asimd" ]
     },
     "fp32_3x1024_omp_16_mlir": {
       "type": "IR-GEN",
-      "benchmark": [ "mlir-gen", "--kernel=const --float-type=f32 --batch=512 --layers=1024,1024,1024,1024 --tiles=32,32,32" ],
+      "benchmark": [ "mlir-gen", "--kernel=const --float-type=f32 --batch=256 --layers=1024,1024,1024,1024 --tiles=32,32,32" ],
       "environment": { "OMP_NUM_THREADS": "16", "KMP_AFFINITY": "granularity=fine,verbose,compact,1,0" },
       "flags": [ "-n", "100", "-run-args='--def-parallel --parallel-task-grid=2,8 --vector-to-kernels --registerBlocking=4,32,1   -aarch64-sve-vector-bits-min=256 -aarch64-sve-vector-bits-max=256'" ],
       "extensions": [ "asimd" ]
@@ -220,28 +220,28 @@
   "mlp_fp32_mlir_vector_kernel_32_sve": {
     "fp32_3x1024_omp_2_mlir": {
       "type": "IR-GEN",
-      "benchmark": [ "mlir-gen", "--kernel=const --bias --relu --float-type=f32 --batch=512 --layers=1024,1024,1024,1024 --tiles=32,32,32" ],
+      "benchmark": [ "mlir-gen", "--kernel=const --bias --relu --float-type=f32 --batch=256 --layers=1024,1024,1024,1024 --tiles=32,32,32" ],
       "environment": { "OMP_NUM_THREADS": "2", "KMP_AFFINITY": "granularity=fine,verbose,compact,1,0" },
       "flags": [ "-n", "100", "-run-args='--def-parallel --parallel-task-grid=8,16 --vector-to-kernels --registerBlocking=4,32,1   -aarch64-sve-vector-bits-min=256 -aarch64-sve-vector-bits-max=256'" ],
       "extensions": [ "asimd" ]
     },
     "fp32_3x1024_omp_4_mlir": {
       "type": "IR-GEN",
-      "benchmark": [ "mlir-gen", "--kernel=const --bias --relu --float-type=f32 --batch=512 --layers=1024,1024,1024,1024 --tiles=32,32,32" ],
+      "benchmark": [ "mlir-gen", "--kernel=const --bias --relu --float-type=f32 --batch=256 --layers=1024,1024,1024,1024 --tiles=32,32,32" ],
       "environment": { "OMP_NUM_THREADS": "4", "KMP_AFFINITY": "granularity=fine,verbose,compact,1,0" },
       "flags": [ "-n", "100", "-run-args='--def-parallel --parallel-task-grid=8,8 --vector-to-kernels --registerBlocking=4,32,1   -aarch64-sve-vector-bits-min=256 -aarch64-sve-vector-bits-max=256'" ],
       "extensions": [ "asimd" ]
     },
     "fp32_3x1024_omp_8_mlir": {
       "type": "IR-GEN",
-      "benchmark": [ "mlir-gen", "--kernel=const --bias --relu --float-type=f32 --batch=512 --layers=1024,1024,1024,1024 --tiles=32,32,32" ],
+      "benchmark": [ "mlir-gen", "--kernel=const --bias --relu --float-type=f32 --batch=256 --layers=1024,1024,1024,1024 --tiles=32,32,32" ],
       "environment": { "OMP_NUM_THREADS": "8", "KMP_AFFINITY": "granularity=fine,verbose,compact,1,0" },
       "flags": [ "-n", "100", "-run-args='--def-parallel --parallel-task-grid=4,8 --vector-to-kernels --registerBlocking=4,32,1   -aarch64-sve-vector-bits-min=256 -aarch64-sve-vector-bits-max=256'" ],
       "extensions": [ "asimd" ]
     },
     "fp32_3x1024_omp_16_mlir": {
       "type": "IR-GEN",
-      "benchmark": [ "mlir-gen", "--kernel=const --bias --relu --float-type=f32 --batch=512 --layers=1024,1024,1024,1024 --tiles=32,32,32" ],
+      "benchmark": [ "mlir-gen", "--kernel=const --bias --relu --float-type=f32 --batch=256 --layers=1024,1024,1024,1024 --tiles=32,32,32" ],
       "environment": { "OMP_NUM_THREADS": "16", "KMP_AFFINITY": "granularity=fine,verbose,compact,1,0" },
       "flags": [ "-n", "100", "-run-args='--def-parallel --parallel-task-grid=2,8 --vector-to-kernels --registerBlocking=4,32,1   -aarch64-sve-vector-bits-min=256 -aarch64-sve-vector-bits-max=256'" ],
       "extensions": [ "asimd" ]