diff --git a/.github/workflows/benchmark.yml b/.github/workflows/benchmark.yml
index 3b37abda35..b823c454c3 100644
--- a/.github/workflows/benchmark.yml
+++ b/.github/workflows/benchmark.yml
@@ -590,6 +590,8 @@ jobs:
       run: test/external/process_replay/reset.py
     - name: openpilot compile3 0.11.0 driving_vision
       run: BENCHMARK_LOG=openpilot_0_11_0_vision PYTHONPATH="." ASSERT_MIN_STEP_TIME=17 DEV=QCOM FLOAT16=1 IMAGE=1 NOLOCALS=1 taskset -c 4-7 python3 examples/openpilot/compile3.py https://github.com/commaai/openpilot/raw/v0.11.0/selfdrive/modeld/models/driving_vision.onnx
+    - name: IR3 openpilot compile3 0.11.0 driving_vision
+      run: BENCHMARK_LOG=ir3_openpilot_0_11_0_vision PYTHONPATH="." ASSERT_MIN_STEP_TIME=17 DEV=QCOM QCOM_IR3=1 FLOAT16=1 IMAGE=1 NOLOCALS=1 taskset -c 4-7 python3 examples/openpilot/compile3.py https://github.com/commaai/openpilot/raw/v0.11.0/selfdrive/modeld/models/driving_vision.onnx
     - name: openpilot compile3 0.11.0 driving_policy
       run: BENCHMARK_LOG=openpilot_0_11_0_policy PYTHONPATH="." ASSERT_MIN_STEP_TIME=3 DEV=QCOM FLOAT16=1 IMAGE=1 NOLOCALS=1 taskset -c 4-7 python3 examples/openpilot/compile3.py https://github.com/commaai/openpilot/raw/v0.11.0/selfdrive/modeld/models/driving_policy.onnx
     - name: openpilot compile3 0.11.0 dmonitoring
diff --git a/tinygrad/runtime/ops_qcom.py b/tinygrad/runtime/ops_qcom.py
index c97d63a9d9..154b69e9ba 100644
--- a/tinygrad/runtime/ops_qcom.py
+++ b/tinygrad/runtime/ops_qcom.py
@@ -202,7 +202,8 @@ class QCOMArgsState(HCQArgsState):
 
     ubos = [b for i,b in enumerate(bufs) for _,dt in prg.buf_dtypes[i] if not isinstance(dt, ImageDType)]
     uavs = [(dt,b) for i,b in enumerate(bufs) for _,dt in prg.buf_dtypes[i] if isinstance(dt, ImageDType)]
-    ibos, texs = uavs[:prg.ibo_cnt], uavs[prg.ibo_cnt:]
+    # NIR can reorder images to different texture slots
+    ibos, texs = uavs[:prg.ibo_cnt], [uavs[prg.ibo_cnt + (prg.tex_to_image[i] if prg.NIR else i)] for i in range(prg.tex_cnt)]
     for cnst_val,cnst_off,cnst_sz in prg.consts_info:
       to_mv(cast(int, self.buf.va_addr) + cnst_off, cnst_sz)[:] = cnst_val.to_bytes(cnst_sz, byteorder='little')
 
@@ -243,6 +244,7 @@ class QCOMProgram(HCQProgram):
       # see https://elixir.bootlin.com/mesa/mesa-25.3.0/source/src/freedreno/ir3/ir3_shader.h#L525
       # and https://elixir.bootlin.com/mesa/mesa-25.3.0/source/src/freedreno/ir3/ir3_compiler_nir.c#L5389
       self.samp_cnt, self.tex_cnt, self.ibo_cnt = (nt:=v.image_mapping.num_tex), nt, v.num_uavs - nt
+      self.tex_to_image = v.image_mapping.tex_to_image[:]
       # IR3 outputs a sampler for every texture (https://elixir.bootlin.com/mesa/mesa-25.3.0/source/src/freedreno/ir3/ir3_compiler_nir.c#L1714)
       self.samplers = [qreg.a6xx_tex_samp_0(wrap_s=(clamp_mode:=mesa.A6XX_TEX_CLAMP_TO_BORDER), wrap_t=clamp_mode, wrap_r=clamp_mode),
                        qreg.a6xx_tex_samp_1(unnorm_coords=True, cubemapseamlessfiltoff=True), 0, 0] * self.samp_cnt
diff --git a/tinygrad/runtime/support/c.py b/tinygrad/runtime/support/c.py
index dddaaa90b5..a4f092d8f8 100644
--- a/tinygrad/runtime/support/c.py
+++ b/tinygrad/runtime/support/c.py
@@ -36,6 +36,8 @@ if TYPE_CHECKING:
     @overload
     def __getitem__(self: Array[_SimpleCData[V], Any], key: int) -> V: ...
     @overload
+    def __getitem__(self: Array[T, Any], key: slice) -> list[T]: ...
+    @overload
     def __getitem__(self: Array[T, Any], key: int) -> T: ...
     def __getitem__(self, key) -> Any: ...
     @overload