[hip][rocm] Fix hipGetDeviceProperties usage after ROCm 6.0 (#16643)
ROCm 6.0 changes the `hipDeviceProp_t` struct, so we would need the new
`hipGetDevicePropertiesR0600()` API to query it--the existing
`hipGetDeviceProperties()`l API won't work for our case with dynamic
libraries.
Related to https://github.com/ROCm/ROCm/issues/2728
diff --git a/experimental/hip/dynamic_symbol_tables.h b/experimental/hip/dynamic_symbol_tables.h
index bbcb1d4..bc6d0b9 100644
--- a/experimental/hip/dynamic_symbol_tables.h
+++ b/experimental/hip/dynamic_symbol_tables.h
@@ -34,7 +34,8 @@
IREE_HAL_HIP_REQUIRED_PFN_DECL(hipFuncSetAttribute, const void *,
hipFuncAttribute, int)
IREE_HAL_HIP_REQUIRED_PFN_DECL(hipGetDeviceCount, int *)
-IREE_HAL_HIP_REQUIRED_PFN_DECL(hipGetDeviceProperties, hipDeviceProp_t *, int)
+IREE_HAL_HIP_OPTIONAL_PFN_DECL(hipGetDevicePropertiesR0600, hipDeviceProp_t *,
+ int)
// hipGetErrorName(hipError_t) and hipGetErrorString(hipError_t) return
// const char* instead of hipError_t so it uses a different macro.
IREE_HAL_HIP_REQUIRED_PFN_STR_DECL(hipGetErrorName, hipError_t)
diff --git a/experimental/hip/hip_driver.c b/experimental/hip/hip_driver.c
index 3026c14..5ade4ed 100644
--- a/experimental/hip/hip_driver.c
+++ b/experimental/hip/hip_driver.c
@@ -232,12 +232,19 @@
iree_hal_driver_t* base_driver, iree_hal_device_id_t device_id,
iree_string_builder_t* builder) {
iree_hal_hip_driver_t* driver = iree_hal_hip_driver_cast(base_driver);
+ if (!driver->hip_symbols.hipGetDevicePropertiesR0600) {
+ // ROCm 6.0 release changes the hipDeviceProp_t struct and would need to use
+ // the matching hipGetDevicePropertiesR0600() API to query it. This symbol
+ // is not available in earlier versions.
+ return iree_ok_status();
+ }
+
hipDevice_t device = IREE_DEVICE_ID_TO_HIPDEVICE(device_id);
hipDeviceProp_t prop;
IREE_HIP_RETURN_IF_ERROR(&driver->hip_symbols,
- hipGetDeviceProperties(&prop, device),
- "hipGetDeviceProperties");
+ hipGetDevicePropertiesR0600(&prop, device),
+ "hipGetDevicePropertiesR0600");
// GPU capabilities and architecture.
IREE_RETURN_IF_ERROR(iree_string_builder_append_format(
diff --git a/experimental/rocm/dynamic_symbol_tables.h b/experimental/rocm/dynamic_symbol_tables.h
index 2cdab9c..323a65d 100644
--- a/experimental/rocm/dynamic_symbol_tables.h
+++ b/experimental/rocm/dynamic_symbol_tables.h
@@ -8,7 +8,7 @@
RC_PFN_DECL(hipCtxDestroy, hipCtx_t)
RC_PFN_DECL(hipDeviceGet, hipDevice_t *, int) // No direct, need to modify
RC_PFN_DECL(hipGetDeviceCount, int *)
-RC_PFN_DECL(hipGetDeviceProperties, hipDeviceProp_t *, int)
+RC_PFN_DECL(hipGetDevicePropertiesR0600, hipDeviceProp_t *, int)
RC_PFN_DECL(hipDeviceGetName, char *, int,
hipDevice_t) // No direct, need to modify
RC_PFN_STR_DECL(
diff --git a/experimental/rocm/rocm_driver.c b/experimental/rocm/rocm_driver.c
index bcec506..4086eac 100644
--- a/experimental/rocm/rocm_driver.c
+++ b/experimental/rocm/rocm_driver.c
@@ -185,8 +185,9 @@
hipDevice_t device = IREE_DEVICE_ID_TO_HIPDEVICE(device_id);
hipDeviceProp_t prop;
- ROCM_RETURN_IF_ERROR(&driver->syms, hipGetDeviceProperties(&prop, device),
- "hipGetDeviceProperties");
+ ROCM_RETURN_IF_ERROR(&driver->syms,
+ hipGetDevicePropertiesR0600(&prop, device),
+ "hipGetDevicePropertiesR0600");
// GPU capabilities and architecture.
IREE_RETURN_IF_ERROR(iree_string_builder_append_format(
