summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
-rw-r--r--backend/src/backend/gen_program.cpp5
-rw-r--r--src/cl_device_data.h54
-rw-r--r--src/cl_device_id.c97
3 files changed, 153 insertions, 3 deletions
diff --git a/backend/src/backend/gen_program.cpp b/backend/src/backend/gen_program.cpp
index e06ed40c..274c99c7 100644
--- a/backend/src/backend/gen_program.cpp
+++ b/backend/src/backend/gen_program.cpp
@@ -209,6 +209,8 @@ namespace gbe {
ctx = GBE_NEW(BxtContext, unit, name, deviceID, relaxMath);
} else if (IS_KABYLAKE(deviceID)) {
ctx = GBE_NEW(KblContext, unit, name, deviceID, relaxMath);
+ } else if (IS_COFFEELAKE(deviceID)) {
+ ctx = GBE_NEW(KblContext, unit, name, deviceID, relaxMath);
} else if (IS_GEMINILAKE(deviceID)) {
ctx = GBE_NEW(GlkContext, unit, name, deviceID, relaxMath);
}
@@ -328,6 +330,7 @@ namespace gbe {
(IS_SKYLAKE(deviceID) && MATCH_SKL_HEADER(binary)) || \
(IS_BROXTON(deviceID) && MATCH_BXT_HEADER(binary)) || \
(IS_KABYLAKE(deviceID) && MATCH_KBL_HEADER(binary)) || \
+ (IS_COFFEELAKE(deviceID) && MATCH_KBL_HEADER(binary)) || \
(IS_GEMINILAKE(deviceID) && MATCH_GLK_HEADER(binary)) \
)
@@ -436,6 +439,8 @@ namespace gbe {
FILL_BXT_HEADER(*binary);
}else if(IS_KABYLAKE(prog->deviceID)){
FILL_KBL_HEADER(*binary);
+ }else if(IS_COFFEELAKE(prog->deviceID)){
+ FILL_KBL_HEADER(*binary);
}else if(IS_GEMINILAKE(prog->deviceID)){
FILL_GLK_HEADER(*binary);
}else {
diff --git a/src/cl_device_data.h b/src/cl_device_data.h
index 123b6192..79f7002e 100644
--- a/src/cl_device_data.h
+++ b/src/cl_device_data.h
@@ -372,7 +372,59 @@
(devid == PCI_CHIP_GLK_3x6 || \
devid == PCI_CHIP_GLK_2x6)
-#define IS_GEN9(devid) (IS_SKYLAKE(devid) || IS_BROXTON(devid) || IS_KABYLAKE(devid) || IS_GEMINILAKE(devid))
+#define PCI_CHIP_COFFEELAKE_S_GT1_1 0x3E90
+#define PCI_CHIP_COFFEELAKE_S_GT1_2 0x3E93
+#define PCI_CHIP_COFFEELAKE_S_GT1_3 0x3E99
+
+#define PCI_CHIP_COFFEELAKE_U_GT1_1 0x3EA1
+#define PCI_CHIP_COFFEELAKE_U_GT1_2 0x3EA4
+
+#define PCI_CHIP_COFFEELAKE_S_GT2_1 0x3E91
+#define PCI_CHIP_COFFEELAKE_S_GT2_2 0x3E92
+#define PCI_CHIP_COFFEELAKE_S_GT2_3 0x3E96
+#define PCI_CHIP_COFFEELAKE_S_GT2_4 0x3E9A
+
+#define PCI_CHIP_COFFEELAKE_H_GT2_1 0x3E94
+#define PCI_CHIP_COFFEELAKE_H_GT2_2 0x3E9B
+
+#define PCI_CHIP_COFFEELAKE_U_GT2_1 0x3EA0
+#define PCI_CHIP_COFFEELAKE_U_GT2_2 0x3EA3
+#define PCI_CHIP_COFFEELAKE_U_GT2_3 0x3EA9
+
+#define PCI_CHIP_COFFEELAKE_U_GT3_1 0x3EA2
+#define PCI_CHIP_COFFEELAKE_U_GT3_2 0x3EA5
+#define PCI_CHIP_COFFEELAKE_U_GT3_3 0x3EA6
+#define PCI_CHIP_COFFEELAKE_U_GT3_4 0x3EA7
+#define PCI_CHIP_COFFEELAKE_U_GT3_5 0x3EA8
+
+#define IS_CFL_GT1(devid) \
+ (devid == PCI_CHIP_COFFEELAKE_S_GT1_1 || \
+ devid == PCI_CHIP_COFFEELAKE_S_GT1_2 || \
+ devid == PCI_CHIP_COFFEELAKE_S_GT1_3 || \
+ devid == PCI_CHIP_COFFEELAKE_U_GT1_1 || \
+ devid == PCI_CHIP_COFFEELAKE_U_GT1_2)
+
+#define IS_CFL_GT2(devid) \
+ (devid == PCI_CHIP_COFFEELAKE_S_GT2_1 || \
+ devid == PCI_CHIP_COFFEELAKE_S_GT2_2 || \
+ devid == PCI_CHIP_COFFEELAKE_S_GT2_3 || \
+ devid == PCI_CHIP_COFFEELAKE_S_GT2_4 || \
+ devid == PCI_CHIP_COFFEELAKE_H_GT2_1 || \
+ devid == PCI_CHIP_COFFEELAKE_H_GT2_2 || \
+ devid == PCI_CHIP_COFFEELAKE_U_GT2_1 || \
+ devid == PCI_CHIP_COFFEELAKE_U_GT2_2 || \
+ devid == PCI_CHIP_COFFEELAKE_U_GT2_3)
+
+#define IS_CFL_GT3(devid) \
+ (devid == PCI_CHIP_COFFEELAKE_U_GT3_1 || \
+ devid == PCI_CHIP_COFFEELAKE_U_GT3_2 || \
+ devid == PCI_CHIP_COFFEELAKE_U_GT3_3 || \
+ devid == PCI_CHIP_COFFEELAKE_U_GT3_4 || \
+ devid == PCI_CHIP_COFFEELAKE_U_GT3_5)
+
+#define IS_COFFEELAKE(devid) (IS_CFL_GT1(devid) || IS_CFL_GT2(devid) || IS_CFL_GT3(devid))
+
+#define IS_GEN9(devid) (IS_SKYLAKE(devid) || IS_BROXTON(devid) || IS_KABYLAKE(devid) || IS_GEMINILAKE(devid) || IS_COFFEELAKE(devid))
#define MAX_OCLVERSION(devid) (IS_GEN9(devid) ? 200 : 120)
diff --git a/src/cl_device_id.c b/src/cl_device_id.c
index a4134404..c20f9eda 100644
--- a/src/cl_device_id.c
+++ b/src/cl_device_id.c
@@ -274,6 +274,36 @@ static struct _cl_device_id intel_glk12eu_device = {
#include "cl_gen9_device.h"
};
+static struct _cl_device_id intel_cfl_gt1_device = {
+ .max_compute_unit = 12,
+ .max_thread_per_unit = 7,
+ .sub_slice_count = 2,
+ .max_work_item_sizes = {512, 512, 512},
+ .max_work_group_size = 256,
+ .max_clock_frequency = 1000,
+#include "cl_gen9_device.h"
+};
+
+static struct _cl_device_id intel_cfl_gt2_device = {
+ .max_compute_unit = 24,
+ .max_thread_per_unit = 7,
+ .sub_slice_count = 3,
+ .max_work_item_sizes = {512, 512, 512},
+ .max_work_group_size = 256,
+ .max_clock_frequency = 1000,
+#include "cl_gen9_device.h"
+};
+
+static struct _cl_device_id intel_cfl_gt3_device = {
+ .max_compute_unit = 48,
+ .max_thread_per_unit = 7,
+ .sub_slice_count = 6,
+ .max_work_item_sizes = {512, 512, 512},
+ .max_work_group_size = 256,
+ .max_clock_frequency = 1000,
+#include "cl_gen9_device.h"
+};
+
LOCAL cl_device_id
cl_get_gt_device(cl_device_type device_type)
{
@@ -785,6 +815,64 @@ glk12eu_break:
cl_intel_platform_enable_extension(ret, cl_khr_fp16_ext_id);
break;
+ case PCI_CHIP_COFFEELAKE_S_GT1_1:
+ case PCI_CHIP_COFFEELAKE_S_GT1_2:
+ case PCI_CHIP_COFFEELAKE_S_GT1_3:
+ DECL_INFO_STRING(cfl_gt1_break, intel_cfl_gt1_device, name, "Intel(R) UHD Graphics Coffee Lake Desktop GT1");
+ case PCI_CHIP_COFFEELAKE_U_GT1_1:
+ case PCI_CHIP_COFFEELAKE_U_GT1_2:
+ DECL_INFO_STRING(cfl_gt1_break, intel_cfl_gt1_device, name, "Intel(R) UHD Graphics Coffee Lake Mobile GT1");
+cfl_gt1_break:
+ intel_cfl_gt1_device.device_id = device_id;
+ intel_cfl_gt1_device.platform = cl_get_platform_default();
+ ret = &intel_cfl_gt1_device;
+ cl_intel_platform_get_default_extension(ret);
+#ifdef ENABLE_FP64
+ cl_intel_platform_enable_extension(ret, cl_khr_fp64_ext_id);
+#endif
+ cl_intel_platform_enable_extension(ret, cl_khr_fp16_ext_id);
+ break;
+
+ case PCI_CHIP_COFFEELAKE_S_GT2_1:
+ case PCI_CHIP_COFFEELAKE_S_GT2_2:
+ case PCI_CHIP_COFFEELAKE_S_GT2_3:
+ case PCI_CHIP_COFFEELAKE_S_GT2_4:
+ DECL_INFO_STRING(cfl_gt2_break, intel_cfl_gt2_device, name, "Intel(R) UHD Graphics Coffee Lake Desktop GT2");
+ case PCI_CHIP_COFFEELAKE_H_GT2_1:
+ case PCI_CHIP_COFFEELAKE_H_GT2_2:
+ DECL_INFO_STRING(cfl_gt2_break, intel_cfl_gt2_device, name, "Intel(R) UHD Graphics Coffee Lake Halo GT2");
+ case PCI_CHIP_COFFEELAKE_U_GT2_1:
+ case PCI_CHIP_COFFEELAKE_U_GT2_2:
+ case PCI_CHIP_COFFEELAKE_U_GT2_3:
+ DECL_INFO_STRING(cfl_gt2_break, intel_cfl_gt2_device, name, "Intel(R) UHD Graphics Coffee Lake Mobile GT2");
+cfl_gt2_break:
+ intel_cfl_gt2_device.device_id = device_id;
+ intel_cfl_gt2_device.platform = cl_get_platform_default();
+ ret = &intel_cfl_gt2_device;
+ cl_intel_platform_get_default_extension(ret);
+#ifdef ENABLE_FP64
+ cl_intel_platform_enable_extension(ret, cl_khr_fp64_ext_id);
+#endif
+ cl_intel_platform_enable_extension(ret, cl_khr_fp16_ext_id);
+ break;
+
+ case PCI_CHIP_COFFEELAKE_U_GT3_1:
+ case PCI_CHIP_COFFEELAKE_U_GT3_2:
+ case PCI_CHIP_COFFEELAKE_U_GT3_3:
+ case PCI_CHIP_COFFEELAKE_U_GT3_4:
+ case PCI_CHIP_COFFEELAKE_U_GT3_5:
+ DECL_INFO_STRING(cfl_gt3_break, intel_cfl_gt3_device, name, "Intel(R) UHD Graphics Coffee Lake Mobile GT3");
+cfl_gt3_break:
+ intel_cfl_gt3_device.device_id = device_id;
+ intel_cfl_gt3_device.platform = cl_get_platform_default();
+ ret = &intel_cfl_gt3_device;
+ cl_intel_platform_get_default_extension(ret);
+#ifdef ENABLE_FP64
+ cl_intel_platform_enable_extension(ret, cl_khr_fp64_ext_id);
+#endif
+ cl_intel_platform_enable_extension(ret, cl_khr_fp16_ext_id);
+ break;
+
case PCI_CHIP_SANDYBRIDGE_BRIDGE:
case PCI_CHIP_SANDYBRIDGE_GT1:
case PCI_CHIP_SANDYBRIDGE_GT2:
@@ -992,7 +1080,10 @@ LOCAL cl_bool is_gen_device(cl_device_id device) {
device == &intel_kbl_gt3_device ||
device == &intel_kbl_gt4_device ||
device == &intel_glk18eu_device ||
- device == &intel_glk12eu_device;
+ device == &intel_glk12eu_device ||
+ device == &intel_cfl_gt1_device ||
+ device == &intel_cfl_gt2_device ||
+ device == &intel_cfl_gt3_device;
}
LOCAL cl_int
@@ -1420,7 +1511,9 @@ cl_device_get_version(cl_device_id device, cl_int *ver)
|| device == &intel_bxt18eu_device || device == &intel_bxt12eu_device || device == &intel_kbl_gt1_device
|| device == &intel_kbl_gt2_device || device == &intel_kbl_gt3_device
|| device == &intel_kbl_gt4_device || device == &intel_kbl_gt15_device
- || device == &intel_glk18eu_device || device == &intel_glk12eu_device) {
+ || device == &intel_glk18eu_device || device == &intel_glk12eu_device
+ || device == &intel_cfl_gt1_device || device == &intel_cfl_gt1_device
+ || device == &intel_cfl_gt3_device) {
*ver = 9;
} else
return CL_INVALID_VALUE;