radeonsi: add support for Vega12

Reviewed-by: Alex Deucher <[email protected]>

8 files changed, 41 insertions, 8 deletions

diff --git a/include/pci_ids/radeonsi_pci_ids.h b/include/pci_ids/radeonsi_pci_ids.h
index 6a3594eabc9..62b130307a3 100644
--- a/include/pci_ids/radeonsi_pci_ids.h
+++ b/include/pci_ids/radeonsi_pci_ids.h
@@ -226,4 +226,10 @@ CHIPSET(0x6868, VEGA10)
 CHIPSET(0x687F, VEGA10)
 CHIPSET(0x686C, VEGA10)
 
+CHIPSET(0x69A0, VEGA12)
+CHIPSET(0x69A1, VEGA12)
+CHIPSET(0x69A2, VEGA12)
+CHIPSET(0x69A3, VEGA12)
+CHIPSET(0x69AF, VEGA12)
+
 CHIPSET(0x15DD, RAVEN)
diff --git a/src/amd/common/ac_llvm_util.c b/src/amd/common/ac_llvm_util.c
index bb9e873af81..f3db1c5a4a4 100644
--- a/src/amd/common/ac_llvm_util.c
+++ b/src/amd/common/ac_llvm_util.c
@@ -114,6 +114,7 @@ const char *ac_get_llvm_processor_name(enum radeon_family family)
 	case CHIP_POLARIS12:
 		return "polaris11";
 	case CHIP_VEGA10:
+	case CHIP_VEGA12:
 	case CHIP_RAVEN:
 		return "gfx900";
 	default:
diff --git a/src/amd/common/ac_surface.c b/src/amd/common/ac_surface.c
index 603b7058bdc..12dfc0cb1f2 100644
--- a/src/amd/common/ac_surface.c
+++ b/src/amd/common/ac_surface.c
@@ -135,6 +135,10 @@ static void addrlib_family_rev_id(enum radeon_family family,
 		*addrlib_family = FAMILY_AI;
 		*addrlib_revid = get_first(AMDGPU_VEGA10_RANGE);
 		break;
+	case CHIP_VEGA12:
+		*addrlib_family = FAMILY_AI;
+		*addrlib_revid = get_first(AMDGPU_VEGA12_RANGE);
+		break;
 	case CHIP_RAVEN:
 		*addrlib_family = FAMILY_RV;
 		*addrlib_revid = get_first(AMDGPU_RAVEN_RANGE);
@@ -905,8 +909,8 @@ static int gfx9_compute_miptree(ADDR_HANDLE addrlib,
 		hin.size = sizeof(ADDR2_COMPUTE_HTILE_INFO_INPUT);
 		hout.size = sizeof(ADDR2_COMPUTE_HTILE_INFO_OUTPUT);
 
-		hin.hTileFlags.pipeAligned = 1;
-		hin.hTileFlags.rbAligned = 1;
+		hin.hTileFlags.pipeAligned = !in->flags.metaPipeUnaligned;
+		hin.hTileFlags.rbAligned = !in->flags.metaRbUnaligned;
 		hin.depthFlags = in->flags;
 		hin.swizzleMode = in->swizzleMode;
 		hin.unalignedWidth = in->width;
@@ -967,8 +971,8 @@ static int gfx9_compute_miptree(ADDR_HANDLE addrlib,
 			dout.size = sizeof(ADDR2_COMPUTE_DCCINFO_OUTPUT);
 			dout.pMipInfo = meta_mip_info;
 
-			din.dccKeyFlags.pipeAligned = 1;
-			din.dccKeyFlags.rbAligned = 1;
+			din.dccKeyFlags.pipeAligned = !in->flags.metaPipeUnaligned;
+			din.dccKeyFlags.rbAligned = !in->flags.metaRbUnaligned;
 			din.colorFlags = in->flags;
 			din.resourceType = in->resourceType;
 			din.swizzleMode = in->swizzleMode;
@@ -1088,8 +1092,14 @@ static int gfx9_compute_miptree(ADDR_HANDLE addrlib,
 			cin.size = sizeof(ADDR2_COMPUTE_CMASK_INFO_INPUT);
 			cout.size = sizeof(ADDR2_COMPUTE_CMASK_INFO_OUTPUT);
 
-			cin.cMaskFlags.pipeAligned = 1;
-			cin.cMaskFlags.rbAligned = 1;
+			if (in->numSamples) {
+				/* FMASK is always aligned. */
+				cin.cMaskFlags.pipeAligned = 1;
+				cin.cMaskFlags.rbAligned = 1;
+			} else {
+				cin.cMaskFlags.pipeAligned = !in->flags.metaPipeUnaligned;
+				cin.cMaskFlags.rbAligned = !in->flags.metaRbUnaligned;
+			}
 			cin.colorFlags = in->flags;
 			cin.resourceType = in->resourceType;
 			cin.unalignedWidth = in->width;
@@ -1116,6 +1126,7 @@ static int gfx9_compute_miptree(ADDR_HANDLE addrlib,
 }
 
 static int gfx9_compute_surface(ADDR_HANDLE addrlib,
+				const struct radeon_info *info,
 				const struct ac_surf_config *config,
 				enum radeon_surf_mode mode,
 				struct radeon_surf *surf)
@@ -1196,6 +1207,10 @@ static int gfx9_compute_surface(ADDR_HANDLE addrlib,
 	else
 		AddrSurfInfoIn.numSlices = config->info.array_size;
 
+	/* This is propagated to HTILE/DCC/CMASK. */
+	AddrSurfInfoIn.flags.metaPipeUnaligned = 0;
+	AddrSurfInfoIn.flags.metaRbUnaligned = 0;
+
 	switch (mode) {
 	case RADEON_SURF_MODE_LINEAR_ALIGNED:
 		assert(config->info.samples <= 1);
@@ -1321,6 +1336,10 @@ static int gfx9_compute_surface(ADDR_HANDLE addrlib,
 			assert(0);
 	}
 
+	/* Temporary workaround to prevent VM faults and hangs. */
+	if (info->family == CHIP_VEGA12)
+		surf->u.gfx9.fmask_size *= 8;
+
 	return 0;
 }
 
@@ -1336,7 +1355,7 @@ int ac_compute_surface(ADDR_HANDLE addrlib, const struct radeon_info *info,
 		return r;
 
 	if (info->chip_class >= GFX9)
-		return gfx9_compute_surface(addrlib, config, mode, surf);
+		return gfx9_compute_surface(addrlib, info, config, mode, surf);
 	else
 		return gfx6_compute_surface(addrlib, info, config, mode, surf);
 }
diff --git a/src/amd/common/amd_family.h b/src/amd/common/amd_family.h
index c62d0aa527a..285111f2a2a 100644
--- a/src/amd/common/amd_family.h
+++ b/src/amd/common/amd_family.h
@@ -93,6 +93,7 @@ enum radeon_family {
     CHIP_POLARIS11,
     CHIP_POLARIS12,
     CHIP_VEGA10,
+    CHIP_VEGA12,
     CHIP_RAVEN,
     CHIP_LAST,
 };
diff --git a/src/gallium/drivers/radeonsi/si_get.c b/src/gallium/drivers/radeonsi/si_get.c
index b4ca5bea943..fc2be33b3e4 100644
--- a/src/gallium/drivers/radeonsi/si_get.c
+++ b/src/gallium/drivers/radeonsi/si_get.c
@@ -75,6 +75,7 @@ const char *si_get_family_name(const struct si_screen *sscreen)
 	case CHIP_POLARIS12: return "AMD POLARIS12";
 	case CHIP_STONEY: return "AMD STONEY";
 	case CHIP_VEGA10: return "AMD VEGA10";
+	case CHIP_VEGA12: return "AMD VEGA12";
 	case CHIP_RAVEN: return "AMD RAVEN";
 	default: return "AMD unknown";
 	}
diff --git a/src/gallium/drivers/radeonsi/si_pipe.c b/src/gallium/drivers/radeonsi/si_pipe.c
index 3d787d58cd1..fa9ee43389a 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.c
+++ b/src/gallium/drivers/radeonsi/si_pipe.c
@@ -828,6 +828,7 @@ struct pipe_screen *radeonsi_screen_create(struct radeon_winsys *ws,
 		sscreen->dpbb_allowed = true;
 	} else {
 		/* Only enable primitive binning on Raven by default. */
+		/* TODO: Investigate if binning is profitable on Vega12. */
 		sscreen->dpbb_allowed = sscreen->info.family == CHIP_RAVEN &&
 					!(sscreen->debug_flags & DBG(NO_DPBB));
 	}
@@ -855,6 +856,7 @@ struct pipe_screen *radeonsi_screen_create(struct radeon_winsys *ws,
 		sscreen->rbplus_allowed =
 			!(sscreen->debug_flags & DBG(NO_RB_PLUS)) &&
 			(sscreen->info.family == CHIP_STONEY ||
+			 sscreen->info.family == CHIP_VEGA12 ||
 			 sscreen->info.family == CHIP_RAVEN);
 	}
 
diff --git a/src/gallium/drivers/radeonsi/si_state.c b/src/gallium/drivers/radeonsi/si_state.c
index 1bfb3c34aa7..b4165a4669b 100644
--- a/src/gallium/drivers/radeonsi/si_state.c
+++ b/src/gallium/drivers/radeonsi/si_state.c
@@ -1675,7 +1675,8 @@ static uint32_t si_translate_texformat(struct pipe_screen *screen,
 
 	if (desc->layout == UTIL_FORMAT_LAYOUT_ETC &&
 	    (sscreen->info.family == CHIP_STONEY ||
-	     sscreen->info.chip_class >= GFX9)) {
+	     sscreen->info.family == CHIP_VEGA10 ||
+	     sscreen->info.family == CHIP_RAVEN)) {
 		switch (format) {
 		case PIPE_FORMAT_ETC1_RGB8:
 		case PIPE_FORMAT_ETC2_RGB8:
@@ -5045,6 +5046,7 @@ static void si_init_config(struct si_context *sctx)
 
 		switch (sctx->b.family) {
 		case CHIP_VEGA10:
+		case CHIP_VEGA12:
 			pc_lines = 4096;
 			break;
 		case CHIP_RAVEN:
diff --git a/src/gallium/drivers/radeonsi/si_state_binning.c b/src/gallium/drivers/radeonsi/si_state_binning.c
index 686701d718f..0f50ea755cb 100644
--- a/src/gallium/drivers/radeonsi/si_state_binning.c
+++ b/src/gallium/drivers/radeonsi/si_state_binning.c
@@ -414,6 +414,7 @@ void si_emit_dpbb_state(struct si_context *sctx, struct r600_atom *state)
 
 	switch (sctx->b.family) {
 	case CHIP_VEGA10:
+	case CHIP_VEGA12:
 	case CHIP_RAVEN:
 		/* Tuned for Raven. Vega might need different values. */
 		context_states_per_bin = 5;