radv/llvm: fix subgroup shuffle for chips without bpermute

bpermute only exists on GFX8+ and only with Wave32 on GFX10. Instead
we have to use readlane with a waterfall loop to defeat the LLVM
backend.

This fixes DOOM Eternal which requires subgroup shuffle.

Cc: <[email protected]>
Signed-off-by: Samuel Pitoiset <[email protected]>
Reviewed-by: Bas Nieuwenhuizen <[email protected]>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4284>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4284>

2 files changed, 30 insertions, 5 deletions

diff --git a/src/amd/llvm/ac_nir_to_llvm.c b/src/amd/llvm/ac_nir_to_llvm.c
index 49627990163..871c6abc17f 100644
--- a/src/amd/llvm/ac_nir_to_llvm.c
+++ b/src/amd/llvm/ac_nir_to_llvm.c
@@ -3950,8 +3950,33 @@ static void visit_intrinsic(struct ac_nir_context *ctx,
 		break;
 	}
 	case nir_intrinsic_shuffle:
-		result = ac_build_shuffle(&ctx->ac, get_src(ctx, instr->src[0]),
-				get_src(ctx, instr->src[1]));
+		if (ctx->ac.chip_class == GFX8 ||
+		    ctx->ac.chip_class == GFX9 ||
+		    (ctx->ac.chip_class == GFX10 && ctx->ac.wave_size == 32)) {
+			result = ac_build_shuffle(&ctx->ac, get_src(ctx, instr->src[0]),
+						  get_src(ctx, instr->src[1]));
+		} else {
+			LLVMValueRef src = get_src(ctx, instr->src[0]);
+			LLVMValueRef index = get_src(ctx, instr->src[1]);
+			LLVMTypeRef type = LLVMTypeOf(src);
+	                struct waterfall_context wctx;
+	                LLVMValueRef index_val;
+
+	                index_val = enter_waterfall(ctx, &wctx, index, true);
+
+			src = LLVMBuildZExt(ctx->ac.builder, src,
+					    ctx->ac.i32, "");
+
+			result = ac_build_intrinsic(&ctx->ac, "llvm.amdgcn.readlane",
+						    ctx->ac.i32,
+						    (LLVMValueRef []) { src, index_val }, 2,
+						    AC_FUNC_ATTR_READNONE |
+						    AC_FUNC_ATTR_CONVERGENT);
+
+			result = LLVMBuildTrunc(ctx->ac.builder, result, type, "");
+
+		        result = exit_waterfall(ctx, &wctx, result);
+		}
 		break;
 	case nir_intrinsic_reduce:
 		result = ac_build_reduce(&ctx->ac,
diff --git a/src/amd/vulkan/radv_device.c b/src/amd/vulkan/radv_device.c
index 2f44b279ac1..1ecac0c1b55 100644
--- a/src/amd/vulkan/radv_device.c
+++ b/src/amd/vulkan/radv_device.c
@@ -1481,9 +1481,9 @@ radv_get_physical_device_properties_1_1(struct radv_physical_device *pdevice,
 					 VK_SUBGROUP_FEATURE_CLUSTERED_BIT |
 					 VK_SUBGROUP_FEATURE_QUAD_BIT;
 
-	if (pdevice->rad_info.chip_class == GFX8 ||
-	    pdevice->rad_info.chip_class == GFX9 ||
-	    (pdevice->rad_info.chip_class == GFX10 && pdevice->use_aco)) {
+	if (((pdevice->rad_info.chip_class == GFX6 ||
+	      pdevice->rad_info.chip_class == GFX7) && !pdevice->use_aco) ||
+	    pdevice->rad_info.chip_class >= GFX8) {
 		p->subgroupSupportedOperations |= VK_SUBGROUP_FEATURE_SHUFFLE_BIT |
 						  VK_SUBGROUP_FEATURE_SHUFFLE_RELATIVE_BIT;
 	}