ac: add 8-bit and 16-bit supports to ac_build_permlane16()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

diff --git a/src/amd/llvm/ac_llvm_build.c b/src/amd/llvm/ac_llvm_build.c
index 54513d79922baf9cb55cb18aade6b235a16dfb32..cf6eda30e2cc61187f4d0d0f849d77c5505428a5 100644 (file)
--- a/src/amd/llvm/ac_llvm_build.c
+++ b/src/amd/llvm/ac_llvm_build.c
@@ -3768,6 +3768,11 @@ static LLVMValueRef
 _ac_build_permlane16(struct ac_llvm_context *ctx, LLVMValueRef src, uint64_t sel,
                     bool exchange_rows, bool bound_ctrl)
 {
+       LLVMTypeRef type = LLVMTypeOf(src);
+       LLVMValueRef result;
+
+       src = LLVMBuildZExt(ctx->builder, src, ctx->i32, "");
+
        LLVMValueRef args[6] = {
                src,
                src,
@@ -3776,10 +3781,13 @@ _ac_build_permlane16(struct ac_llvm_context *ctx, LLVMValueRef src, uint64_t sel
                ctx->i1true, /* fi */
                bound_ctrl ? ctx->i1true : ctx->i1false,
        };
-       return ac_build_intrinsic(ctx, exchange_rows ? "llvm.amdgcn.permlanex16"
-                                                    : "llvm.amdgcn.permlane16",
-                                 ctx->i32, args, 6,
-                                 AC_FUNC_ATTR_READNONE | AC_FUNC_ATTR_CONVERGENT);
+
+       result = ac_build_intrinsic(ctx, exchange_rows ? "llvm.amdgcn.permlanex16"
+                                                      : "llvm.amdgcn.permlane16",
+                                   ctx->i32, args, 6,
+                                   AC_FUNC_ATTR_READNONE | AC_FUNC_ATTR_CONVERGENT);
+
+       return LLVMBuildTrunc(ctx->builder, result, type, "");
 }
 
 static LLVMValueRef
@@ -3790,10 +3798,7 @@ ac_build_permlane16(struct ac_llvm_context *ctx, LLVMValueRef src, uint64_t sel,
        src = ac_to_integer(ctx, src);
        unsigned bits = LLVMGetIntTypeWidth(LLVMTypeOf(src));
        LLVMValueRef ret;
-       if (bits == 32) {
-               ret = _ac_build_permlane16(ctx, src, sel, exchange_rows,
-                                          bound_ctrl);
-       } else {
+       if (bits > 32) {
                assert(bits % 32 == 0);
                LLVMTypeRef vec_type = LLVMVectorType(ctx->i32, bits / 32);
                LLVMValueRef src_vector =
@@ -3812,6 +3817,9 @@ ac_build_permlane16(struct ac_llvm_context *ctx, LLVMValueRef src, uint64_t sel,
                                                     LLVMConstInt(ctx->i32, i,
                                                                  0), "");
                }
+       } else {
+               ret = _ac_build_permlane16(ctx, src, sel, exchange_rows,
+                                          bound_ctrl);
        }
        return LLVMBuildBitCast(ctx->builder, ret, src_type, "");
 }