ac: add ac_build_bit_count() helper

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

diff --git a/src/amd/common/ac_llvm_build.c b/src/amd/common/ac_llvm_build.c
index 1f5112e9929491086c8c1fd2774a3aba907823d9..4fbe0ddb9c5ebe39778083574603ef3e915f7d13 100644 (file)
--- a/src/amd/common/ac_llvm_build.c
+++ b/src/amd/common/ac_llvm_build.c
@@ -2107,6 +2107,34 @@ LLVMValueRef ac_build_fsign(struct ac_llvm_context *ctx, LLVMValueRef src0,
        return val;
 }
 
+LLVMValueRef ac_build_bit_count(struct ac_llvm_context *ctx, LLVMValueRef src0)
+{
+       LLVMValueRef result;
+       unsigned bitsize;
+
+       bitsize = ac_get_elem_bits(ctx, LLVMTypeOf(src0));
+
+       switch (bitsize) {
+       case 64:
+               result = ac_build_intrinsic(ctx, "llvm.ctpop.i64", ctx->i64,
+                                           (LLVMValueRef []) { src0 }, 1,
+                                           AC_FUNC_ATTR_READNONE);
+
+               result = LLVMBuildTrunc(ctx->builder, result, ctx->i32, "");
+               break;
+       case 32:
+               result = ac_build_intrinsic(ctx, "llvm.ctpop.i32", ctx->i32,
+                                           (LLVMValueRef []) { src0 }, 1,
+                                           AC_FUNC_ATTR_READNONE);
+               break;
+       default:
+               unreachable(!"invalid bitsize");
+               break;
+       }
+
+       return result;
+}
+
 #define AC_EXP_TARGET          0
 #define AC_EXP_ENABLED_CHANNELS 1
 #define AC_EXP_OUT0            2

diff --git a/src/amd/common/ac_llvm_build.h b/src/amd/common/ac_llvm_build.h
index 0d261bae0977edf1166275faad6917e4bb09fe27..98a61a2405426d0e5b86b8dbe3b52d0f1118f009 100644 (file)
--- a/src/amd/common/ac_llvm_build.h
+++ b/src/amd/common/ac_llvm_build.h
@@ -424,6 +424,8 @@ LLVMValueRef ac_build_isign(struct ac_llvm_context *ctx, LLVMValueRef src0,
 LLVMValueRef ac_build_fsign(struct ac_llvm_context *ctx, LLVMValueRef src0,
                            unsigned bitsize);
 
+LLVMValueRef ac_build_bit_count(struct ac_llvm_context *ctx, LLVMValueRef src0);
+
 void ac_optimize_vs_outputs(struct ac_llvm_context *ac,
                            LLVMValueRef main_fn,
                            uint8_t *vs_output_param_offset,

diff --git a/src/amd/common/ac_nir_to_llvm.c b/src/amd/common/ac_nir_to_llvm.c
index a3f26ab46c98940ca5c8c86bf30ee03b595669f4..d79ff4399b1fe42f9071393b331f479e9aa28b45 100644 (file)
--- a/src/amd/common/ac_nir_to_llvm.c
+++ b/src/amd/common/ac_nir_to_llvm.c
@@ -839,12 +839,7 @@ static void visit_alu(struct ac_nir_context *ctx, const nir_alu_instr *instr)
                result = ac_build_intrinsic(&ctx->ac, "llvm.bitreverse.i32", ctx->ac.i32, src, 1, AC_FUNC_ATTR_READNONE);
                break;
        case nir_op_bit_count:
-               if (ac_get_elem_bits(&ctx->ac, LLVMTypeOf(src[0])) == 32)
-                       result = ac_build_intrinsic(&ctx->ac, "llvm.ctpop.i32", ctx->ac.i32, src, 1, AC_FUNC_ATTR_READNONE);
-               else {
-                       result = ac_build_intrinsic(&ctx->ac, "llvm.ctpop.i64", ctx->ac.i64, src, 1, AC_FUNC_ATTR_READNONE);
-                       result = LLVMBuildTrunc(ctx->ac.builder, result, ctx->ac.i32, "");
-               }
+               result = ac_build_bit_count(&ctx->ac, src[0]);
                break;
        case nir_op_vec2:
        case nir_op_vec3: