radeon/llvm: add support for texture offsets, fix TEX_LD

Signed-off-by: Vadim Girlin <vadimgirlin@gmail.com>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>

src/gallium/drivers/radeon/AMDGPUIntrinsics.td

@@ -42,7 +42,7 @@ let TargetPrefix = "AMDGPU", isTarget = 1 in {
   def int_AMDGPU_mullit : Intrinsic<[llvm_v4f32_ty], [llvm_float_ty, llvm_float_ty, llvm_float_ty], [IntrNoMem]>;
   def int_AMDGPU_tex : Intrinsic<[llvm_v4f32_ty], [llvm_v4f32_ty, llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
   def int_AMDGPU_txb : Intrinsic<[llvm_v4f32_ty], [llvm_v4f32_ty, llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
-  def int_AMDGPU_txf : Intrinsic<[llvm_v4f32_ty], [llvm_v4f32_ty, llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
+  def int_AMDGPU_txf : Intrinsic<[llvm_v4f32_ty], [llvm_v4f32_ty, llvm_i32_ty, llvm_i32_ty, llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
   def int_AMDGPU_txq : Intrinsic<[llvm_v4f32_ty], [llvm_v4f32_ty, llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
   def int_AMDGPU_txd : Intrinsic<[llvm_v4f32_ty], [llvm_v4f32_ty, llvm_v4f32_ty, llvm_v4f32_ty, llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
   def int_AMDGPU_txl : Intrinsic<[llvm_v4f32_ty], [llvm_v4f32_ty, llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;

src/gallium/drivers/radeon/R600CodeEmitter.cpp

@@ -456,9 +456,11 @@ void R600CodeEmitter::emitALU(MachineInstr &MI, unsigned numSrc)
 void R600CodeEmitter::emitTexInstr(MachineInstr &MI)
 {
 
-  int64_t sampler = MI.getOperand(2).getImm();
-  int64_t textureType = MI.getOperand(3).getImm();
   unsigned opcode = MI.getOpcode();
+  bool hasOffsets = (opcode == AMDIL::TEX_LD);
+  unsigned op_offset = hasOffsets ? 3 : 0;
+  int64_t sampler = MI.getOperand(op_offset+2).getImm();
+  int64_t textureType = MI.getOperand(op_offset+3).getImm();
   unsigned srcSelect[4] = {0, 1, 2, 3};
 
   // Emit instruction type
@@ -518,10 +520,11 @@ void R600CodeEmitter::emitTexInstr(MachineInstr &MI)
   }
 
   // XXX: Emit offsets
-  emitByte(0); // X
-  emitByte(0); // Y
-  emitByte(0); // Z
-  // There is no OFFSET_W
+  if (hasOffsets)
+	  for (unsigned i = 2; i < 5; i++)
+		  emitByte(MI.getOperand(i).getImm()<<1);
+  else
+	  emitNullBytes(3);
 
   // Emit sampler id
   emitByte(sampler);

src/gallium/drivers/radeon/R600Instructions.td

@@ -430,8 +430,11 @@ def CNDE_INT : R600_3OP <
 
 def TEX_LD : R600_TEX <
   0x03, "TEX_LD",
-  [(set R600_Reg128:$dst, (int_AMDGPU_txf R600_Reg128:$src0, imm:$src1, imm:$src2))]
->;
+  [(set R600_Reg128:$dst, (int_AMDGPU_txf R600_Reg128:$src0, imm:$src1, imm:$src2, imm:$src3, imm:$src4, imm:$src5))]
+> {
+let AsmString = "TEX_LD $dst, $src0, $src1, $src2, $src3, $src4, $src5";
+let InOperandList = (ins R600_Reg128:$src0, i32imm:$src1, i32imm:$src2, i32imm:$src3, i32imm:$src4, i32imm:$src5);
+}
 
 def TEX_GET_TEXTURE_RESINFO : R600_TEX <
   0x04, "TEX_GET_TEXTURE_RESINFO",

src/gallium/drivers/radeon/radeon_setup_tgsi_llvm.c

@@ -581,6 +581,7 @@ static void txd_fetch_args(
 	emit_data->dst_type = LLVMVectorType(bld_base->base.elem_type, 4);
 }
 
+
 static void txp_fetch_args(
 	struct lp_build_tgsi_context * bld_base,
 	struct lp_build_emit_data * emit_data)
@@ -643,6 +644,40 @@ static void tex_fetch_args(
 	}
 }
 
+static void txf_fetch_args(
+	struct lp_build_tgsi_context * bld_base,
+	struct lp_build_emit_data * emit_data)
+{
+	const struct tgsi_full_instruction * inst = emit_data->inst;
+	struct lp_build_tgsi_soa_context *bld = lp_soa_context(bld_base);
+	const struct tgsi_texture_offset * off = inst->TexOffsets;
+	LLVMTypeRef offset_type = bld_base->int_bld.elem_type;
+
+	/* fetch tex coords */
+	tex_fetch_args(bld_base, emit_data);
+
+	/* fetch tex offsets */
+	if (inst->Texture.NumOffsets) {
+		assert(inst->Texture.NumOffsets == 1);
+
+		emit_data->args[1] = LLVMConstBitCast(
+			bld->immediates[off->Index][off->SwizzleX],
+			offset_type);
+		emit_data->args[2] = LLVMConstBitCast(
+			bld->immediates[off->Index][off->SwizzleY],
+			offset_type);
+		emit_data->args[3] = LLVMConstBitCast(
+			bld->immediates[off->Index][off->SwizzleZ],
+			offset_type);
+	} else {
+		emit_data->args[1] = bld_base->int_bld.zero;
+		emit_data->args[2] = bld_base->int_bld.zero;
+		emit_data->args[3] = bld_base->int_bld.zero;
+	}
+
+	emit_data->arg_count = 4;
+}
+
 static void emit_icmp(
 		const struct lp_build_tgsi_action * action,
 		struct lp_build_tgsi_context * bld_base,
@@ -1029,7 +1064,7 @@ void radeon_llvm_context_init(struct radeon_llvm_context * ctx)
 	bld_base->op_actions[TGSI_OPCODE_UMIN].intr_name = "llvm.AMDGPU.umin";
 	bld_base->op_actions[TGSI_OPCODE_UMAX].emit = build_tgsi_intrinsic_nomem;
 	bld_base->op_actions[TGSI_OPCODE_UMAX].intr_name = "llvm.AMDGPU.umax";
-	bld_base->op_actions[TGSI_OPCODE_TXF].fetch_args = tex_fetch_args;
+	bld_base->op_actions[TGSI_OPCODE_TXF].fetch_args = txf_fetch_args;
 	bld_base->op_actions[TGSI_OPCODE_TXF].intr_name = "llvm.AMDGPU.txf";
 	bld_base->op_actions[TGSI_OPCODE_TXQ].fetch_args = tex_fetch_args;
 	bld_base->op_actions[TGSI_OPCODE_TXQ].intr_name = "llvm.AMDGPU.txq";