ac/llvm: add better code for fsign

[mesa.git] / src / amd / llvm / ac_llvm_build.h

diff --git a/src/amd/llvm/ac_llvm_build.h b/src/amd/llvm/ac_llvm_build.h
index 772054efecde2d8dd02eed161526c1e6884bcd64..6491843cd78241cbe1fffaa7e0dabb0fb0c3d0b9 100644 (file)
--- a/src/amd/llvm/ac_llvm_build.h
+++ b/src/amd/llvm/ac_llvm_build.h
@@ -79,6 +79,9 @@ struct ac_llvm_context {
        LLVMTypeRef f32;
        LLVMTypeRef f64;
        LLVMTypeRef v2i16;
+       LLVMTypeRef v4i16;
+       LLVMTypeRef v2f16;
+       LLVMTypeRef v4f16;
        LLVMTypeRef v2i32;
        LLVMTypeRef v3i32;
        LLVMTypeRef v4i32;
@@ -108,6 +111,12 @@ struct ac_llvm_context {
        LLVMValueRef i1true;
        LLVMValueRef i1false;
 
+       /* Temporary helper to implement demote_to_helper:
+         * True = live lanes
+         * False = demoted lanes
+         */
+       LLVMValueRef postponed_kill;
+
        /* Since ac_nir_translate makes a local copy of ac_llvm_context, there
         * are two ac_llvm_contexts. Declare a pointer here, so that the control
         * flow stack is shared by both ac_llvm_contexts.
@@ -117,8 +126,6 @@ struct ac_llvm_context {
        unsigned range_md_kind;
        unsigned invariant_load_md_kind;
        unsigned uniform_md_kind;
-       unsigned fpmath_md_kind;
-       LLVMValueRef fpmath_md_2p5_ulp;
        LLVMValueRef empty_md;
 
        enum chip_class chip_class;
@@ -177,7 +184,8 @@ void ac_build_s_barrier(struct ac_llvm_context *ctx);
 void ac_build_optimization_barrier(struct ac_llvm_context *ctx,
                                   LLVMValueRef *pvgpr);
 
-LLVMValueRef ac_build_shader_clock(struct ac_llvm_context *ctx);
+LLVMValueRef ac_build_shader_clock(struct ac_llvm_context *ctx,
+                                  nir_scope scope);
 
 LLVMValueRef ac_build_ballot(struct ac_llvm_context *ctx, LLVMValueRef value);
 LLVMValueRef ac_get_i1_sgpr_mask(struct ac_llvm_context *ctx,
@@ -310,7 +318,6 @@ ac_build_buffer_store_format(struct ac_llvm_context *ctx,
                             LLVMValueRef data,
                             LLVMValueRef vindex,
                             LLVMValueRef voffset,
-                            unsigned num_channels,
                             unsigned cache_policy);
 
 LLVMValueRef
@@ -331,7 +338,8 @@ LLVMValueRef ac_build_buffer_load_format(struct ac_llvm_context *ctx,
                                         LLVMValueRef voffset,
                                         unsigned num_channels,
                                         unsigned cache_policy,
-                                        bool can_speculate);
+                                        bool can_speculate,
+                                        bool d16);
 
 LLVMValueRef
 ac_build_tbuffer_load_short(struct ac_llvm_context *ctx,
@@ -546,6 +554,7 @@ struct ac_image_args {
        unsigned cache_policy : 3;
        bool unorm : 1;
        bool level_zero : 1;
+       bool d16 : 1; /* data and return values are 16-bit, requires GFX8+ */
        unsigned attributes; /* additional call-site specific AC_FUNC_ATTRs */
 
        LLVMValueRef resource;
@@ -557,6 +566,7 @@ struct ac_image_args {
        LLVMValueRef derivs[6];
        LLVMValueRef coords[4];
        LLVMValueRef lod; // also used by ac_image_get_resinfo
+       LLVMValueRef min_lod;
 };
 
 LLVMValueRef ac_build_image_opcode(struct ac_llvm_context *ctx,
@@ -587,17 +597,8 @@ void ac_build_waitcnt(struct ac_llvm_context *ctx, unsigned wait_flags);
 
 LLVMValueRef ac_build_fract(struct ac_llvm_context *ctx, LLVMValueRef src0,
                           unsigned bitsize);
-
-LLVMValueRef ac_build_fmed3(struct ac_llvm_context *ctx, LLVMValueRef src0,
-                           LLVMValueRef src1, LLVMValueRef src2,
-                           unsigned bitsize);
-
-LLVMValueRef ac_build_isign(struct ac_llvm_context *ctx, LLVMValueRef src0,
-                           unsigned bitsize);
-
-LLVMValueRef ac_build_fsign(struct ac_llvm_context *ctx, LLVMValueRef src0,
-                           unsigned bitsize);
-
+LLVMValueRef ac_build_isign(struct ac_llvm_context *ctx, LLVMValueRef src0);
+LLVMValueRef ac_build_fsign(struct ac_llvm_context *ctx, LLVMValueRef src);
 LLVMValueRef ac_build_bit_count(struct ac_llvm_context *ctx, LLVMValueRef src0);
 
 LLVMValueRef ac_build_bitfield_reverse(struct ac_llvm_context *ctx,
@@ -607,6 +608,7 @@ void ac_optimize_vs_outputs(struct ac_llvm_context *ac,
                            LLVMValueRef main_fn,
                            uint8_t *vs_output_param_offset,
                            uint32_t num_outputs,
+                           uint32_t skip_output_mask,
                            uint8_t *num_param_exports);
 void ac_init_exec_full_mask(struct ac_llvm_context *ctx);
 
@@ -737,6 +739,9 @@ ac_build_ddxy_interp(struct ac_llvm_context *ctx, LLVMValueRef interp_ij);
 LLVMValueRef
 ac_build_load_helper_invocation(struct ac_llvm_context *ctx);
 
+LLVMValueRef
+ac_build_is_helper_invocation(struct ac_llvm_context *ctx);
+
 LLVMValueRef ac_build_call(struct ac_llvm_context *ctx, LLVMValueRef func,
                           LLVMValueRef *args, unsigned num_args);