[Mesa-dev] [PATCH 2/3] ac/llvm: consolidate find lsb function.

[Dave Airlie]

From: Dave Airlie <airlied at redhat.com>

This was the same between si and ac.

Signed-off-by: Dave Airlie <airlied at redhat.com>
---
 src/amd/common/ac_llvm_build.c                    | 31 +++++++++++++++++++++++
 src/amd/common/ac_llvm_build.h                    |  4 +++
 src/amd/common/ac_nir_to_llvm.c                   | 30 +---------------------
 src/gallium/drivers/radeonsi/si_shader_tgsi_alu.c | 27 ++------------------
 4 files changed, 38 insertions(+), 54 deletions(-)

diff --git a/src/amd/common/ac_llvm_build.c b/src/amd/common/ac_llvm_build.c
index 2b1f15b..242712e 100644
--- a/src/amd/common/ac_llvm_build.c
+++ b/src/amd/common/ac_llvm_build.c
@@ -1775,3 +1775,34 @@ ac_lds_store(struct ac_llvm_context *ctx,
 	ac_build_indexed_store(ctx, ctx->lds,
 			       dw_addr, value);
 }
+
+LLVMValueRef ac_find_lsb(struct ac_llvm_context *ctx,
+			 LLVMTypeRef dst_type,
+			 LLVMValueRef src0)
+{
+	LLVMValueRef params[2] = {
+		src0,
+
+		/* The value of 1 means that ffs(x=0) = undef, so LLVM won't
+		 * add special code to check for x=0. The reason is that
+		 * the LLVM behavior for x=0 is different from what we
+		 * need here. However, LLVM also assumes that ffs(x) is
+		 * in [0, 31], but GLSL expects that ffs(0) = -1, so
+		 * a conditional assignment to handle 0 is still required.
+		 *
+		 * The hardware already implements the correct behavior.
+		 */
+		LLVMConstInt(ctx->i1, 1, false),
+	};
+
+	LLVMValueRef lsb = ac_build_intrinsic(ctx, "llvm.cttz.i32", ctx->i32,
+					      params, 2,
+					      AC_FUNC_ATTR_READNONE);
+
+	/* TODO: We need an intrinsic to skip this conditional. */
+	/* Check for zero: */
+	return LLVMBuildSelect(ctx->builder, LLVMBuildICmp(ctx->builder,
+							   LLVMIntEQ, src0,
+							   ctx->i32_0, ""),
+			       LLVMConstInt(ctx->i32, -1, 0), lsb, "");
+}
diff --git a/src/amd/common/ac_llvm_build.h b/src/amd/common/ac_llvm_build.h
index b47d51a..f790619 100644
--- a/src/amd/common/ac_llvm_build.h
+++ b/src/amd/common/ac_llvm_build.h
@@ -297,6 +297,10 @@ LLVMValueRef ac_lds_load(struct ac_llvm_context *ctx,
 			 LLVMValueRef dw_addr);
 void ac_lds_store(struct ac_llvm_context *ctx,
 		  LLVMValueRef dw_addr, LLVMValueRef value);
+
+LLVMValueRef ac_find_lsb(struct ac_llvm_context *ctx,
+			 LLVMTypeRef dst_type,
+			 LLVMValueRef src0);
 #ifdef __cplusplus
 }
 #endif
diff --git a/src/amd/common/ac_nir_to_llvm.c b/src/amd/common/ac_nir_to_llvm.c
index f78f486..0167755 100644
--- a/src/amd/common/ac_nir_to_llvm.c
+++ b/src/amd/common/ac_nir_to_llvm.c
@@ -1227,34 +1227,6 @@ static LLVMValueRef emit_bcsel(struct ac_llvm_context *ctx,
 	return LLVMBuildSelect(ctx->builder, v, src1, src2, "");
 }
 
-static LLVMValueRef emit_find_lsb(struct ac_llvm_context *ctx,
-				  LLVMValueRef src0)
-{
-	LLVMValueRef params[2] = {
-		src0,
-
-		/* The value of 1 means that ffs(x=0) = undef, so LLVM won't
-		 * add special code to check for x=0. The reason is that
-		 * the LLVM behavior for x=0 is different from what we
-		 * need here.
-		 *
-		 * The hardware already implements the correct behavior.
-		 */
-		LLVMConstInt(ctx->i1, 1, false),
-	};
-
-	LLVMValueRef lsb = ac_build_intrinsic(ctx, "llvm.cttz.i32", ctx->i32,
-					      params, 2,
-					      AC_FUNC_ATTR_READNONE);
-
-	/* TODO: We need an intrinsic to skip this conditional. */
-	/* Check for zero: */
-	return LLVMBuildSelect(ctx->builder, LLVMBuildICmp(ctx->builder,
-							   LLVMIntEQ, src0,
-							   ctx->i32_0, ""),
-			       LLVMConstInt(ctx->i32, -1, 0), lsb, "");
-}
-
 static LLVMValueRef emit_ifind_msb(struct ac_llvm_context *ctx,
 				   LLVMValueRef src0)
 {
@@ -1895,7 +1867,7 @@ static void visit_alu(struct ac_nir_context *ctx, const nir_alu_instr *instr)
 		break;
 	case nir_op_find_lsb:
 		src[0] = ac_to_integer(&ctx->ac, src[0]);
-		result = emit_find_lsb(&ctx->ac, src[0]);
+		result = ac_find_lsb(&ctx->ac, ctx->ac.i32, src[0]);
 		break;
 	case nir_op_ufind_msb:
 		src[0] = ac_to_integer(&ctx->ac, src[0]);
diff --git a/src/gallium/drivers/radeonsi/si_shader_tgsi_alu.c b/src/gallium/drivers/radeonsi/si_shader_tgsi_alu.c
index 913b6c3..a27586b 100644
--- a/src/gallium/drivers/radeonsi/si_shader_tgsi_alu.c
+++ b/src/gallium/drivers/radeonsi/si_shader_tgsi_alu.c
@@ -535,31 +535,8 @@ static void emit_lsb(const struct lp_build_tgsi_action *action,
 		     struct lp_build_emit_data *emit_data)
 {
 	struct si_shader_context *ctx = si_shader_context(bld_base);
-	LLVMValueRef args[2] = {
-		emit_data->args[0],
-
-		/* The value of 1 means that ffs(x=0) = undef, so LLVM won't
-		 * add special code to check for x=0. The reason is that
-		 * the LLVM behavior for x=0 is different from what we
-		 * need here. However, LLVM also assumes that ffs(x) is
-		 * in [0, 31], but GLSL expects that ffs(0) = -1, so
-		 * a conditional assignment to handle 0 is still required.
-		 */
-		LLVMConstInt(ctx->i1, 1, 0)
-	};
-
-	LLVMValueRef lsb =
-		lp_build_intrinsic(ctx->ac.builder, "llvm.cttz.i32",
-				emit_data->dst_type, args, ARRAY_SIZE(args),
-				LP_FUNC_ATTR_READNONE);
-
-	/* TODO: We need an intrinsic to skip this conditional. */
-	/* Check for zero: */
-	emit_data->output[emit_data->chan] =
-		LLVMBuildSelect(ctx->ac.builder,
-				LLVMBuildICmp(ctx->ac.builder, LLVMIntEQ, args[0],
-					      ctx->i32_0, ""),
-				LLVMConstInt(ctx->i32, -1, 0), lsb, "");
+
+	emit_data->output[emit_data->chan] = ac_find_lsb(&ctx->ac, emit_data->dst_type, emit_data->args[0]);
 }
 
 /* Find the last bit set. */
-- 
2.9.5